Essays on Job Search, Unemployment, and Regulatory Compliance

By

G6kqe Bagbug

B.A., Istanbul University, 2005 M.A., Istanbul University, 2008 M.S., Massachusetts Institute of Technology, 2015

SUBMITTED TO THE SLOAN SCHOOL OF MANAGEMENT IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY IN MANAGEMENT

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

JUNE 2017

C2017 Massachusetts Institute of Technology. All rights reserved. I Signature redacted Signature of Author: Department of Management May 5, 2017 Signature redacted Certified by: Thomas A. Kochan George Maverick Bunker Professor of Management/ Thesis Co-Supe/visor Signature redacted Certified by: /&oberto M ern ez William F. Pounds Professor in ag ent Thesis C -Sup isor Signature redacted Accepted by: INSTITUTE MASSACHUSETTS Catherine Tucker Sloan Distinguished Professor of Management MASSACHUSETTS INSTITUTE OF TECHNOLOGY co Professor of Marketing LU Chair, MIT Sloan PhD Program JUN 262017 I 0 LIBRARIES 77 Massachusetts Avenue Cambridge, MA 02139 MfTLibraries http://Iibraries.mit.edu/ask

DISCLAIMER NOTICE

Due to the condition of the original material, there are unavoidable flaws in this reproduction. We have made every effort possible to provide you with the best copy available.

Thank you.

The images contained in this document are of the best quality available. Essays on Job Search, Unemployment, and Regulatory Compliance

by

Gik9e Bagbug Submitted to the Sloan School of Management on May 5, 2017 in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Management

Abstract

This dissertation is composed of four essays, each studying limits to the means that are famously known to be effective. In the first essay, I investigate the effect of using social ties on the quality of opportunities pursued in job search. Using fixed effects models, I show that when the job seeker uses contacts, she pursues jobs that pay less than the jobs she pursues through formal methods. In addition, the analysis show that networks limit the geographical and occupational diversity of jobs pursued. In the second essay, using a mixed methods approach I examine how the negative emotional toll of long-term unemployment (LTU) is shaped by the interaction of gender and marital status. The interviews reveal a pattern with more marital tensions that exacerbate the emotional toll of LTU reported by married men than married women. The analysis of survey data show that overall marriages are helpful to the wellbeing of both unemployed men and women. Yet, for married men but not for married women, the analysis revealed that the significance of the benefits derived from marriage disappears once household income is controlled. The third essay examines whether introducing choice into a mandatory training program makes any difference in Unemployment Insurance recipients' job search performance. Using a field experiment design, I compare job search outcomes of individuals who have unconstrained workshop choices to others who only have a constrained option. Analyses show that providing the option of selecting which workshop to attend does not improve outcomes for unemployed. In the fourth essay, using data from safety inspections in laboratories at a large university, I investigate whether conducting semi-annual inspections and reporting findings back to responsible constituencies decreases the number of violations. The analyses show that the Environmental, Health, and Safety Management System did not reduce the number of violations. Rather, the results show a widening gap between compliant and non-compliant scientists. Using both lab-level quantitative data and interviews with inspectors and EHS personnel, I delineate the factors that impede the success of the system.

Thesis Co-Supervisor: Thomas A. Kochan Title: George Maverick Bunker Professor of Management

Thesis Co-Supervisor: Roberto M. Fernandez Title: William F. Pounds Professor in Management

1 To Atif Balbug

2 Acknowledgements

Completing a doctoral program is not an easy journey. One needs a supportive committee more than anything. I was very lucky in this sense. I first want to thank Tom Kochan. He always encouraged me to see the broader picture and to think about the policy implications of my research. Beyond academic guidance, Tom provided me any kind of support a student needs. Words are not enough to express my gratitude to him. I thank Roberto Fernandez who helped me develop a taste for good research. His time and effort that he gave to me facilitated my transition from a consumer to a producer of knowledge. I also would like to thank Emilio Castilla who helped me improve my quantitative research skills. His attention to details has always been very helpful. I have had an angel on my shoulder: Susan Silbey. Susan not only guided my intellectual growth but also showed tremendous care to me and my family. Paul Osterman has always been very helpful with his sharp eyes for seeing puzzles and alternative explanations. To thank Ofer Sharone enough, I must write another acknowledgement only dedicated to him. Ofer, my role model, is an example of a scholar who simultaneously does good research and makes difference in people's lives. I have learned from him a lot and will continue learning. I am enormously grateful to the whole IWER community including my professors Bob McKersie, Erin Kelly, and Matt Amengual and my friends Alberto Fuentes, Alex Kowalski, Andrew Weaver, Aruna Ranganathan, Ben Rissing, Duanyi Yang, George Ward, Mahreen Khan, Maja Tampe, Ryan Hammond, and Will Kimball. Friends always remind the beautiful aspects of the world. I was very lucky to be admitted to the program in the same year with Hye Jin Rho. Without her, I would have felt very lonely. I owe so much to Christine Riordan's thoughtful friendship. My friends, Arvind Karunakaran, Burak Dura, Can Ulusel, Elliott Greenblatt, Hyejun Kim, Minjae Kim, Josh Krieger, Orhan Celiker, Ozge Karanfil, Paul Tillberg, Santiago Campero, Sema Ermez, Shirley Poyau, Stefan Beljean, made my life here more exciting. I also thank my friends at Turkish Student Association and Eastgate Family Housing. During my Ph.D., I was very lucky to get help from many nice people at MIT and in Cambridge. I owe enormous thanks to Dave Buck, Katy Bertman, Jacky McGoldrick, Ayn Cavicchi, Hillary Ross, Davin Schnappauf, Sarah Massey, and Sharon Cayley. Finally, I thank my parents for their endless support. My deepest gratitude is, of course, to my wife Suzan. She is the co-producer of this happy ending.

3 Table of Contents 1. Do Social Networks Lead to Better Opportunities to Pursue? Evidence from Job A pplications ...... 7 1.1. Introduction ...... 7 1.2. Theoretical and Empirical Background ...... 8 1.3. The Search Process...... 10 1.4. The Current Study ...... 12 1.5. Methods and Data...... 15 1.5.1. Study V ariables ...... 16 1.5.1.1. Search Methods ...... 16 1.5.1.2. Job Q uality ...... 17 1.6 R esults ...... 18 1.6.1. Indirect Tests (Between-Individual Analyses)...... 18 1.6.2. Direct Tests (Within-Individual Analyses)...... 19 1.6.3. Robustness Checks ...... 19 1.6.3.1. Checks with wage ...... 19 1.6.3.2. Check with a sub-sample ...... 20 1.6.3.3. Check with another proxy for job quality ...... 20 1.6.4. Analyses of Occupational and Spatial Scope ...... 21 1.7 D iscussion ...... 22 1.8. References ...... 25 1.9. Tables ...... 28 1.10. Figures ...... 39 1.11. Appendix ...... 41 2. The Emotional Toll of Long-Term Unemployment: Examining the Interaction Effects of Gender and Marital Status ...... 45 2.1. Introduction ...... 45 2.2. The Great Recession and Long-Term Unemployment ...... 46 2.3. Gender, Marital Status and the Emotional Toll of Unemployment ...... 49 2.4. Qualitative Data and Analysis ...... 52 2.5. Quantitative Data and Analysis ...... 59 2.5.1. Descriptive Results ...... 60 2.5.2. Cross-sectional A nalyses ...... 61 2.5.3. Panel Data Analysis ...... 61 2.6. Discussion and Conclusion ...... 65 2.7. R eferences ...... 67

4 2.8. Tables ...... 71 2.9 . F igures ...... 81 3. Activation Programs for Unemployed Insurance Recipients: Comparison of Two Training Delivery Formats ...... 83 3.1. Introduction ...... 83 3.2. Job search Interventions ...... 84 3.3. Interventions and Other Programs for UI Recipients ...... 85 3.4. Does training delivery format matter?...... 87 3.5. T he current study ...... 89 3.6. Methods ...... 90 3.6.1. Sample and Procedure ...... 90 3.6.2. Study V ariables ...... 91 3.7. Results ...... 91 3.8. D iscussion ...... 93 3.9. R eferences ...... 95 3.10. T ables ...... 100 3.11 Figures ...... 108 3.12 Appendix ...... 109 4. Why Is Regulatory Compliance Difficult? Variable Performance and the Insulation of Economically Resourceful Actors ...... 11 4.1. Introduction ...... 111 4.2. Regulatory Compliance through Self-Regulation...... 112 4.2.1. A Form of Self-regulation: EHS Management Systems ...... 113 4.2.2. EHS Management Systems in Academic Laboratories ...... 115 4.3. The Current Study: Academic Laboratories at Eastern University...... 117 4.3.1. Setting ...... 117 4.3.2. Data: The number of safety violations ...... 118 4.4. Results ...... 119 4.4.1. Descriptive Analysis of Violations ...... 119 4.4.2. Inferential Analysis of Violations ...... 120 4.5. Qualitative Data Analysis ...... 122 4.6. Discussion ...... 128 4.7. R eferences ...... 131 4 .8 . Tables ...... 134

5 4.9. Figures ...... 141 4.10. Appendix ...... 144

6 1. Do Social Networks Lead to Better Opportunities to Pursue? Evidence from Job Applications

1.1. Introduction In the labor market both job seekers and employers often utilize social networks in the search process. For job seekers personal contacts are a source of job-related information that may not be otherwise available. For employers contacts and referrals provide difficult to obtain information about the job applicants' productivity and fit. Job seekers' use of social networks in their job search has been extensively studied (for reviews, see, e.g., Fernandez, Castilla, & Moore, 2000, Castilla, 2005, Castilla, Lan, & Rissing, 2013). Researchers have repeatedly found that individuals commonly use social contacts while searching for a job and that social contacts help individuals find employment (Pellizzari, 2010). However, little is known about the precise role of social networks in the search process, and whether the use of social networks is associated with particular kinds of job opportunities. One important but unanswered question in the literature is whether job opportunities pursued through social networks are better (or worse) than those pursued through other methods. This question still remains unanswered partly because researchers lack data on the pool of jobs to which job seekers apply before securing a position. While the job search process is often treated as a black box, the relationship between the use of networks and search outcomes has received considerable research attention. Prior research shows that using social contacts helps individuals find work (Granovetter, 1995). However, when it comes to the quality of the job found through contacts the picture is blurry (Pellizzari, 2010). For instance, while early empirical studies showed that the jobs obtained through social connections were associated with higher wages (see, e.g., Boxman, De Graaf, & Flap, 1991), recent research has found the opposite (see, e.g., Antoninis, 2006). In this study, we propose a close examination of the job search process to help us shed light on this mixed picture in the existing literature. We use unique data on individual search behavior obtained from an applicant tracking system used by hiring companies. Our main focus is to assess whether and how the quality of jobs to which job seekers apply varies depending on the search method. Because this field study dataset includes information on job applications made by the same individual to different jobs using different search methods, we are able to explore within

7 individual variation in job search behavior and directly test whether the same job seeker chooses to apply to jobs with different qualities depending on the search method she uses. The fixed effects analyses help address the issue of unobserved heterogeneity and thus provide a clearer picture of the effect of the chosen job search method on the quality of job pursued. Our analyses show that compared to applications to jobs using formal methods, when job seekers apply for jobs using their social contacts they apply to worse jobs. The jobs that are applied to through networks pay less than jobs that are applied to using formal methods. Furthermore, we are able to examine two mechanisms that might help explain these findings: geographical and occupational diversity of job applications. We find that when job seekers apply for jobs using social networks, they apply to lower number of distinct occupations and to jobs located in a lower number of states compared to when they apply using formal methods. The remainder of the paper is organized as follows. In the next section, we provide some theoretical and empirical background to understand the important question of whether job search quality systematically differs depending on the method used by the job seeker. We then describe our research setting and the job search process under study. We follow with the presentation of our data and quantitative analyses. We conclude with a discussion of the implications of our study for better understanding today's job search process and the role of social contacts within this process.

1.2. Theoretical and Empirical Background According to the Current Population Survey, 26% of unemployed individuals reported that they contacted friends or relatives during their job search (Bureau of Labor Statistics, 2016). If we also consider connections made with acquaintances for job search purposes, this percentage would undoubtedly be much higher. Previous scholarly work has repeatedly confirmed that the use of social ties is common among job seekers and that networks are involved in about half of the job matches (Corcoran, Datcher, & Duncan, 1980; Granovetter, 1995; Marsden & Campbell, 1990). While the use of social networks to find a job is common, its benefits to job seekers is less clear. Previous studies have shown that using social contacts accelerates finding a job. The use of contacts reduce unemployment duration by one to three months on average (Bentolila, Michelacci, & Suarez, 2010). However, findings concerning the quality of jobs obtained through networks are mixed in the existing literature. While early studies show that jobs obtained via

8 social contacts are associated with higher starting wages (Corcoran et al., 1980; Simon & Warner, 1992), more recent research reports the opposite results (Antoninis, 2006; Bentolila et al., 2010; Loury, 2006; Mouw, 2003). The mixed findings regarding wages are mirrored in studies focusing on other job attributes. While one set of studies find some evidence that the jobs found through networks provide higher status than the jobs that are found through formal channels (e.g., Lin, Ensel, Vaughn, 1981), another group of studies finds no advantages for the use of contacts in the job search once other factors were taken into account (see, e.g., Bridges & Villemez, 1986; de Graaf & Flap, 1988). It is important to note that in addition to producing inconclusive empirical evidence, previous studies suffer from serious methodological problems. First, due to their designs, most studies cannot rule out the possibility of unobserved heterogeneity among individuals that might introduce spurious results. Second, past studies tend to select on the dependent variable by focusing only on successful applications which result in offer or hire decisions. Since unsuccessful job applications might also have been made through the same method that yielded successful applications, taking only successful applications into account can lead to inaccurate conclusions (for these methodological discussions, see, e.g., Fernandez & Weinberg, 1997; Montgomery, 1992; Mouw, 2003). Both positive and negative predictions regarding the effect of networks on job quality can find theoretical bases in the scholarly literature. On the one hand, some scholars suggest that social contacts provide better job opportunities than formal methods due to the information conduit role of networks (Rees, 1966). In a labor market where uncertainty is prevalent, social connections help acquire useful information about job opportunities. Thus, the informed job seeker will make a better decision on job choice, leading to better matches. In contrast, social networks generate job opportunities faster than formal methods. Therefore, finding a job easier through networks may convince a job seeker to take a job which does not match with her productivity, ultimately leading to worse matches. Bentolila and his colleagues (2010) provide empirical evidence for this theoretical model. They find that social contacts reduce unemployment duration, however at the same time contacts generate wage discounts of at least 2.5%. Their findings show that job seekers who found a job with the help of contacts often end up with jobs which were out of the job seekers' primary occupational field. Thus, the use of contacts leads to a productivity mismatch since individuals trade higher wages for easier access

9 to jobs. In their review of the literature, loannides and Loury (2004) point out that the main reason for variation among findings concerning the effect of networks on job quality is the heterogeneity involved in the job matches occurring through networks. These authors emphasize that sources of heterogeneity such as contacts, workers, and employers need to be considered in job search-social networks studies. As they note, "wage premiums and wage penalties associated with finding jobs through personal contacts are the joint outcome of firms' recruitment efforts and individuals' job search." (loannides & Loury, 2004, p. 1085). In this study, we are in a rare research position to examine the supply side of the market by focusing on individual search behavior. This approach allows us to isolate two main sources of heterogeneity. First, because we conduct job seeker fixed effect analyses, our results do not suffer from heterogeneity among individuals. Second, because we study job applications which occur prior to any interaction with the potential employer, the heterogeneity among employers is not a concern for our particular study of the effect of social networks on job quality. Consequently, our novel empirical approach allows us to estimate the direct effects of using a particular job search method when pursuing job opportunities.

1.3 The Search Process Job search has become an integral part of lives of workers. Compared to past, people today hold more jobs over their careers and stay shorter with one employer. According to the Associated Press-NORD Center for Public Affairs Research 2016 report, 40% of America's baby boomers held an average of twelve jobs from ages 18 to 48. Half of those aged 65 and above, however only a third of those aged 50 to 64 have stayed with the same employer for at least two decades. In January 2014, the average tenure with the current employer was 7.9 years for people aged 45 to 54, compared to 10.4 years for those aged 55 to 64. We see more action in the external market today (Bidwell & Briscoe, 2010). Individuals experience many more job search spells compared to past due to a variety of reasons. Periodic economic downturns lead to high unemployment rates (Elsby, Bart, & Sahin, 2010). Professionals see boundaryless career as an opportunity to strengthen their skills and bargaining power (Arthur & Rousseau, 1996). Non-standard work arrangements including independent contracting and temporary work reduced the duration of employment contract (Weil, 2014).

10 These factors and many others have made workers spend significant amount of their time for job seeking. While searching for jobs, job seekers use a range of methods including both formal and informal sources. One common method is the use of social networks such as friends, family members, colleagues, and acquaintances. Not only job seekers but also employers utilize information networks. Firms that use a referral program benefit from it. The use of referrals lowers screening costs, generates better applicant pools, and leads to high post-hire productivity (Fernandez & Weinberg, 1997; Fernandez, Castilla, & Moore, 2000; Simon & Warner, 1992). Job seekers' use of social networks has been extensively studied. However in this extensive literature, a large fraction of studies have focused on search outcomes. Even though scholars have repeatedly called for process-oriented studies that investigate individual search behavior, few studies have attempted to open up the black box of job search to provide insights on what is actually happening during the search. In one recent study of job search behavior, Pager and Pedulla (2015) investigate how African-American job seekers differ from White job seekers in targeting (or avoiding) particular job sectors in order to deal with discrimination in the labor market. In contrast to the economic models that propose that disadvantaged groups would self-select into certain jobs to overcome discrimination (Heckman, 1998), these authors show that African-American job seekers widen their search and apply to a broader range of job openings compared to White job seekers. In another study on search process, Kudlyak and her colleagues (2013) examine how job seekers direct their applications over the course of job search. They find that, in the initial period of search, job seekers apply for jobs that are compatible with their education, however, as search continues, they apply for jobs that require less education. Barbulescu (2015) study the effectiveness of different kinds of contacts across stages of MBA job search. She concludes that job seekers with less occupationally focused networks apply to more job types and thus on aggregate obtain more interviews. Obukhova and Lan (2013) take an important step in investigating the relationship between job search methods and employment outcomes by studying individual search behavior. Using fixed effects models, they study the number of applications submitted and the number of interviews and offers received by the same graduating university student who used different search methods. As the authors emphasize, their study provides a useful starting point for

11 comparison across methods while handling the problem of unobserved heterogeneity at the same time: "We can compare the outcomes the same individual achieves through contacts and formal methods" (p. 2207). These fixed effects models show that using contacts does improve job search outcomes for this population. In this study, we not only build on the recent studies of job search, but also overcome some of their empirical limitations. Prior empirical work has tended to sample new university graduates (Obukhova & Lan, 2013) or MBA students (Barbulescu, 2015, Greenberg & Fernandez, 2016). Since university graduates are looking for their first job, it is difficult to generalize their findings to the general population of job seekers. Moreover, if we are interested in investigating the effects of social networks on job search success, a student sample has a few limitations. First, one can argue that social networks as job finding channels are likely to be less important for new graduates since they have not yet build connections with the world of work. Second, it seems reasonable to assume that most students would rely on university career centers as a starting point for their search. Thus, those students who also use their social networks may likely be a self-select group that is putting in extra effort into their search. Because any such extra effort would naturally lead to more and/or better job opportunities, the findings are not surprising. On the other hand, MBA students have unique characteristics which are different from the general job seeker population. First, they mostly depend on alumni network. Second, they search for jobs in very few industries with more than fifty percent of students going into consulting and financial services'. Third, in the MBA market, the cost of search is low since it is mediated by universities (Sterling, 2014). Thus, MBA graduates are looking for work in specific labor markets and their networks might function differently than the general population. Another important limitation of these studies is that they rely on individuals' self-report. Therefore, recall biases and memory errors are important concerns here. Our current study overcomes these limitations by studying records of actual job applications of individuals from a large sample of general job seekers.

1.4 The Current Study As mentioned before, a large fraction of studies on the role of social networks in hiring focused on search outcomes. This obsession with search outcomes is problematic mainly for

I http://www.hbs.edu/recruiting/data/Pages/industry.aspx (retrieved on July 14, 2016).

12 three reasons. First, job search is a multi-staged process. In the initial stage, job seekers identify job opportunities. Then, they make applications to the jobs that they are interested in. Finally, they prepare for and attend interviews. Krueger and Mueller (2011) studied how job seekers spend their time during this process and found that only 5% of time is spent at the last phase which is the interview and offer stage. The authors found that job seekers commit 25% of their time to making applications and the remaining 70% to identifying job opportunities. Thus, a large fraction of search is dedicated to opportunity identification and pursuit, which were often ignored by previous research. Second, the large fraction of search is unsuccessful. Job seekers rarely get the first job that they apply to. In the year we observe search behavior in this study, the average job search duration was ten weeks (BLS, 2010) and time spent in job searching in a typical week was approximately twelve hours (Krueger & Mueller, 2011). Thus, an average job seeker had spent 120 hours to find a job at the time period we study. If researchers only study the jobs that yield interviews or offers, they are at risk of ignoring the large portion of search which does not lead to positive outcomes. Pager and Pedulla (2015) explain this as follows: "The emphasis on wage offers in job search ignores the large fraction of search activity that does not result in a job offer. Decisions by job seekers about where to search -based on some combination of preferences and perceived opportunity- represent an important constraint on the subsequent distribution of offers and, ultimately, an individual's placement in the labor market" (p. 1007). Third, the role of social networks is qualitatively different in each stage of job search (Barbulescu, 2015). In the first stage, social connections may provide job leads while at the last stage they may help getting the job by putting a good word for the candidate. Thus, making inferences about the usefulness of networks for the whole process by only looking at the final stage would be misleading. In this study, we conceptualize and examine "opportunity identification and pursuit" to study the role of networks in the search process. We define opportunity identification and pursuit as the time and effort job seekers invest to learn about and pursue a job. In the current study, we are specifically interested in how the use of social connections affect the quality of opportunity identification and pursuit. Job search is not a uniform activity and job search methods differ in costs (Addison & Portugal, 2002). Each method requires different amounts of time and physical and mental

13 resources. The level of effort required for surfing the Internet to find out vacancies is different from the level of effort required for building new social relationships with others who might be helpful in directing to job leads. While looking for a job, seekers often exploit a wide range of sources including social connections. Social networks provide information about job opportunities not available otherwise. This information may extend the search on extensive and/or intensive margin (Rees, 1966). Networks may help identify a new job opportunity (i.e., extensive margin) or provide in-depth information about a particular opportunity (i.e., intensive margin). Thus, the use of networks has information value, ultimately leading to better opportunities. On the other hand, people's social networks are not usually diverse and they tend to be occupation or location specific. The more individuals gain experience in a specific occupational field, their social contacts tend to be from that field (McDonald, 2011). Information holders with strong within-industry connections have more information to share and actually share them (Marin, 2013). Besides, people often build relationships locally (Mok & Wellman, 2007; Wellman, 1996). Social networks create job opportunities faster than formal methods however aforementioned characteristics of networks make them locally bound sources which impede occupational and spatial mobility. "The availability of social contacts and the opportunity of finding a job more easily may convince a worker to undertake a career in professions, sectors or locations where his abilities are not fully exploited." (Bentolila et al., 2010; p. 20). In addition to the question of how networks affect the quality of opportunities identified and pursued, in this study we look at how networks affect occupational and spatial diversity of these opportunities. As an attempt to answer these questions, we direct our attention to the search process itself. Rather than looking at subsequent labor market outcomes, we investigate the proximal consequence of using a search method, that is, the possible set of jobs available to individual job seekers. We specifically address the following question in this study: Do personal networks do a better job in steering job seekers into high-quality and more diverse job openings than formal methods? This study has several strengths. First, we look at the job search process itself rather than looking at subsequent search outcomes, which might be the results of confounding factors. Second, rather than paying attention to the quantity (number of applications submitted through each method) as done in prior research, we focus on the quality of such jobs. Third, using fixed

14 effects models, we eliminate the threat of unobserved heterogeneity. Fourth, we do not consider only the applications that turned into offers, we consequently do not select on the dependent variable; instead we are taking into account the whole range of applications, both successful and unsuccessful, made by job seekers in the studied period. Fifth, in contrast to previous research which used data on intertemporal job searches (e.g., Mouw, 2002; 2003), we use contemporaneous search data (i.e., search in the same spell). Therefore, our results are not affected by changes in individual characteristics. Sixth, our data include actual applications made by job seekers, thus eliminating self-report errors. Finally, our large sample size improves the reliability (and possibly generalizability) of our findings.

1.5 Methods and Data The data we analyze in this study come from a U.S. private company which provides an applicant tracking system (ATS) to companies and organizations that are hiring candidates. The dataset consists of the entire population of applications made to 561 different organizations through this ATS over one year (from September 2010 to September 2011). Applicant tracking systems are commonly used by companies nowadays (Cappelli, 2013). These systems help firms manage applications and process hiring phases easier and cheaper. They also facilitate communication between incumbents (e.g., HR department, hiring managers, upper level management), and help manage Equal Employment Opportunity related data. Our data include job applications made to hiring companies that share the same ATS. Our unit of analysis is the application made to a particular position. The number of applications in the whole data set is 5,747,997. These applications had been made by 3,478,720 different candidates. Seventy-six percent of the candidates submitted one application using this ATS. The number of candidates who submitted more than one application is 824,501. The variables included in our dataset are job application identification number (ID), the date on which the job advertisement was posted online, the company ID, the job title, the level of education required for the job, and the job seeker ID. Key to this study, the dataset also contains a variable recording the original source of job information. We now turn to describing these key variables in detail.

15 1.5.1. Study Variables 1.5.1.1. Search Methods. The data include information on the source through which the job seeker learned about the job opportunity. In the original dataset, the information on sources is very specific and is entered in open-text format (e.g., "job board at courthouse"; "newspaper print ad"). For the purpose of the analyses we present here, we coded these 57,649 unique entries into five broad job search categories. Table 1 provides the frequencies of the source of information in all applications. The first category, formal methods, includes all formal methods such as the Internet, employment agencies, and newspaper ads. In coding refererals, we distinguished between employee referrals (that is, the practice of recruiting new workers via employee referrals) from non-employee referrals since the outcomes resulting from these two sources of referrals might differ. The literature suggests that employee referrals are often preferred over other candidates since they have already been pre-screened by the referrer employee who has a good knowledge of the work setting and whose reputation depends on the quality of the candidate she refers (Rubineau & Fernandez, 2013). Additionally, using employee referrals may accelerate the socialization of newcomers into the company (Fernandez et al., 2000). In our data, the employee referral category consists of cases where the job seeker is referred by an employee of the hiring organization. The non-employee referral category includes social connections that do not work in the targeted organization. We also created a distinct category for internal candidates since this group is in a unique situation; they often seek promotion and wage increase. As discussed by Fernandez and Weinberg (1997), internal candidates are an interesting comparison group for the referrals. Since internal candidates are in a more advantageous position than external referrals, they form an upper-bound comparision group. The last category we created is "other methods" and this category includes observations which cannot be placed into the other categories, such as "prefer not to answer" or "available upon request." For the main descriptive and inferential analyses, we treated this category as missing and did not include it in the analyses. The distribution of methods in our population of job opportunities show that 84% of all applications had been made through formal methods. This distribution of search methods in our data does not substantially differ from distributions reported in other studies. For instance, Krueger and Mueller (2011) found that unemployed

16 people dedicated only 9% of their time to contacting networks. The distribution of methods used by individuals who made more than one application is provided in Table 2.

1.5.1.2. Job Quality. In this study, our aim is to investigate the effect of using different job search methods on the quality of the job to which job seekers apply. We operationalize the quality of jobs by wage estimates. As Montgomery (1992) describes, "When the nonpecuniary aspects of employment differ across jobs, the 'wage' might be interpreted as a broader index of job quality" (p. 586). However, because our data do not contain information about the actual wages offered for the job, we deduced wage information from job titles. We adopted the approach developed by Pager and Pedulla (2015) in order to determine the quality of jobs using job titles. The data include 130,776 unique job titles in open-text format (e.g., "transportation planner," "restaurant manager"). First, we coded these job titles into 96 Standard Occupational Classification (SOC) codes generated by the Bureau of Labor Statistics. Since companies have different ways of labeling jobs, we only coded those titles that exactly matched SOC titles to avoid miscategorization. 2 We left out titles that were ambiguous or difficult to assign to an occupational category. Thus, we were able to code 11,702 job titles which cover 2,430,630 job applications (42% of all applications in the whole data set). To determine the quality of the job that job seekers applied to, we matched the occupational codes to wage estimates for each category provided by Occupational Employment Statistics (OES) Survey (for more information, go to www.bls.gov/oes/). The occupational categories into which we coded job titles, their distribution in the data, as well as their respective annual median wages obtained from OES, can be found in the Appendix A. In our analyses, we used natural logarithms of annual median wages. An example illustrates the job title coding and wage matching process. The dataset includes applications made to a job entitled "occupational therapist." The occupational code in the 2010 Standard Occupational Classification for this job is 29-1122. The broad group to which this occupation belongs is called "therapists," and the code for this group is 29-1120. The name of the minor group that includes the "Therapists" broad group is "Health Diagnosing and Treating Practitioners" and its code is 29-1000. By looking at Occupational Employment

2The Bureau of Labor Statistics also provides "Major Groups" categorization including 23 categories. We used "Minor Groups" categorization of 96 categories to better capture the variation.

17 Statistics, we figured out that the annual median wage at the national level for this minor group of "Health Diagnosing and Treating Practitioners" is $73,410. We coded the wages for other evident job titles in this way. The wages for coded applications range from $18,460 to $110,740, with the mean of $56,622. These numbers indicate that the jobs covered in the data represent a wide spectrum of jobs from low-paying jobs to high-paying ones. In addition, as seen in Appendix A, every occupational category is represented in the data with larger categories being office and administrative support occupations (27%), management occupations (24%), and business and financial operations occupations (14%).

1.6 Results We begin by providing descriptive statistics for our key study variables. Table 3 shows means and standard deviations as well as the correlations. These descriptive statistics provide an initial overview of the data. Several points are worth mentioning in this table. First, all correlations among variables are significant at the 0.001 level. Second, referrals -both employee referrals and non-employee referrals- have negative associations with the national median wage. The wages of the jobs that people applied to via referrals are lower than those of jobs they applied through other means. Figure 1 shows mean wage per application by search channel. As seen in the figure, mean wage for formal methods is $56,633, followed by non-employee referral with the mean of $54,732, which is followed by employee referral with the mean of $53,345. Figure 2 shows the distribution of mean wage per application by different channels of formal methods. For inferential analyses, we first conducted indirect tests (henceforth, referred as between-individual analyses). Then we ran direct tests (within-individual analyses). These analyses are described in detail in the next three sub-sections.

1.6.1. Indirect Tests (Between-Individual Analyses) We first ran an indirect test for our dependent variable, the logarithm of median wages, henceforth log-wages. These results are shown in Table 4.3 We regressed log-wages on search methods for observations for which we have wage information (N=2,242,525). The OLS

3We report robust estimators (Huber-White sandwich estimators).

18 regression results show that the use of non-employee referrals leads to a 3 percent decrease in median wages compared to formal methods, and the use of employee referrals leads to a 6 percent decrease in median wages compared to formal methods. These findings from between-individual analyses suggest that when job seekers use social contacts they apply for lower quality jobs than when they use formal methods. This said, these are preliminary and not necessarily convincing results yet as we do not have any controls for the human and social capital of job seekers.

1.6.2. Direct Tests (Within-Individual Analyses) Since we have longitudinal data of individuals who were observed in at least two job applications made through different search methods, we are able to run individual fixed effect models to investigate the effect of using different search methods on job quality for the same individual. The within-individual analyses net out all observed and unobserved individual differences among job seekers, including human and social capital as well as demographic characteristics such as race and gender (for similar empirical approaches in labor market studies, see, e.g., Mouw, 2006; Obukhova & Lan, 2013; Yakubovich, 2005). We ran fixed effects model with the dependent variable log-wages (see Table 5). The results of this model show that when the job seeker uses a non-employee referrer, the job she applies to pays 1 percent less than the job she applies to through formal methods. When she uses an employee referrer, this percentage increases to 2 percent. The coefficient for the internal candidate category is positive and significant (p<0.01). This finding is not surprising given that internal candidates often seek promotions.

1.6.3. Robustness Checks 1.6.3.1. Checks with wage: To verify our findings, we conducted several robustness checks. First, we ran the same fixed effects model with log-wages by adding the "Other" category in the model. As mentioned in the Methods section, this category includes miscellaneous entries in the data that cannot be categorized in any of the search method category (e.g., "prefer not to answer", "N/A"). As seen in Table 6, adding this category into the regression did not change the results. Second, we combined non-employee referral and employee referral categories into one

19 category as "Referral." Regressing log-wages on this category yielded same results as well (see Table 7).

1.6.3.2. Check with a sub-sample: In the literature, social networks are thought to provide more benefits to high-educated, high-skilled individuals than low educated, low-skilled ones (e.g., Lin, 1999; Marsden, 2001). We also tested this argument with an additional analysis. In the data set, we have no information on any individual-level characteristics of job seekers. However, because we do have information on the required education level for the position, we coded a candidate as graduate degree holder if she made at least one application to a job which requires a graduate degree. Since candidates who do not have a graduate degree would not apply for a job which requires graduate degree, we were able to estimate the education level for a particular group of candidates in the data set. Note that because individuals who apply to jobs that require a high school degree can hold a high school degree or a higher degree, we were not able to apply this method to candidates who never applied to a graduate degree job, but applied to jobs that require high school degree.4 This flagging of individuals who presumably hold graduate degree allowed us to test the argument that high- skilled individuals achieve better labor market outcomes as a result of using social networks, but low-skilled individuals do not. Contrary to this argument, we find that graduate degree holders experience a wage penalty when they use their social contacts (see Table 8).

1.6.3.3. Check with another proxy for job quality: Our data include information on the required level of education for the job that is posted in the job advertisement. In the dataset, we have the information of required level of education for 3,678,168 observations (64% of all applications in the whole data set). As can be seen in Table 9, this variable is recorded using six categories: High-school diploma, 2-year degree, 3- year degree, 4-year degree, post graduate, and doctoral degree. Among the jobs for which we have required education information, almost 44% of applications were made to positions that require high school diploma, and 43% of applications were made to openings that require a four- year degree.

4Same applies to job applications made to jobs that require college degree.

20 Since the required education level provide information about the quality of the job (the more required education the better job), as a robustness check, we looked at whether the use of referrals lead to jobs that require less or more education. In the within-job seeker fixed effects model (see Table 10), we regressed required education level for the job on search methods. The results show that the job to which a job seeker applies as a non-employee referral requires less education than the job to which she applies through formal methods. It is important to note that although employee-referral category also has a negative coefficient, the effect is extremely small and not significant (p>0.01).

1.6.4. Analyses of Occupational and Spatial Scope One possible mechanism that might help explain our findings is the spatial narrowing effect of social networks when job seekers search for jobs. In particular, if individuals' social connections show limited geographic dispersion, using social contacts in job search might lead to limited number of job opportunities in different locations, and ultimately lower quality of jobs on average. Based on job location information available in our data set, we compared the number of different states that the same job seeker applied to using different search methods. The first column of Table 11 provides results from this analysis. We find that when job seekers used social networks (both non-employee referral and employee referral), they tend to apply to jobs in a lower number of states compared to when they apply to jobs using formal methods. Thus, because individuals' social networks are geographically constrained, they seem to produce job opportunities in a lower number of states compared to opportunities generated through formal methods such as the Internet and employment agencies. Especially when searching for jobs using the Internet, job seekers are able to access information about job opportunities in different states where they may or may not have social ties. Another possible mechanism is the narrowed occupational scope of the search as a result of using networks. Individuals do not have social contacts in every occupation. Depending heavily on contacts might prevent job seekers from exploiting a wide range of opportunities in different occupations. Our analyses show that when job seekers use their contacts, they apply to lower number of distinct occupational categories (see Table 11, column 2).

21 1.7 Discussion While individuals make decisions about their job search strategies, various social, economic and physical constraints and opportunities substantially affect the resulting distribution of job opportunities (Pager & Pedulla, 2015). By looking at the supply-side, the goal of this study is to shed light on the direct effects of social connections on search behavior itself. Specifically, we explore the question of whether job opportunities pursued through social networks are better or worse than those pursued through formal methods such as employment agencies, newspaper advertisements, and the Internet. In this study, we have responded to calls for more search process oriented research (Pager & Pedulla, 2015; Barbulescu, 2015). One important and distinctive aspect of this study is that our dataset contains records of actual individual search behavior. By running individual fixed effects models, we were able to examine how the same job seeker is steered into different job opportunities as a result of using different methods. Therefore, both employer and job seeker heterogeneity are not concerns in our study. The results of this study show that when job seekers use their social contacts they apply to worse jobs in terms of observable characteristics. The jobs that are applied to through networks require less education and pay less than those applied to through formal channels. Our results are in line with findings from recent studies that utilized tighter methodologies to figure out the direct effects of social networks on labor market outcomes (e.g., Greenberg & Fernandez, 2016). In contrast with the common knowledge about the benefits of social networks, this recent line of studies have reported that job seekers are worse off in monetary terms when they use their connections. Similar to recent empirical findings, we find that social connections of job seekers direct them to a different segment of the labor market than where they search through formal channels, which ultimately might lead to lower quality matches. We also find support for the explanation that because individuals' social networks are geographically constrained, they produce opportunities in a lower number of states compared to opportunities generated through formal methods. Because a significant portion of social interactions occur between physically close individuals (Wellman, 1996), the geographical breadth of jobs learned through social ties is likely to be narrower than jobs learned through formal methods. In addition, our findings provide support for occupational homophily (McPherson, Smith-Lovin, & Cook, 2001). The use of networks limits the number of distinct

22 jobs that job seekers apply to. Consequently, the geographical and occupational narrowness of search may lead to a discrepancy between the capabilities of the seeker and the requirements of the job she accepts, contributing to productivity mismatch. Networks provide valuable information on non-monetary aspects of the job. It is possible that job seekers might be substituting non-monetary information that they obtain through networks for monetary rewards. Previous research shows that job seekers who use their social connections gather high-quality information about the job and the organization than job seekers who use formal methods (Granovetter, 1995). It is plausible to expect that through social networks, job seekers might obtain better information about the unobservable characteristics of the job (i.e., job-person fit), which ultimately influences their application decisions even if the job itself pays less. Recent research has provided support for this argument. Using European data, Franzen and Hangartner (2006) show that individuals who found a job through social connections experience monetary disadvantages but enjoy non-monetary advantages. The jobs that they found through networks are more compatible with their abilities and provide better career prospects. This finding is consistent with the literature showing a positive association between the use of social networks in hiring and the filling of job vacancies with workers who are a good fit in terms of skills and preferences (see, e.g., Fernandez et al., 2000; Castilla, 2005). Similarly, in a recent study, Greenberg and Fernandez (2016) find that although MBA graduates' search through networks result in offers with lower pay, these graduates are more likely to accept these offers due to perceived growth potential in these jobs. Our results also provide insight on this issue. The fixed effects analyses show that the use of employee referrals lead to 2% decrease in log-wages whereas the use of non-employee referrals lead to 1% decrease. Because employee referrals provide deeper information about the job and the organization than non-employee referrals, this difference in reductions confirms findings from recent studies that as the information acquired intensifies, the monetary benefit traded off increases. One possible explanation for our results is that job seekers might be overestimating their capabilities and thus might be applying to higher-quality jobs than they can really handle. Their contacts might have a more realistic view of job seekers' competencies. Falk, Huffman, and Sunde (2006) question the standard search theory which assumes that individuals are perfectly informed about their relative abilities. Their findings show that individuals are uncertain about

23 their abilities at the beginning of the search and as the search continues they update their beliefs about their abilities. Almost 65 years ago, Reynolds and Shister, in their book Job Horizons, found that non- wage factors are of great importance for workers. They showed that job seekers give much weight to factors like "physical characteristics of the job," "the degree of independence and control associated with the job," "the fairness of treatment", "the quality of relationships with fellow workers" and "the degree of job interest." It is information about these kinds of job attributes that are conveyed through social contacts and which may not be otherwise available (Fernandez & Weinberg, 1997). Personal contacts might tell "if prospective workmates are congenial, if the boss is neurotic, and if the company is moving forward or is stagnant" (see Granovetter, 1995, p. 13). Moreover, our results speak to social resources theory as well. This theory proposes that individuals with personal resources such as a high-education or a high-status family background would obtain greater benefits from using social contacts than individuals who do not have these resources (Lin 1999, Lin and Dumin 1986). However, the current study shows that this advantage does not appear for highly educated individuals at the application stage, introducing boundary conditions for the applicability of social resources theory. This study also adds to studies emphasizing the need for within-individual study designs to improve our understanding of the role of networks in the job search (see, e.g., Mouw, 2006; Obukhova & Lan, 2013; Yakubovich, 2005). Since the within-individual method controls for between-individual heterogeneity, this methodology can help clarify network causality in the labor market. Finally, this study has practical implications for the study of job search processes and outcomes. Our findings imply that formal methods do appear to do a better job than informal methods in steering job seekers into pursuing better jobs. In the contemporary job search and recruitment arena, the most widely used formal method is the Internet. Our findings also imply that job seekers can take the advantage of using online job search boards actively in order to be informed about high quality job opportunities spanning through geographies and occupational fields. Our results also draw attention to downsides of heavy dependence on networks. It is important to make job seekers and career counselors aware of these potential downsides. At the macro level, policies that loose heavy dependence on networks and that promote occupational and spatial mobility would be effective.

24 1.8. References Addison, J. T., & Portugal, P. (2002). Job search methods and outcomes. Oxford Economic Papers,54(3), 505-533. Antoninis, M. (2006). The wage effects from the use of personal contacts as hiring channels. Journal of Economic Behavior & Organization, 59(1), 133-146. Arthur, M. B., & Rousseau, D. M. (2001). The boundaryless career:A new employment principlefor a new organizationalera. New York: Oxford University Press. Barbulescu, R. (2015). The strength of many kinds of ties: Unpacking the role of social contacts across stages of the job search process. OrganizationScience, 26(4), 1040-1058. Bentolila, S., Michelacci, C., & Suarez, J. (2010). Social contacts and occupational choice. Economica, 77(305), 20-45. Bidwell, M., & Briscoe, F. (2010). The dynamics of interorganizational careers. Organization Science, 21(5), 1034-1053. Boxman, E. A., De Graaf, P. M., & Flap, H. D. (1991). The impact of social and human capital on the income attainment of Dutch managers. Social Networks, 13(1), 51-73. Bridges, W. P., & Villemez, W. J. (1986). Informal hiring and income in the labor market. American SociologicalReview, 51(4), 574-582. Bureau of Labor Statistics (2016) Current Population Survey. Retrieved from https://www.bls.gov/cps/cpsaat33.htm Cappelli, Peter (2013). We can now automate hiring. Is that good? Harvard Business Review. Retrieved from https://hbr.org/2013/12/we-can-now-automate-hiring-is-that-good Castilla, E. J. (2005). Social Networks and Employee Performance in a Call Center. American Journal of Sociology, 110(5), 1243-1283. Castilla, E. J., Lan, G. J., & Rissing, B. A. (2013). Social networks and employment: Mechanisms. Sociology Compass, 7(12), 999-1012. Corcoran, M., Datcher, L., & Duncan, G. (1980). Information and influence networks in labor markets. In G. Duncan & J. Morgan (Eds.), Five thousandAmerican families: Patternsof economic progress (1-37). Ann Arbor, MI: Institute for Social Research. De Graaf, N. D., & Flap, H. D. (1988). "With a little help from my friends": Social resources as an explanation of occupational status and income in West Germany, The Netherlands, and the United States. Social Forces, 67(2), 452-472. Elsby, M. W., Hobijn, B., & Sahin, A. (2010). The labor market in the Great Recession. In D. Romer & J. Wolfers (Eds). Brookings Paperson Economic Activity, 1-48. Washington, DC: Brookings Inst. Falk, A., Huffman, D., & Sunde, U. (2006). Self-Confidence and Search. IZA DP No. 2525. Institutefor the Study of Labor. Fernandez, R. M., & Weinberg, N. (1997). Sifting and sorting: Personal contacts and hiring in a retail bank. American Sociological Review, 62(6), 883-902. Fernandez, R. M., Castilla, E. J., & Moore, P. (2000). Social capital at work: Networks and employment at a phone center. American Journalof Sociology, 105(5), 1288-1356. Franzen, A., & Hangartner, D. (2006). Social networks and labour market outcomes: The non- monetary benefits of social capital. EuropeanSociological Review, 22(4), 353-368. Granovetter, M. (1995). Getting a job: A study of contacts and careers. Chicago, IL: University of Chicago Press. Greenberg, J., & Fernandez, R. M. (2016). The Strength of Weak Ties in MBA Job Search: A Within-Person Test. SociologicalScience, 3, 296-316.

25 Heckman, J. (1998). Detecting discrimination. Journal of Economic Perspectives 12(2):101-16. Holzer, H. J. (1988). Search method use by unemployed youth. Journal of Labor Economics, 6(1), 1-20. loannides, Y. M., & Datcher Loury, L. (2004). Job information networks, neighborhood effects, and inequality. Journal of Economic Literature,42(4), 1056-1093. Krueger, A. B., & Mueller, A. (2011). Job search, emotional well-being, and job finding in a period of mass unemployment: Evidence from high-frequency longitudinal data. Brookings Papers on Economic Activity, 1, 1-57. Kudlyak, M., Lkhagvasuren, D., & Sysuyev, R. (2013). Systematic Job Search: New Evidence from Individual Job Application Data. FederalReserve Bank of Richmond Working Paper, 12(03): 1-40. Kuhn, P., & Skuterud, M. (2004). Internet job search and unemployment durations. The American Economic Review, 94(1), 218-232. Lin, N. (1999). Social networks and status attainment. Annual Review of Sociology, 25(1), 467- 487. Lin, N., & Dumin, M. (1986). Access to occupations through social ties. Social Networks, 8(4), 365-385. Lin, N., Ensel, W. M., & Vaughn, J. C. (1981). Social resources and strength of ties: Structural factors in occupational status attainment. American SociologicalReview, 46(4), 393-405. Loury, L. D. (2006). Some contacts are more equal than others: Informal networks, job tenure, and wages. Journal of Labor Economics, 24(2), 299-318. Marin, A. (2013). Who can tell? Network diversity, within-industry networks, and opportunities to share job information. In SociologicalForum, 28(2), 350-372. Marsden, P. V. (2001) Interpersonal ties, social capital, and employer staffing practices. N. Lin N, K. S. Cook, & R. S. Burt (Eds.) Social Capital:Theory and Research, 105-25. New York: Aldine de Gruyter. Marsden, P. V., & Campbell, K. (1990). Recruitment and selection processes: The organization side of job searches. R. Breiger (Eds), Social Mobility and Social Structure, 59-79. New York: Cambridge University Press. Marsden, P. V., & Gorman, E. H. (2001). Social networks, job changes, and recruitment. I. Berg, A. L. Kalleberg, (eds). Sourcebook of Labor Markets: Evolving Structures and Processes. Kluwer Academic/Plenum Publishers, New York, 467-502. McDonald, S. (2011). What you know or who you know? Occupation-specific work experience and job matching through social networks. Social Science Research, 40(6), 1664-1675. McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27(1), 415-444. Mok, D., & Wellman, B. (2007). Did distance matter before the Internet? Interpersonal contact and support in the 1970s. Social Networks, 29(3), 430-461. Montgomery, J. D. (1992). Job search and network composition: Implications of the strength-of- weak-ties hypothesis. American SociologicalReview, 57(5), 586-596. Mouw, T. (2002). Racial differences in the effects of job contacts: Conflicting evidence from cross-sectional and longitudinal data. Social Science Research, 31(4), 511-538. Mouw, T. (2003). Social capital and finding a job: Do contacts matter? American Sociological Review, 68(6), 868-898. Mouw, T. (2006). Estimating the causal effect of social capital: A review of recent research. Annual Review of Sociology, 32, 79-102.

26 Obukhova, E., & Lan, G. (2013). Do job seekers benefit from contacts? A direct test with contemporaneous searches. Management Science, 59(10), 2204-2216. Pager, D., & Pedulla, D. S. (2015). Race, self-selection, and the job search process. American Journalof Sociology, 120(4), 1005-1054. Pellizzari, M. (2010). Do friends and relatives really help in getting a good job? ILR Review, 63(3), 494-510. Rees, A. (1966). Information networks in labor markets. The American Economic Review, 56(1/2), 559-566. Reynolds, L., & Shister, J. (1949). Job horizons. New York, NY: Harper Bros. Rubineau, B., & Fernandez, R. M. (2013). Missing links: Referrer behavior and job segregation. Management Science, 59(11), 2470-2489. Simon, C. J., & Warner, J. T. (1992). Matchmaker, matchmaker: The effect of old boy networks on job match quality, earnings, and tenure. JournalofLabor Economics, 10(3), 306-330. Sterling, A. D. (2014). Friendships and search behavior in labor markets. Management Science, 60(9), 2341-2354. Wang, S. Y. (2013). Marriage networks, nepotism, and labor market outcomes in China. American Economic Journal:Applied Economics, 5(3), 91-112. Weil, D. (2014). The fissured workplace. Harvard University Press. Wellman, B. (1996). Are personal communities local? A Dumptarian reconsideration. Social Networks, 18(4), 347-354. Yakubovich, V. (2005). Weak ties, information, and influence: How workers find jobs in a local Russian labor market. American SociologicalReview, 70(3), 408-421.

27 1.9. Tables

Table 1. Distribution of Search Methods in All Job Applications Source of information Freg. Percent Formal methods 4,824,479 83.93 Non-employee referral 221,703 3.86 Employee referral 75,206 1.31 Internal candidate 152,807 2.66 Other 473,802 8.24 Total 5,747,997 100%

28 Table 2. Distribution of Methods for Individuals Who Made More Than One Application Source of information Number of individuals Only social networks 16,936 Only formal methods 695,893 Formal methods + social networks 60,926 Internal candidate 7,415 Total 824,501

29 Table 3. Means, Standard Deviations and Correlations among Study Variables Required Non- Log education Formal employee Employee N of obs. Mean SD Min Max wages level methods referral referral

Log wages 2,430,630 10.84 0.46 9.82 11.61

Required education level 3,678,168 2.05 0.99 1 4 0.54*

Formal methods 5,274,195 0.91 0.27 0 1 0.005* 0.04*

Non- employee referral 5,274,195 0.04 0.20 0 1 -0.01* -0.03* -0.68*

Employee referral 5,274,195 0.01 0.11 0 1 -0.01* -0.01* -0.39* -0.02*

Internal candidate 5,274,195 0.02 0.16 0 1 0.01* -0.02* -0.56* -0.03* -0.02* Note. Because we have different number of missing values for the variables, number of pairs for each correlation coefficient might differ. For the required education level, higher score indicates higher education. *p<0 .0 0 1

30 Table 4. OLS Regression Model Predicting the Logarithm of Wage

Log-wage Non-employee referral -0.03* (0.001) Employee referral -0.06** (0.002) Internal candidate 0.05** (0.001) Constant 10.84** (0.0003)

0.0007 Number of Observations 2,242,525 Note. Standard errors are in parentheses. Formal methods is the omitted category. ** p<0.01, *p<0.05

31 Table 5. Fixed Effects Models Predicting Quality of Jobs Log-wage

Non-employee referral -0.01** (0.003) Employee referral -0.02** (-0.005) Internal candidate 0.04** (0.004) Constant 10.84** (0.0002)

Number of observations 2,242,525 Number of groups 1,635,564 Note. Standard errors are in parentheses. Formal methods is the omitted category. * *p<0 .0 1

32 Table 6. Robustness Check: Fixed Effects Model with "Other" Log-wage Non-employee referral -0.01 * * (0.003) Employee referral -0.02* * (-0.005) Internal candidate 0.04* (0.004) Other 0.01* (0.002) Number of observations 2,430,630 Number of groups 824,501 Note. Standard errors are in parentheses. Formal methods is the omitted category. **p<0.01, *p<0.05

33 Table 7. Robustness Check: Fixed Effects Model with "Referral" Log-wage Referral -0.014** (0.002) Constant 10.85** (0.000) Number of observations 2,242,525 Number of groups 60,926 Note. Standard errors are in parentheses. Formal methods is the omitted category. * *p<0 .0 1

34 Table 8. Robustness Check: Fixed Effects Model with Graduate Degree Holders Log-wage Non-employee referral -0.0327* (-0.0103)

Employee referral -0.126* (-0.0187)

Internal candidate 0.00179 (-0.014)

Constant 10.95* (-0.00113)

Number of observations 116,962 Number of groups 55,914 Note. Standard errors are in parentheses. Formal methods is the omitted category. * p<0.0 1

35 Table 9. Distribution of Required Education Level for the Job in All Applications Required education level for the job Freq. Percent High school 1,602,076 43.56 2 year degree 359,531 9.77 3 year degree 20,808 0.57 4 year degree 1,577,679 42.89 Post graduate 103,944 2.83 Doctoral 14,130 0.38

Total 3,678,168 100%

36 Table 10. Fixed Effects Models Predicting Required Education Required Education Level Non-employee referral -0.11** (0.005) Employee referral -0.0007 (0.010) Internal candidate -0.14** (0.006) Constant 2.07** (0.0005)

Number of observations 3,382,252

Number of groups 2,308,530 Note. Standard errors are in parentheses. Formal methods is the omitted category. **p< 0 .0 1

37 Table 11. Fixed Effects Regression Models Predicting Geographical and Occupational Dispersion Number of Number of Distinct Occupational States Categories Non-employee referral -0.48*** -0.78 ** * (0.007) (0.020) Employee referral -0.49* * * -0.88*** (0.013) (0.036) Internal candidate -0.39*** -0.51*** (0.009) (0.026) Constant 1.24*** 1.52*** (0.0008) (0.002)

Number of observations 3,329,160 2,242,525

Number of groups 824, 501 824, 501 Note. Standard errors are in parentheses. Formal methods is the omitted category. ***p<. 0 0 1

38 1.10. Figures

Figure 1. Mean Wage in Dollars per Application by Search Channel

60000.00 58976.19

58000.00 56633.83

56000.00 54732.35

54000.00 53345.44

52000.00

50000.00 Employee Non-employee Formal methods Internal referral referral candidate

39 Figure 2. Mean Wage in Dollars per Application by Formal Methods 70000 62531 56691 56836 57408 58301 600(X) 54096 48173 50000 42932 40000 30000 20(XX) ...... 10000 I 0

A&

&

40 1.11. Appendix Appendix A. Distribution of Occupational Titles and Respective Annual Medial Wages (N=2,430,630) under Study SOC Annual Median Code Occupation Title Frequency Percent Wage (from OES) 11-1000 Top Executives 30,709 1.26 99,550 11-2000 Advertising, Marketing, Promotions, Public Relations, and Sales Managers 78,258 3.22 107,950 11-3000 Operations Specialties Managers 118,146 4.86 100,720 11-9000 Other Management Occupations 404,934 16.66 81,940 13-1000 Business Operations Specialists 174,765 7.19 62,230 13-2000 Financial Specialists 166,706 6.86 63,070 15-1100 Computer Occupations 76,024 3.13 76,270 15-2000 Mathematical Science Occupations 13,448 0.55 76,270 17-1000 Architects, Surveyors, and Cartographers 307 0.01 66,380 17-2000 Engineers 15,739 0.65 86,200 17-3000 Drafters, Engineering Technicians, and Mapping Technicians 12,354 0.51 51,720 19-1000 Life Scientists 15,355 0.63 68,780 19-2000 Physical Scientists 9,856 0.41 74,850 19-3000 Social Scientists and Related Workers 3,293 0.14 69,290 19-4000 Life, Physical, and Social Science Technicians 15,990 0.66 41,130 Counselors, Social Workers, and Other Community and Social Service 21-1000 Specialists 24,029 0.99 40,400 23-1000 Lawyers, Judges, and Related Workers 596 0.02 110,740 23-2000 Legal Support Workers 12,190 0.50 46,850 25-1000 Postsecondary Teachers 1,025 0.04 62,390 25-2000 Preschool, Primary, Secondary, and Special Education School Teachers 4,756 0.20 52,240 25-3000 Other Teachers and Instructors 27,471 1.13 30,100 25-4000 Librarians, Curators, and Archivists 3,372 0.14 44,320 25-9000 Librarians, Curators, and Archivists 5,990 0.25 44,320 27-1000 Art and Design Workers 43,206 1.78 42,250 27-2000 Entertainers and Performers, Sports and Related Workers 232 0.01 38,530 27-3000 Media and Communication Workers 14,846 0.61 50,930 27-4000 Media and Communication Equipment Workers 10,902 0.45 39,450 29-1000 Health Diagnosing and Treating Practitioners 65,584 2.70 73,410 29-2000 Health Technologists and Technicians 37,872 1.56 40,380 29-9000 Other Healthcare Practitioners and Technical Occupations 7,118 0.29 54,130

41 31-1000 Nursing, Psychiatric, and Home Health Aides 20,656 0.85 23,030 31-2000 Occupational Therapy and Physical Therapist Assistants and Aides 1,612 0.07 42,120 31-9000 Other Healthcare Support Occupations 19,321 0.79 30,620 33-2000 Fire Fighting and Prevention Workers 238 0.01 45,590 33-3000 Law Enforcement Workers 3,447 0.14 49,770 33-9000 Other Protective Service Workers 1,648 0.07 24,230 35-1000 Supervisors of Food Preparation and Serving Workers 3,499 0.14 30,120 35-2000 Cooks and Food Preparation Workers 4,040 0.17 20,090 35-3000 Food and Beverage Serving Workers 9,930 0.41 18,460 35-9000 Other Food Preparation and Serving Related Workers 3,841 0.16 18,530 37-1000 Supervisors of Building and Grounds Cleaning and Maintenance Workers 4,614 0.19 37,610 37-2000 Building Cleaning and Pest Control Workers 10,948 0.45 21,570 37-3000 Grounds Maintenance Workers 1,697 0.07 23,970 39-2000 Animal Care and Service Workers 2,219 0.09 19,970 39-3000 Entertainment Attendants and Related Workers 2,640 0.11 18,850 39-5000 Personal Appearance Workers 42 0 22,190 39-6000 Baggage Porters, Bellhops, and Concierges 2,969 0.12 22,880 39-7000 Tour and Travel Guides 6,567 0.27 24,430 39-9000 Other Personal Care and Service Workers 12,329 0.51 20,630 41-1000 Supervisors of Sales Workers 635 0.03 39,350 41-2000 Retail Sales Workers 5,967 0.25 19,890 41-3000 Sales Representatives, Services 84,652 3.48 51,080 41-9000 Other Sales and Related Workers 3,266 0.13 28,600 43-1000 Supervisors of Office and Administrative Support Workers 5,823 0.24 49,330 43-2000 Communications Equipment Operators 3,054 0.13 25,930 43-3000 Financial Clerks 88,478 3.64 32,680 43-4000 Information and Record Clerks 200,110 8.23 29,710 43-5000 Material Recording, Scheduling, Dispatching, and Distributing Workers 12,011 0.49 28,380 43-6000 Secretaries and Administrative Assistants 235,148 9.67 35,330 43-9000 Other Office and Administrative Support Workers 117,449 4.83 28,590 45-2000 Agricultural Workers 329 0.01 19,020 47-1000 Supervisors of Construction and Extraction Workers 3,110 0.13 59,700 47-2000 Construction Trades Workers 6,414 0.26 38,970 47-3000 Helpers, Construction Trades 478 0.02 26,570 47-4000 Other Construction and Related Workers 645 0.03 39,620

42 47-5000 Extraction Workers 5,455 0.22 40,650 49-2000 Electrical and Electronic Equipment Mechanics, Installers, and Repairers 44,931 1.85 46,550 49-3000 Vehicle and Mobile Equipment Mechanics, Installers, and Repairers 114 0 38,470 49-9000 Other Installation, Maintenance, and Repair Occupations 31,202 1.28 38,750 51-2000 Assemblers and Fabricators 9,330 0.38 28,580 51-3000 Food Processing Workers 234 0.01 24,380 51-4000 Metal Workers and Plastic Workers 7,597 0.31 35,380 51-5100 Printing Workers 25 0 34,100 51-6000 Textile, Apparel, and Furnishings Workers 8,293 0.34 22,000 51-8000 Plant and System Operators 479 0.02 53,820 51-9000 Other Production Occupations 282 0.01 29,330 53-1000 Supervisors of Transportation and Material Moving Workers 593 0.02 48,910 53-2000 Air Transportation Workers 4,946 0.20 68,210 53-3000 Motor Vehicle Operators 19,294 0.79 32,800 53-4000 Rail Transportation Workers 14,869 0.61 52,910 53-6000 Other Transportation Workers 3,410 0.14 21,600 53-7000 Material Moving Workers 4,677 0.19 24,230

43 44 2. The Emotional Toll of Long-Term Unemployment: Examining the Interaction Effects of Gender and Marital Status1

2.1. Introduction Among the most pernicious and enduring effects of the Great Recession is the rise of long-term unemployment. While the rate of long-term unemployment has declined from its Great Recession peak, as of 2015 the percent of the unemployed who are long-term unemployed remains at levels unseen in over six decades. A well-established literature associates long-term unemployment with a variety of social ills including poverty, increased risk of physical and mental health problems, deteriorating emotional wellbeing, high suicide and mortality rates, domestic violence, divorce, and academic underperformance of kids (Sullivan & Von Wachter, 2009; Van Horn, 2013). Given the historically high rate of long-term unemployment and its associated negative consequences it is important for researchers to develop nuanced understandings of how such effects vary and the extent to which these effects are mediated by other institutions. This paper draws on mixed methods to consider how the negative emotional toll that typically accompanies spells of long-term unemployment differs for men and women, and more specifically, whether and how marital status interacts with gender in shaping this emotional toll. Our qualitative data show that marriages are often helpful in buffering the emotional toll of long- term unemployment for both men and women, but for approximately half of our interviewees a range of marital tensions arose due to their unemployment. Looking more closely at such tensions reveals a strikingly gendered pattern in which marital tensions related to the provider role were only reported by married men and not by married women. Our analysis of survey data also shows that overall marriages are helpful to the wellbeing of both unemployed men and women. Yet, for married men but not for married women, the analysis reveals that the significance of the benefits derived from marriage disappears once we control for household income. Taken together these findings make an important contribution to the existing literature by deepening our understanding of how gender and marital status shape and mediate the

' Basbug, G. & Sharone, 0. (2017). The Emotional Toll of Long-term Unemployment: Examining the Interaction Effects of Gender and Marital Status. In A. Kalleberg & T. von Wachter (Eds), RSF Journal of the Social Sciences, 3(3). Russell Sage Foundation, 112 East 64th Street, New York, NY 10065. Reprinted with Permission.

45 emotional toll of long-term unemployment. Since this emotional toll is strongly associated with negative effects on physical and mental health, as well as job search discouragement, the findings in this paper have important implications for policymakers and practitioners working to address the fallout from the ongoing crisis of long-term unemployment.

2.2. The Great Recession and Long-Term Unemployment The Great Recession is the longest and most devastating economic downturn since the Great Depression (Elsby, Hobijin, & Sahin, 2010; Sum, Khatiwada, McLaughlin, & Palma, 2009). From May 2007 to October 2009, the labor force lost over 7.5 million jobs and the unemployment rate climbed from 4.8 percent to 10.1 percent (Grusky, Western, & Wimer, 2011; Katz, 2010). Compared to previous recessions, during and after the Great Recession, job loss was much higher, reemployment rates were lower, spells of unemployment were longer, the rate of underemployment was higher, and the increase in the number of discouraged workers was substantial (Farber, 2011). The level of economic dislocation and financial devastation brought on by the Great Recession was accompanied by an enormous toll on emotional wellbeing. Among the most salient indicators of this negative emotional toll are the effects on families with rising divorce rates (Morgan, Cumberworth, & Wimer, 2011), increased domestic violence (Schneider, Harknett, & McLanahan, 2014), reduced fertility rates (Schneider & Hastings, 2014), and diminished child development (McLanahan, Tach, & Schneider, 2013). One of the distinguishing characteristics of the Great Recession was the rise of long-term unemployment (defined by the U.S. Bureau of Labor Statistics as unemployment lasting 27 weeks or longer). In May 2015 the Bureau of Labor Statistics' data showed that 29 percent of the unemployed were long-term unemployed compared to 18 percent in 20072. In February 2017, this rate declined only to about 24 percent, even though the overall unemployment rate had declined to 4.7 percent. In March 2010, the number of long-term unemployed individuals reached over 6.5 million representing 44.1 percent of the unemployed population, and the long- term unemployment rate increased to 4.3 percent, much higher than the prior postwar era peak of 2.6 percent in 1983 (Katz, 2010).

2http://www.bls.gov/news.release/empsit.nrO.htlm

46 The picture is even more distressing if we additionally consider the estimated number of long-term unemployed workers who became discouraged and dropped out of the labor force since the Great Recession. In March 2015 the Economic Policy Institute estimated that the U.S. has 3 million "missing workers,"3 defined as potential workers who are neither working nor looking for work due to weak labor market conditions. Qualitative data suggest that a sizeable proportion of missing workers withdrew from the labor market after a prolonged and unsuccessful search (Sharone, 2013). The ongoing crisis of long-term unemployment can also be seen by examining patterns in unemployment duration. While the mean unemployment duration in past recessions was 20 weeks, mean duration increased to 35 weeks in the Great Recession (Farber, 2011). Long-term unemployment affects men and women across varied segments of American society. It is widespread across different education levels, gender and race groups as well as occupations and industries (Economic Policy Institute, 2014). Recent studies show that long- term unemployment is not limited to individuals who are inflexible and therefore closed to opportunities in other occupations or industries, or to individuals who lack advanced education (Sharone et. al., 2015). Although for college-educated individuals the rate of long-term unemployment is lower than for individuals without a college degree, the college-educated long- term unemployment rate is twice as high as before the Great Recession. Among the few predictors of long-term unemployment is age. More than half of all unemployed workers over the age of 55 are long-term unemployed across all levels of education (Evangelist & Christman, 2013). The severe and negative emotional toll of long-term unemployment is well-established, and has been shown to be as harmful to wellbeing as the loss of income (Winkelmann & Winkelmann, 1988; Sharone, 2013). A line of qualitative studies dating back to the Great Depression (e.g., Bakke, 1933) consistently show that prolonged unemployment in the U.S. frequently leads to diminished self-esteem and increased self-blame, which is particularly intense among white-collar workers (Chen, 2015; Newman, 1988; Sharone, 2013; Smith, 2001). For example, Newman's (1999) study of unemployed managers found that most managers perceived their unemployment as a personal failing. Sharone (2013) likewise describes the emotional toll

' http://www.epi.org/blog/another-month-same-story-job-openings-data-little-changed-in-february/

47 of long-term unemployed white-collar workers who come to feel that they are "flawed" or "defective." Meta-analytic reviews of quantitative studies also consistently show that unemployed individuals suffer from significantly lower levels of mental health compared to employed individuals (McKee-Ryan, Song, Wanberg, & Kinicki 2005). Longitudinal studies show that when individuals become unemployed their psychological health deteriorates while upon reemployment their well-being recovers. In addition, meta-analytic findings show that as unemployment duration increases so does the negative emotional toll and the risk for mental and physical health issues (Paul & Moser, 2009). For example, according to a Gallup poll, the depression rate among long-term unemployed workers is 18 percent compared to 12.4 and 5.6 percent for all unemployed and employed workers, respectively. 4 Underlying the negative emotional toll of unemployment is stress induced by the loss of income and savings, the exhaustion of unemployment benefits, as well as the loss of self-esteem and self-confidence due to the experience of repeated employer rejections. The longer individuals remain unemployed the more they experience deprivation (Brief, Konovsky, Goodwin, & Link, 1995), exhaustion of coping resources (Kinicki, Prussia, & McKee-Ryan, 2000), and even personality changes (Boyce, Wood, Daly, & Sedikides, 2015). Krueger and Mueller (2011) found that unemployed workers' unhappiness increases the longer they stay unemployed, and such unhappiness is particularly salient when they are engaged in job searching activities. The association between unemployment and negative physical health outcomes is also established by a long line of research and meta-analytic analyses (e.g., Brand, 2015; McKee- Ryan et al., 2005; Paul & Moser, 2009; Sullivan & von Wachter, 2009). A number of studies link the emotional toll of unemployment described above-and particularly the experience of self-blame and internalization of stigma among unemployed workers-with negative health outcomes (Eales, 1989; Creed & Dee Bartrum, 2007; Rantakeisu, Starrin, & Hagquist, 1999). Over the past decade a rapidly growing literature on the connection between stigma and health shows that the internalization of stigma is linked, among other mechanisms, to increased diastolic blood pressure (Harrell, Hall, & Taliaferro, 2003; Guyll, Matthews, & Bromberger 2001) and increased cortisol output (Dettenborn, Tietze, Bruckner, & Kirschbaum, 2010; Townsend, Major, Gangi, & Mendes, 2011). These physical responses to internalized stigma

4 http://www.gallup.com/poll/171044/depression-rates-higher-among-long-term-unemployed.aspx

48 are described in this literature as connected to experiences that threaten peoples' identity and status in social hierarchies, and specifically to experiences involving conditions of uncertainty about outcomes, and in which a successful outcome involves a process of "social evaluation" (Keating, 2009, p. 75; Major & O'Brien, 2005). The core experience of unemployment is precisely one of repeatedly subjecting oneself to social evaluation under conditions of great uncertainty, which upon repeated employer rejections frequently leads unemployed workers to perceive themselves as "flawed" (Newman, 1999; Sharone, 2013). A further mechanism linking unemployment and negative physical health is social isolation (Kawachi & Berkman, 2001; Keating 2009; Umberson, Corsnoe, & Reczek, 2010).

2.3. Gender, Marital Status and the Emotional Toll of Unemployment An important but understudied question is how the negative emotional toll of long-term unemployment varies by gender and marital status. The qualitative literature on this issue yields mixed results. Older studies dating back to the 1930s suggest that unemployment is particularly undermining of married men due to the effects of the traditional male breadwinner role. For example, Komarovsky's (1940, p. 74) classic study of long-term unemployment in the Great Depression, using in-depth interviews, finds that "man experiences a deep frustration because in his own eyes he fails to fulfill what is the central duty of his life, the very touchstone of his manhood -the role of family provider." According to Komarovsky this emotional toll was more acute for men whose exclusive identity was as providers than for men who also identified as fathers and husbands. A similar finding about the particular vulnerability of married men was reported by Newman's (1999) study of unemployed managers in the 1980s. Newman (1999, p. 139) explains: "Unemployment strikes at the heart of the masculine ideal... [having] failed at the task that most clearly defines his role, he suffers a loss of identity as a man." Like Komarovsky (1940), Newman (1999) also found that not all men were equally vulnerable to this emotional toll; specifically, gay men who were not married were much less likely to report the anxiety and anguish expressed by married men. A more recent qualitative study examining the emotional toll of long-term unemployment challenges the applicability of Komarovsky and Newman's studies in the current era and claims that with changing gender roles it is now married women, more than married men, who are vulnerable to self-blame and the emotional toll of identity loss. According to Lane (2011, p.

49 121) men in the twenty first century "have an alternative standard of masculinity" where relying on a partner's income is evidence of progressive manhood." According to Lane (2011, p. 117) the unemployed men she interviewed took it as a badge of honor that they were "secure enough in [their] manhood to comfortably rely on [their] wife's income." For women, on the other hand, their increasingly career-based identities mean that relying on their husbands' income is devastating to their self-worth-evoking uncomfortable feelings of being needy and dependent. Thus, in comparing the experiences of unemployed married men and women, Lane (2011, 126) concludes that "women's feelings of dwindling self-worth mirror those of male managers in [Newman's] study far more so than did the comments of their male peers." While Lane's claims are interesting, more research is needed to determine whether a historical reversal has indeed occurred. One limitation of Lane's (2011) study is that none of the men who "took pride" in letting their wife support them were part of a dual earner couple with children. This is significant because other contemporary studies show that the traditional male breadwinner role expectations surfaces most clearly when married couples have children (Gerson, 2011). Turning to survey studies that compare the unemployment experiences of men and women we also find mixed results. Some studies find that men experience more distress associated with unemployment than women, while others find no difference (Broman, Hamilton, Hoffman, & Mavaddat, 1995; Leana & Feldman, 1991). Meta-analytic findings suggest that men are more negatively affected by unemployment than women (Paul & Moser, 2009). Yet several other prior studies show that it takes more time for women to find a job after a period of unemployment, and that long-term unemployed women's re-employment opportunities come with a greater loss of prior salary than is the case for long-term unemployed men (Snyder & Nowak, 1984). A different line of studies explores how marital status mediates the negative emotional toll of unemployment. The dominant perspective in the psychology literature is that marriage provides critical social support and acts as a shield in difficult times such as unemployment. Prior studies that examine the effect of being married on unemployment typically focus on the social support aspect of the marriage relationship. For example, cross-sectional studies consistently find that married unemployed individuals are in a better psychological state than single unemployed individuals (Cooke & Rousseau, 1984; Leana & Feldman, 1991). These findings are consistent with sociological studies suggesting that marriages benefits both men and

50 women and enhances mental health (Simon, 2002). In their meta-analytic study, McKee and her colleagues (2005) found that married unemployed individuals are more satisfied with their lives than single unemployed ones. Similarly, using British panel data, Clark and Oswald (1994) found that married unemployed individuals experience less mental health distress than single unemployed individuals. Alongside the studies discussed above showing the supportive effects of marriages, the literature also reveals that some unemployed individuals experience negative social support from their spouses, including findings of social undermining which refers to negative evaluation of the person in terms of his or her attributes or actions (Vinokur & van Ryn, 1993). One of the reasons for such negative social support in marriages is the economic strain experienced in times of unemployment. Economic hardship changes the quality of the relationship and creates conflict in married couples (Conger, Rueter, & Elder, 1999). For example, Hoffman and Duncan (1995) found that a husband's low income significantly increases the risk of marital dissolution. Similarly, Jensen and Smith (1990) found that in a married couple if the husband is unemployed, divorce is more likely to occur but this is not the case when the wife is unemployed. However, more recent studies find that either a husband or wife's unemployment leads to increased likelihood of marital dissolution (Hansen, 2005; Jalovaara, 2003). Another complicating factor in the current era of stagnating wages is that increased economic interdependence between married men and women prevent both husbands and wives from contemplating separation (McManus & DiPrete, 2001). Although the studies described above analyze the effects of gender and marital status on the emotional toll of unemployment, prior quantitative studies have not considered how marital status interacts with gender in mediating this emotional toll. The interaction of gender and marital status is likely important given the well-documented significant effects of marriage on wellbeing, and given the persistence of gender differences in the experiences of men and women in contemporary society. In light of the existing literature on the effects of marriage we expect that for both long-term unemployed men and women marriage will offer some protective benefits and buffer against the emotional toll of unemployment. Marriage is a ready-made support system. In addition to emotional support, marriages also play a social and economic insurance role. However, based on prior qualitative studies such as Komarovsky (1940) and Newman (1999) discussed above, for long-term unemployed men, we hypothesize that marriage

51 is a double-edged sword. While on the one hand it provides varied forms of support as described above, on the other hand-to the extent that stereotypical gendered expectations persist which focus the male role in the family on breadwinning-long-term unemployment is likely to create tensions. Specifically, we expect that the inability of long-term unemployed men to fulfill the breadwinner role will generate marital distress that will offset the supportive benefits of marriages for men. Given that this tension is premised on the persistence of stereotypical male gender roles, we expect that this particular interaction of gender and marital status will affect men more than women. With respect to women, given the well-documented research on the labor market obstacles facing long-term unemployed women, we expect that the emotional toll of long-term unemployment will be as difficult, if not more difficult for women than for men, but that in the case of women marriages will follow the general pattern of having a salubrious effect.

2.4. Qualitative Data and Analysis To begin our exploration of how the emotional toll of long-term unemployment varies by gender, and whether and how it is mediated by marital status, we turn to our qualitative data. The qualitative data discussed in this paper are derived from 50 in-depth interviews conducted with long-term unemployed job seekers in 2013 and 2014. To recruit interview subjects we reached out to job seekers through Boston area career centers, networking groups, and libraries, and invited job seekers to sign up for the opportunity to participate in research in exchange for receiving free support to be provided by volunteer career coaches and counselors. Interested job seekers were asked to complete a short survey in order for us to determine whether they met the following criteria: (i) unemployed six months or longer, (ii) between the ages of 40-65, (iii) white-collar occupations, and (iv) looking for work in the Boston area. While over 800 job seekers signed up for the opportunity to participate only 125 met the criteria for participation. Our sample of 125 unemployed job seekers was 55 percent male and 45 percent female, with 70 percent of the sample being married and with a mean age of 54. Our recruiting focused on unemployed workers over the age of 40 because, as previously discussed, older workers are more likely than younger workers to get trapped in long-term unemployment. While we expect that due to self-selection our sample consists of unemployed job seekers with higher than average levels of motivation to continue with their job search and to receive job search support, we do not expect this particular selection bias to affect our findings regarding how gender and marital

52 status mediate the emotional toll of long-term unemployment. In-depth interviews with a randomly selected subset of 50 long-term unemployed job seekers were conducted prior to any of the job seekers receiving support. These semi-structured, in-depth interviews were conducted either in-person or by telephone and lasted approximately 60-120 minutes each. We asked participants questions regarding their experience looking for work and the effect of unemployment on their wellbeing and personal relationships. The in-depth interviews revealed the intense negative emotions that long-term unemployed job seekers experience. Consistent with prior research both male and female job seekers discussed the emotional toll of financial stress and diminishing self-esteem. For example, Deborah, a fifty-two year old marketing executive, discussed how job loss has entailed a "profound loss of self-esteem and confidence." She explained: "My job was a huge part of my identity. Huge part of how people saw me. It's hard not to feel like a throw-away." Steven, a fifty six year old computer engineer, succinctly described similar feelings: "I'm embarrassed. I'm humiliated. I feel like a loser." A vast literature, as previously described, finds that marriages are generally supportive of individuals' emotional wellbeing. Consistent with this literature, compared to the married unemployed workers we interviewed, those who were unmarried more frequently described painful feelings of intense isolation and the absence of support. For example, Abbey, a single woman, discussed how "I have no one supporting me. I'm on my own." While Abbey has friends and extended family, she explains that "it's not like I'm talking to them every day. I really feel the weight of this is on myself. It's really emotionally devastating to feel that sort of isolation." Gene, a widower, described unemployment as being the "darkest period in my life," and isolation as the most difficult aspect of this period since his social life had been fully centered around his former work. Being home alone "I do not talk to anyone on a daily basis." He describes the temporary reprieve from isolation when he goes to a networking support group: "These meetings are helpful sometimes for the camaraderie for a couple hours." But, then he adds, "I find myself as the meeting is closing ... I sit there and go, 'This isn't gonna last much longer. Maybe another twenty minutes. Then I'm going to get in my car and go home. It's going to be dark." Both unmarried men and women described how as a result of their unemployment their friendships have become more fragile, and thus friends are less available as a source of support.

53 Frank discussed how in his case his friends have withdrawn: "People that I thought I knew, some have just dropped off the radar and have literally told me not to contact them. It's like they are saying 'there is something wrong with you. You're not working and we don't want to talk to you."' Tammy's friends have not cut off contact but, she explains, they are "all married so a lot of times they're tied up with other obligations," and even when they do get together, Tammy feels like "people don't really want to know all the gory details about how depressing it is. They will say 'Oh, that's awful, its depressing, let's talk about something else." In addition to the discomforts and stigmas that lead friends to withdraw from unemployed workers, other factors lead unemployed workers to withdraw from friends. In part such withdrawal is due to financial constraints as being with friends usually involves spending money. Rob put it this way: "I don't have the money to go out and do what my friends are doing." But withdrawing from friends is also the result of social unease rooted in the loss of status. Jack explains: You feel uncomfortable about where you are in your life versus where your friends are. Some of these folks have jobs, ranging from vice presidents to directors, doctors. So you tend to maybe not socialize as much as you would like to or as you did in the past. Some single unemployed workers also discussed the fact that although they would like to find a romantic partner, dating was simply not feasible during unemployment. Denise, with a laugh, rhetorically asked: "I'm already facing rejection right and left on the job front, who needs more rejection?" But then she added, with a more serious note: What's the first question anybody asks you when they meet you: 'What do you do for a living'? Well, I don't do anything because I'm one of these people who had the misfortune to be laid off. I can't even see any gentleman taking me seriously. David likewise reported wishing to be in a relationship but then added, "a part of me feels too

unworthy to be . . . I feel 'less than.' By contrast to these unmarried job seeker's descriptions of isolation and lack of social support, approximately half of the married job seekers we interviewed described varied forms of positive supports from their spouses. For example, Ryan shared: "I would not have been able to make it through this period without my wife. My wife could not be more supportive." Ryan explained that in the moments when he felt most discouraged his wife would remain "really

54 positive and hopeful" and "she'd say 'oh this is going to be your week, I can see it." Jen described her supportive relationship with her husband of 30 years. Despite the fact that at the time of the interview Jen and her husband were about to lose their home to foreclosure, she found much solace in the marriage, explaining: We have been able to face the terror together and in so doing, we have built a bond that is amazingly durable. Nobody should have to go through what we've been through, but [the marriage] is a tremendous balm. While unmarried unemployed workers were more likely to describe painful isolation than married workers, marriage does not always alleviate the isolation felt by unemployed workers. In fact, about half our married interviewees also reported experiencing the emotional toll of unemployment in isolation. Steven, who is previously quoted as feeling "embarrassed" and "humiliated" explained that despite the fact that he is married, he feels "very much alone" reflecting on how "it's very hard to talk to my wife about it because she's never been through anything like it." Albert similarly shares: "My wife has no clue what it's like." The experience of emotional isolation within marriages is not limited to men. Linda, a fifty-eight year old woman who has worked for over twenty years in software sales, explains: "My husband doesn't understand what I'm going through." Because Linda's husband's salary is sufficient to financially support the both of them he has difficulties relating to Linda's negative emotions of identity loss and her growing sense of financial vulnerability. While our qualitative data suggest that both men and women may experience emotional isolation within marriages, when looking more closely at the narratives of married unemployed workers a strikingly gendered pattern emerges in our interview data with about a third of the men, but no women, describing marital tensions due to disagreements over job search intensity and what kind of job the unemployed spouse should seek. Tensions over job search intensity frequently took the form of men reporting that their wives did not think they were exerting sufficient effort to find work. For example, Larry is a software engineering manager specializing in speech recognition. Larry and his wife are under significant financial pressure because they have two school-aged children and a large mortgage. Larry describes the strain this has put on his marriage: [My wife] says 'You're not doing enough.' What the hell? I can only do so much. She doesn't understand it. She hasn't been through that herself.

55 There is a lot of pressure if you're in a family. A different but related kind of tension was discussed by Richard, a fifty year old public relations professional. Richard and his wife also have young children and feel intense financial stress due to Richard's unemployment. Richard explains: "I've never really been in a position where I wasn't sure whether I could pay the mortgage or buy my food or keep my car repaired." His family is dependent on his income since his wife "doesn't make much money. She's mostly volunteering." Unlike Larry's case, the source of the marital tension, as Richard describes it, is not that he is not doing enough on his job search but that he is doing too much job searching and not enough to contribute his share to the "second shift" (Hochschild, 1989). Richard put it this way: My wife feels neglected because I'm wrapped up in job search related things and not spending enough time just doing other things which are a part of daily life. She might think I'm not doing anything but looking for a job is more than 40 hours. You always have to be on. That's a stress. The marital tensions described by Richard and Larry are partly rooted in the difficulties spouses have of understanding the experience of unemployment. As previously discussed, the feeling that spouses "don't understand" is frequently reported by both men and women. The distinctly gendered pattern is the lack of understanding coupled with the suggestion that the job seeker is not doing enough on their search, or is not doing enough at home, which was only reported by male job seekers. Given the abundant research showing that married women do a disproportionate amount of house and care work in families (e.g. Gerson, 2011) it is perhaps not surprising that women with unemployed husbands may expect more support on the home front. At the same time, for men like Richard who struggle to do the "work" of job searching (Sharone 2013) under the stress of unemployment, this spousal expectation may signal that his wife "might think I'm not doing anything." Interestingly, some unemployed women we interviewed did report the kind of tension described by Richard but not in the context of their marriages. Instead these tensions arose with mothers, sisters, or close friends who were disappointed when not receiving support from unemployed women who they had presumed had time on their hands. For example, Nadine, an unemployed unmarried woman, discussed changes in her relationship with her mother and sister: They both expect me to pick up a lot more of the pieces. My mom will now

56 ask would you mind stopping at the post office and mailing this packages? The subtext is that since 'you're not doing anything why don't you help me?' They don't understand the pressures of looking for work. Another kind of marital tension that was generally only reported by men focuses on the kind of job the unemployed spouse should seek. A good example is Warren, a sixty-two year old environmental scientist, with a seven-year old child. Despite the fact that Warren's wife works fulltime his family is under considerable financial stress in the absence of his income. Warren reported that his wife has pressured him to seek any available job, whether or not in his field, including low paying retail position. Warren explained: It's a huge stress on our marriage. My relationship with my wife is really fraught with difficulty because I didn't realize how much she really wanted a breadwinner in the mix ... My wife has been pressuring me to get a job in retail sales, selling camping gear or something. In his long career as an environmental scientist Warren has taught at Ivy League universities and conducted research with multimillion dollar grants. While Warren has expanded the breadth of his search beyond teaching and research to include university administration, he has maintained a focus on positions that would allow him to continue making a contribution in his field of expertise. Yet, the unexpected toll on his marriage from his hesitation to consider jobs like retail sales weighs heavily on Warren as he confides: "I don't want to let my loved ones down. That's a huge thing." While in Warren's case marital tensions arose from his spouse feeling that his search was too narrow, other men report marital tensions arising from spouses thinking their husbands' searches are too broad. James, a fifty-three year old former corporate manager has expanded his search to include finance-related positions in non-profit companies. James believes that his corporate finance experience can be a valuable asset to a non-profit company, and that down the road such a position would provide more security and meaning than his former work even if this would involve an initial pay cut. Yet, James is experiencing tensions in his marriage because his wife does not support his search including non-profit companies. James explains: I feel I should take a lower paying job and she's like, 'Oh my god, that's so low.' She doesn't want me to take it. She's thinking we're going nowhere with me taking a lower paying job. I feel like I have a longer range view of

57 things than her... She's just looking for how much they are going to pay me right now. And I'm looking at how much I'm going to end up getting after 5-10 years. While the marital tensions described by Warren and James were not unusual among long-term unemployed men, they were strikingly absent from the accounts of marital tensions described by women. In both of these cases the marital tensions arising from unemployment are implicitly linked to the extent to which the unemployed male is perceived to be fulfilling the breadwinner role. For Tom, a fifty-five year-old unemployed sales agent, the link of breadwinning and the state of his marriage was made explicit. Tom noted that the strain in his marriage has been somewhat mitigated due to savings that he had accumulated when he had a job, which means "I've still been able to provide." But, after a pause, he added: "Absolutely, if I wasn't able to provide, I probably would not be married today." To summarize, as we review the qualitative data, we find that consistent with the existing literature on the salubrious effects of marriages, about half of our married unemployed workers benefited from forms of positive support that were less frequently available to unmarried unemployed workers. Yet, our qualitative data also reveal two kinds of marital tensions that exacerbate the emotional toll of long-term unemployment. First, both men and women report how marital tensions may arise due to spouses being unable to understand and relate to the difficult experiences of unemployment. Second, unemployed married men but not women describe marital tensions arising due to spouses feeling that the unemployed spouse is not doing enough on their search, not doing enough at home, or not looking for the right kind of job. It is important to note that the second kind of marital tension, which was only reported by men, was in almost every case accompanied by descriptions of the family's severe economic stress. This is perhaps not surprising given that tensions surrounding issues like the appropriate level of job search intensity and appropriate job targets are more likely to arise under conditions of economic duress. Although the qualitative data are derived from a relatively small sample, this striking pattern does support our hypothesis that while both men and women generally benefit from being married, in the context of unemployment, due to a lingering male breadwinner expectations for men the salubrious benefits of marriage are counterweighed by marital tensions that intensify the emotional toll of long-term unemployment. In the remainder of this paper we explore this hypothesis using a larger dataset and quantitative analysis.

58 2.5. Quantitative Data and Analysis To further examine the effects and interactions of gender and marital status in mediating the emotional toll of long-term unemployment, and specifically to explore the hypothesis generated by our qualitative data and the literature discussed above, we exploit a set of high frequency data collected right after the end of Great Recession from a sample of unemployed workers in New Jersey. The survey was conducted by the Princeton University Survey Research Center from October 2009 to April 2010. The sample consists of 6,025 unemployed individuals who were recipients of unemployment insurance in New Jersey. The unemployed individuals in this study were surveyed for 12 consecutive weeks, with a subset of respondents who were long- term unemployed at the time of study surveyed for an additional 12 weeks for a total of 24 weeks. In total 39,201 surveys were completed. Table 1 provides information on sample characteristics. The survey consisted of two parts: an initial survey which was administered in the first week and collected information on demographics and income, and a weekly survey which was administered in the first week and in each subsequent week and which gathered a wide range of information about respondents' ongoing job search activities, time use, reservation wages, job offers and emotional states (for a detailed description of the survey, see Krueger & Mueller, 2011). Although at the time of the survey New Jersey's overall unemployment rate was similar to the national unemployment rate, it is important to note that New Jersey's long-term unemployment rate, with 40 percent of its unemployed workers being long-term unemployed, was among the highest in the United States. The New Jersey dataset is well-suited for addressing our theoretical and empirical questions regarding the effects of gender and marital status in shaping the experience of unemployment because it includes information on the emotional state of unemployed respondents as measured by life satisfaction, as well as time spent in a negative mood on a weekly basis. Life satisfaction was measured with the question of "Taking all things together, how satisfied are you with your life as a whole these days"? Respondents were asked to pick a level of life satisfaction from a 4-point scale ranging from very satisfied to not at all satisfied. The survey also measured respondents' mood by asking: "Now we would like to know how you feel and what mood you are in when you are at home. When you are at home, what percentage of the time are you: 'in a bad mood', 'a little low or irritable mood', 'in a mildly pleasant mood', 'in

59 a very good mood"'? Respondents were asked to indicate the percentage of time that they experienced each mood category.

2.5.1. Descriptive Results We begin our analysis by examining descriptive statistics about the reported level of life satisfaction among the unemployed respondents. Table 2 breaks down these responses by gender and marital status. To provide some context for interpreting the responses of the unemployed individuals in the New Jersey survey we also present the responses of employed individuals to the same life satisfaction question from Princeton Affect and Time Use Survey (PATS). PATS is a national telephone survey that was conducted in the spring of 2006 by the Gallup Organization. While the whole PATS sample consists of nearly 4,000 respondents, for purposes of our comparison with the New Jersey data we limit our analysis to the responses of employed individuals in the PATS data. As seen in Table 2, the differences in the life satisfaction of employed and unemployed individuals are dramatic. Only 5.5 percent of unemployed single females are very satisfied with their lives compared to 37 percent for employed single females. Overall married females are more satisfied than single females regardless of employment status, but again the effect of unemployment is enormous with the percent of unemployed married women who are very satisfied with their lives at 8.7 percent compared to 55 percent for employed married females. Table 2 shows a similar pattern for men with immense differences in life satisfaction between employed and unemployed men, and as is the case with women, married men report higher levels of life satisfaction than single men. While the comparison between the PATS and New Jersey data is limited by the fact that PATS data were collected three years earlier and are national in scope (as opposed to only focusing on New Jersey), the dramatic differences in the answers to the identical questions about life satisfaction provide a useful baseline for contextualizing the emotional state of unemployed individuals in the New Jersey survey. For additional descriptive statistics on how the emotional toll of unemployment varies by gender and marital status we draw on the New Jersey survey data to compare the means of the responses given to the life satisfaction question and the percentage of time respondents reported spending in a bad mood for each group (single female, married female, single male, and married male). In Table 3 ANOVA statistics show that there is a significant difference between groups.

60 For both unemployed men and unemployed women those who are married report more life satisfaction and less time in a bad mood than single unemployed individuals. Our descriptive results are not surprising. We see that (1) consistent with the literature on the emotional toll of unemployment, employed individuals are indeed far more satisfied with their lives than unemployed individuals, and (2) consistent with the literature on the salubrious effects of marriage, married unemployed and employed individuals are more satisfied with their lives when compared to single employed and unemployed individuals.

2.5.2. Cross-sectional Analyses We next analyze whether the interaction of gender and marital status significantly predicts individuals' life satisfaction and mood. Using data from the initial survey of unemployed New Jersey workers, as seen in Table 4, we first regressed life satisfaction on gender, marital status and a host of control variables such as education, number of children, household income, race, age, and unemployment duration. In this model levels of life satisfaction are higher for unemployed women than for unemployed men, and higher for married than single individuals of both genders. When we added the interaction of gender and marital status we see that the interaction term is significantly and negatively related to life satisfaction. This result shows that single males experience the lowest levels of life satisfaction followed by single females. Married women reported higher levels of satisfaction than married men. The interaction term is also significantly and negatively related to time spent in a good mood. Single males again, as a group, spend the least amount of time in a good mood, followed by single females. Married women report spending the most time in a good mood, followed by married men. The cross-sectional analyses of the data support our descriptive findings.

2.5.3. Panel Data Analysis One of the distinguishing characteristics of the Great Recession, as previously discussed, is the crisis of long-term unemployment that came in its wake. For this reason we are particularly interested in exploring how the experience of unemployment changes over time as unemployment duration increases. Because the New Jersey survey of unemployed individuals provides data on the emotional state of respondents on a weekly basis, we are able to run individual fixed effect models to see how the emotional toll of unemployment unfolds over time

61 for married and single men and women. Using panel data also helps reduce the likelihood that our findings are driven by omitted variables, which have the potential to be in relationship with the independent variable. By using fixed effect models we can eliminate unobserved heterogeneity across individuals, which might introduce spurious results. In tables 5 and 6, we present fixed effect models for single females, married females, single males, and married males with the dependent variables of life satisfaction and time spent in bad mood, respectively. Table 5 shows that as unemployment duration increases life satisfaction increases for all groups except single males, but the effect sizes are extremely small and therefore not very meaningful. Much more striking are the findings with regard to time spent in bad mood. As seen in the Table 6, as the duration of unemployment increases, time spent in bad mood significantly increases for all groups. In short, our findings consistently show that for married and single men and women time spent in a bad mood dramatically increases with longer periods of unemployment, and because we use fixed effects models looking at the same individuals over time we can eliminate the possibility that these findings are driven by observed and unobserved differences among individuals. Looking more closely at these findings to explore whether the extent of the increase in time spent in bad mood significantly differs across groups, we use our panel data to run random effects models which include the interaction of unemployment duration with married and single men and women. As seen in Table 7, the interaction of unemployment duration with gender and marital status reveals a significant difference between single men and married men. The increase in time spent in bad mood as unemployment continues is significantly higher for single men than married men. However, it is important to note that although the increase in time spent in bad mood as unemployment continues is likewise higher for single females than for married females, this difference is not significant. Taken as a whole, the findings described above suggest that marriage mediates and diminishes the negative emotional toll of unemployment for both men and women. In the final step of our analysis we attempt to dig deeper to understand what is it is about marriage that is

5While this finding is unexpected given that other indicators of negative wellbeing intensify with increased unemployment, it is consistent with Krueger and Mueller (2011) which analyzed the same data and found that self-reported mood worsens as unemployment duration increases but life satisfaction does not. This finding is also consistent with other prior studies finding some evidence of habituation to unemployment as individuals remain unemployed (Clark, 2006; Winkelmann & Winkelmann, 1998).

62 helpful. Our analysis here is guided by our qualitative findings which suggest that marriages may be less supportive when married couples feel acute financial stress. Moreover, the pattern in the qualitative findings suggest that marital tensions as a result of financial stress are more likely to arise in couples with an unemployed man as compared to an unemployed woman. To explore whether the salubrious effects of marriage vary by gender under conditions of financial stress, we separated our analysis for married and single men and women. For this analysis, following Kahneman, Krueger, Schkade, Schwarz, and Stone (2006), we created a new dependent variable called "negative mood" by combining time spent in bad mood and time spent in a low/irritable mood. We first examine the relationship among marital status, economic strain, and negative mood for men. As can be seen in Table 8 Model 1, when we regress marital status on negative mood, and control for unemployment duration, we find that, as expected, single men spend significantly more time in negative moods than married men. However, in Model 2, when we add household income to the regression we find that after controlling for household income marital status became insignificant. In other words, once we control for the extra household income that typically comes from being married, there is no significant difference between the time spent in a negative mood for single men and married men. To see the effects of other variables in Model 3 we included all other control variables, except household income, and still found the significant difference between single men and married men with regard to time in a negative mood. In Model 4, when we added household income along with the other control variables, the marital status variable once again became insignificant. One interpretation of this striking finding is that for unemployed men the benefits of marriage derive more from added income than from other forms of intangible or emotional supports. We next ran the same models for women and, as can be seen in Table 9, the results are strikingly different. In Model 1, when we regress marital status on negative mood, and control for unemployment duration, we find that single women, just like single men, spend significantly more time in bad moods when compared to their married counterparts. The difference between men and women arise when we control for household income in Model 2. Unlike men, women benefited from being married even after controlling for household income, as well as after controlling for all other variables (Models 3 and 4). The differences in the effects of marriage for men and women suggested by Tables 8 and 9 are broadly consistent with the pattern observed

63 in our qualitative data, which suggested more marital tensions arising for couples under financial stress with an unemployed male. Finally, since our findings suggest that a lingering male breadwinner role increases the emotional toll of unemployment for married men, and since according to some existing studies the salience of the male breadwinner role varies by education levels-and specifically has diminished for married college-educated men (e.g. Lane, 2011)-to explore variations by levels of education we looked at whether men and women's education level moderates the relationship between unemployment duration and negative mood. Our regression analyses using panel data in Tables 8 and 9 show that overall unemployed college-educated workers (which for purposes of Table 8 and 9 we define as workers with at least some college education) spend significantly less time in a negative mood than unemployed workers who are not college-educated. However, differences emerge between men and women when we control for income and other demographic variables. Among men, after controlling for income and demographic variables, college-educated unemployed workers spent significantly less time in a negative mood than unemployed male workers who did not go to college. By contrast, among women, after controlling for income and demographic variables, college-educated unemployed workers do not significantly differ in time spent in a negative mood from unemployed female workers who did not attend college. We further examined whether education level moderates the effect of unemployment duration on negative mood for men and women separately. For this analysis, we regressed negative mood on the interaction terms of unemployment duration with education level dummies. As seen in Table 10, there is a significant difference for men between the reference category of some high school and the higher education categories. As the education level increases, this difference gets bigger. However, we do not see a similar pattern for women. Figure 1 graphically shows the difference by gender in whether time spent in a negative mood changes with level of education. The graph shows that for men, but not for women, time spent in negative mood decreases as education level increases. These findings lend some support to claims in the literature that the salience of the male breadwinner role varies by education levels, and specifically, that the male breadwinner expectations, and the emotional toll that accompanies such expectations during times of unemployment, are less salient for married college-educated men than for men without a college education.

64 2.6. Discussion and Conclusion Although the Great Recession officially ended in June 2009, its disastrous effects are still with us. Strikingly, the rate of long-term unemployment remains at levels unseen in the post-war era, wreaking havoc on the finances and wellbeing of millions of American families. Voluminous research has shown that long-term unemployment is associated with a variety of social ills including, among others, job search discouragement leading workers to drop out of the workforce, and deteriorating mental and physical health. As discussed in the introduction to this paper prior research has linked both job search discouragement (Sharone, 2013) and the deterioration of health (e.g., Creed & Dee Bartrum, 2007; Rantakeisu et al., 1999) to the severe and negative emotional toll of long-term unemployment. Drawing on both qualitative interview data as well as survey data this paper breaks new ground by examining how the negative emotional toll of long-term unemployment is shaped by the interaction of gender and marital status. A well-established literature shows the salubrious effects of marriage for wellbeing. Yet, given the ongoing debate in the qualitative literature about the degree to which marriage may in some cases exacerbate the emotional toll of unemployment for married men (e.g., Newman, 1998) or for married women (e.g., Lane, 2011), it is surprising that to our knowledge no prior survey study has focused on the interaction of gender and marital status in mediating the emotional toll of long-term unemployment. Our in-depth interviews with long-term unemployed job seekers in the Boston area reveal a suggestive pattern. As can be expected, given the existing literature on the positive effects of marriage, many interviewees discuss the benefits of marriage. Yet, approximately half the interviewees also discussed marital tensions that exacerbated the emotional toll of long-term unemployment. Although both unemployed men and women reported some tensions in their marriages arising from the fact that their spouses had difficulties understanding their prolonged unemployment, it was only men who described a second kind of tension involving spouses who suggested the unemployed male was not doing enough on their search, not doing enough at home, or not looking for the right kind of job. This pattern in the qualitative data, along with prior studies, led us to hypothesize that for unemployed men the generally supportive effects of being married would be counterweighed by marital tensions that tend to intensify the emotional toll of long-term unemployment.

65 Turning to our analysis of survey data of unemployed job seekers in New Jersey we confirm the immense emotional toll of unemployment showing large differences in life satisfaction between employed and unemployed individuals, as well as how negative moods increases over time with prolonged unemployment. How does the interaction of gender and marital status shape this emotional toll? Overall our data suggest that marriages are helpful to the wellbeing of both unemployed men and women. Yet, an interesting finding emerges when we attempt to understand what it is about being married that is particularly helpful to wellbeing. Marriages may be helpful in various ways, including as a source of emotional comfort and support and/or as a form of economic insurance with spousal contributions to household income. Strikingly, when looking at the emotional toll of unemployment for men (as measured by time spent in a negative mood) our analysis revealed there were no significant benefits to being married once we controlled for household income. While further research is needed to examine this finding more closely, the survey data we analyzed suggest that the main emotional benefit of marriage for unemployed men derives from increased household income. For unemployed women, by contrast, the benefits of being married for their emotional wellbeing remained significant even after controlling for household income. These findings are broadly consistent with the pattern in our qualitative data showing heightened marital tensions among married couples with an unemployed male. This study has important implications for policymakers and practitioners working to address the negative consequences of the ongoing crisis of long-term unemployment. In thinking about responses to the negative emotional toll generated by unemployment-which to an important extent underlies issues such as job search discouragement and deteriorating mental and physical health-traditionally available sources of emotional support during times of crisis, most notably spouses, may not be well positioned to provide such support in the specific context of financial stress and long-term unemployment. This points to the importance of policies and practices that bolster public support institutions for long-term unemployed job seekers and to the need to reverse the trend of defunding public sources of such support (McKenna, McHugh, & Wentworth, 2012).

66 2.7. References Bakke, E. W. (1933). The unemployed man: A social study. New York: E. P. Dutton & Co. Bales. Boyce, C. J., Wood, A. M., Daly, M., & Sedikides, C. (2015). Personality change following unemployment. Journal ofApplied Psychology, 100(4), 991. Brand, J. E. (2015). The far-reaching impact of job loss and unemployment. Annual Review of Sociology, 41, 359-375. Brief, A. P., Konovsky, M. A., Goodwin, R., & Link, K. (1995). Inferring the meaning of work from the effects of unemployment. JournalofApplied Social Psychology, 25(8), 693- 711. Broman, C. L., Hamilton, V. L., Hoffman, W. S., & Mavaddat, R. (1995). Race, gender, and the response to stress: Autoworkers' vulnerability to long-term unemployment. American Journal of Community Psychology, 23(6), 813-842. Bureau of Labor Statistics. "Unemployment Data". Accessed May 9, 2015. http://www.bls.gov/news.release/empsit.nrO.htlm Census Bureau. "Families Data". Accessed May 9, 2015. http://www.census.gov/hhes/families/data/cps2014FG.html Chen, V. T. (2015). Cut loose: Jobless and hopeless in an unfair economy. Berkeley: University of California Press. Clark, A. E., & Oswald, A. J. (1994). Unhappiness and unemployment. The Economic Journal, 104(424), 648-659. Clark, A. E. (2003). Unemployment as a social norm: Psychological evidence from panel data. Journal of labor economics, 21(2), 323-351. Clark, A. (2006). A note on unhappiness and unemployment duration. Applied Economics Quarterly, 52(4), 291-308. Conger, R. D., Rueter, M. A., & Elder Jr, G. H. (1999). Couple resilience to economic pressure. Journalof Personalityand Social Psychology, 76(1), 54. Cooke, R. A., & Rousseau, D. M. (1984). Stress and strain from family roles and work-role expectations. Journalof Applied Psychology, 69(2), 252. Creed, P., & Bartrum, D. (2007). Explanations for deteriorating wellbeing in unemployed people: Specific unemployment theories and beyond. In T. Kieselbach, A. Winelield, C. Boyd & S. Anderson (Eds), Unemployment and Health: Internationaland InterdisciplinaryPerspectives, (1-20). Bowen Hills: Australian Academic Press. Dettenborn, L., Tietze, A., Bruckner, F., & Kirschbaum, C. (2010). Higher cortisol content in hair among long-term unemployed individuals compared to controls. Psychoneuroendocrinology,35(9), 1404-1409. Eales, M. J. (1989). Shame among unemployed men. Social Science & Medicine, 28(8), 783- 789. Economic Policy Instittue. "Missing workers". Accessed May 9, 2015. http://www.epi.org/blog/another-month-same-story-job-openings-data-little-changed-in- february/ Elsby, M. W., Hobijn, B., & Sahin, A. (2010). The labor market in the Great Recession. Brookings PaperEconomic Activity, 1-48. Evangelist, M., & Christman, A. (2013). Scarring effects: Demographics of the long-term unemployed and the danger of ignoring the jobs deficit. National Employment Law ProjectBriefing Paper.

67 Farber, H. S. (2011). Job loss in the Great Recession: Historical perspective from the displaced workers survey, 1984-2010. NationalBureau of Economic Research Working Paper, 17040. Gallup. "Depression Rates". Accessed May 9, 2015. http://www.gallup.com/poll/171044/depression-rates-higher-among-long-term- unemployed.aspx Gerson, K. (2009). The unfinished revolution: Coming of age in a new era of gender, work, and family. New York: Oxford University Press. Grusky, D. B., Western, B., & Wimer, C. (2011). The Consequences of the Great Recession. In D. Grusky, B. Western, & C. Wimer (Eds.) The GreatRecession, 3-20. New York, NY: Russell Sage Foundation Press. Guyll, M., Matthews, K. A., & Bromberger, J. T. (2001). Discrimination and unfair treatment: Relationship to cardiovascular reactivity among African American and European American women. Health Psychology, 20(5), 315. Hansen, H. T. (2005). Unemployment and marital dissolution: A panel data study of Norway. European Sociological Review, 21(2), 135-148. Harrell, J. P., Hall, S., & Taliaferro, J. (2003). Physiological responses to racism and discrimination: An assessment of the evidence. American Journalof Public Health, 93(2), 243-248. Hochschild, A. (1989). The Second Shift: Working Parentsand the Revolution at Home. New York: Viking Penguin. Hoffman, S. D., & Duncan, G. J. (1995). The effect of incomes, wages, and AFDC benefits on marital disruption. Journal of Human Resources, 30(1), 19-41. Jalovaara, M. (2003). The joint effects of marriage partners' socioeconomic positions on the risk of divorce. Demography, 40(1), 67-81. Jensen, P., & Smith, N. (1990). Unemployment and marital dissolution. Journal of Population Economics, 3(3), 215-229. Kahneman, D., Krueger, A. B., Schkade, D., Schwarz, N., & Stone, A. A. (2006). Would you be happier if you were richer? A focusing illusion. Science, 312(5782), 1908-1910. Katz, L. (2010). "Long-term Unemployment in the Great Recession." Testimony for the Joint Economic Committee U.S. Congress Hearing on "Long-Term Unemployment: Causes, Consequences and Solutions" Kawachi, I., & Berkman, L. F. (2001). Social ties and mental health. Journalof Urban health, 78(3), 458-467. Keating, D. (2009). Social interaction in human development: Pathway to health and capabilities. In P. Hall & M. Lamont (Eds). Successfld Societies: How Institutions and CultureAffect Health, 53-81. New York: Cambridge University Press. Kinicki, A. J., Prussia, G. E., & McKee-Ryan, F. M. (2000). A panel study of coping with involuntary job loss. Academy of Management Journal,43(1), 90-100. Komarovsky, M. (1940). The Unemployed Man and His Family. New York, NY: Octagon Books. Krueger, A. B., & Mueller, A. (2011). Job search, emotional well-being, and job finding in a period of mass unemployment: Evidence from high-frequency longitudinal data. Brookings Papers on Economic Activity, 2011(1), 1-57. Lane, C. M. (2011). A Company of One: Insecurity, Independence, and the New World of White-Collar Unemployment. Ithaca, NY: Cornell University Press

68 Leana, C. R., & Feldman, D. C. (1991). Gender differences in responses to unemployment. Journalof Vocational Behavior, 38(1), 65-77. Major, B., & Obrien, L. T. (2005). The social psychology of stigma. Annual Review of Psychology, 56, 393-421. McKee-Ryan, F., Song, Z., Wanberg, C. R., & Kinicki, A. J. (2005). Psychological and physical well-being during unemployment: A meta-analytic study. Journal ofApplied Psychology, 90(1), 53. McKenna, C., McHugh, R., & Wentworth, G. (2012). Getting real: Time to re-invest in the public employment service. National Employment Law ProjectBriefing Paper. McLanahan, S., Tach, L., & Schneider, D. (2013). The causal effects of father absence. Annual Review of Sociology, 39, 399-427. McManus, P. A., & DiPrete, T. A. (2001). Losers and winners: The financial consequences of separation and divorce for men. American SociologicalReview, 66(2), 246-268. Morgan, S. P., Cumberworth, E., & Wimer, C. (2011). The great recessions influence on fertility, marriage, divorce, and cohabitation. In D. Grusky, B. Western, & C. Wimer (Eds.). The GreatRecession, 220-245. New York, NY: Russell Sage Foundation Press. Newman, K. S. (1998). FallingFrom Grace: DownwardMobility In The Age Of Affluence. New York, NY: Free Press. Paul, K. I., & Moser, K. (2009). Unemployment impairs mental health: Meta-analyses. Journal of Vocational Behavior, 74(3), 264-282. Rantakeisu, U., Starrin, B., & Hagquist, C. (1999). Financial hardship and shame: A tentative model to understand the social and health effects of unemployment. British Journalof Social Work, 29(6), 877-901. Schneider, D., & Hastings, 0. P. (2014). The effect of economic shocks on poor women's marriage and non-marital fertility: Evidence from the great recession. Paper presented at the Annual Meeting of the American Sociological Association. San Francisco. Schneider, D., Harknett, K., & McLanahan, S. (2014). Intimate partner violence in the great recession. Paper presented at the Annual Meeting of the American Sociological Association. San Francisco. Simon, R. W. (2002). Revisiting the relationships among gender, marital status, and mental health. American Journal of Sociology, 107(4), 1065-1096. Sharone, 0. (2013). Flawed System/Flawed Self: Job Searching and Unemployment Experiences. Chicago, IL: University of Chicago Press. Sharone, 0., Ghayad, R., Basbug, G., Vasquez, A., & Rosin, M. (2015). Supporting experienced LTU professionals. In C. Van Horn, T. Greene, & T. Edwards (Eds.). Transforming U.S. Workforce Development Policiesfor the 21st Century. Upjohn Institute Press. Smith, V. (2001). Crossing the GreatDivide: Worker Risk and Opportunity in the New Economy. Ithaca, NY: Cornell University Press. Snyder, K. A., & Nowak, T. C. (1984). Job loss and demoralization: Do women fare better than men? InternationalJournal of Mental Health, 13(1-2), 92-106. Sullivan, D., & Von Wachter, T. (2009). Job displacement and mortality: An analysis using administrative data. The QuarterlyJournal of Economics, 124(3), 1265-1306. Sum, A., Khatiwada, I., McLaughlin, J., & Palma, S. (2009). The economic recession of 2007- 2009: A comparativeperspective on its duration and the severity of its labor market impacts. Northeastern University: Center for Labor Market Studies. Townsend, S. S., Major, B., Gangi, C. E., & Mendes, W. B. (2011). From "in the air" to "under

69 the skin": Cortisol responses to social identity threat. Personalityand Social Psychology Bulletin, 37(2), 151-164. Umberson, D., Crosnoe, R., & Reczek, C. (2010). Social relationships and health behavior across the life course. Annual Review of Sociology, 36, 139-157. Van Horn, E. C. (2013). Working scared (or not at all): The lost decade, great recession, and restoring the shatteredamerican dream. U.K.: Rowan and Littlefield. Vinokur, A. D., & Van Ryn, M. (1993). Social support and undermining in close relationships: Their independent effects on the mental health of unemployed persons. Journal of Personality and Social Psychology 65(2): 350-359. Winkelmann, L., & Winkelmann, R. (1998). Why are the unemployed so unhappy? Evidence from panel data." Economica 65(257): 1-15.

70 2.8. Tables Table 1. Sample Characteristics of NJ sample Demographics Percent of total Female 52.1 Marital Status Married 48.9 Single 28.0 Separated 3.3 Divorced 12.1 Widowed 1.9 Domestic partnership 5.7 Age in years 24 or less 6.8 25-34 21.3 35-44 21.1 45-54 26.8 55 or over 24.0 Race White 68.0 Black 15.3 Other 5.4 Ethnicity Hispanic 9.1 Non-Hispanic 80.8 Education Less than high school 7.0 High school 26.0 Some college 26.4 College 40.7 Unemployment duration (weeks of UL paid) 0-9 11.2 10-19 11.2 20-29 11.9 30-39 10.8 40-49 11.3 50-59 10.1 60 or more 33.6 N 6,025

71 Table 2. Life Satisfaction of Unemploy ed and Employed Individuals Single Female Married Femal e Single Male Married Male

very satisfied 5.5% (37%) 8.7% (55%) 5.9% (34%) 9.0% (48%)

satisfied 35.8% (52%) 49.9% (38%) 31.0% (54%) 40.6% (43%)

not satisfied 46.5% (9%) 35.8% (6%) 51.7% (10%) 41.8% (8%)

not at all satisfied 12.2% (2%) 5.7% (1%) 11.4% (3%) 8.7% (1%)

Note. Percentages in parentheses are nationwide employed individuals' responses from Princeton Affect and Time Use Survey. Sample size for the PATS is 2,048.

72 Table 3. Means and ANOVA Results for Life Satisfaction and Percent of Time Spent in Bad Mood Life Satisfaction Percentage of time spent in a bad mood at home (1=not at all satisfied, 4=very satisfied)

Single Female 2.34 12.98 (0.76) (16.33) Married Female 2.61 10.69 (0.72) (14.93) Single Male 2.31 14.49 (0.74) (17.78) Married Male 2.49 12.36 (0.77) (16.57)

F score 49.013** 12.151**

Total 6011 5939 **p<. 0 0 1 Note. Standard deviations are in parentheses.

73 Table 4. OLS Results with Entry Survey Data

Life Satisfaction Time in Bad Mood Time in Good Mood Education 0.018* 0.020* -0.358* -0.374* -0.289 -0.241 (0.008) (0.008) (0.179) (0.180) (0.293) (0.294) Number of Children 0.0141 0.0151 0.197 0.181 -0.000 0.045 (0.008) (0.008) (0.173) (0.173) (0.282) (0.283) Household Income 0.022** 0.022** -0.393** -0.386** 0.289* 0.268* (0.003) (0.003) (0.076) (0.076) (0.125) (0.125) Unemployment Duration -0.001** -0.001** 0.023* 0.023* -0.048* -0.049** (0.000) (0.000) (0.008) (0.008) (0.140) (0.014) Female 0.094** 0.133** -2.125** -2.669** 2.063* 3.635** (0.020) (0.027) (0.432) (0.581) (0.706) (0.949) Black -0.061* -0.057* 0.161 0.109 5.597* * 5.749** (0.029) (0.029) (0.634) (0.635) (1.035) (1.037) Asian -0.0861 -0.086 -3.12* 3.124* -1.898 -1.910 (0.047) (0.047) (1.018) (1.018) (1.663) (1.662) American Indian -0.046 -0.054 1.612 1.728 5.102 4.767 (0.181) (0.181) (3.88) (3.883) (6.340) (6.339) Pacific islander -0.191 -0.183 -3.823 -3.924 7.762 8.053 (0.179) (0.179) (3.96) (3.964) (6.474) (6.472) Hispanic 0.0611 0.0621 0.443 0.420 4.969** 5.036** (0.035) (0.025) (0.775) (0.775) (1.266) (1.266) Age groups 0.003 0.004 -1.012** -1.03** -0.086 -0.032 (0.004) (0.004) (0.096) (0.097) (0.158) (0.159) Single -0.124** -0.077* -0.279 -0.952 -1.319 0.624 (0.023) (0.032) (0.510) (0.701) (0.833) (1.144)

Female X Single -0.086* 1.233 -3.564* (0.040) (0.881) (1.438)

Constant 2.266** 2.237** 22.694** 23.101** 25.822** 24.645** (0.051) (0.053) (1.107) (1.144) (1.808) (1.869)

R2 0.0411 0.0419 0.0432 0.0436 0.0138 0.0148 *p<.05, **<.01 Note. Standard errors are in parentheses.

74 Table 5. Fixed Effect Models for DV: Life Satisfaction Single Female Married Female Single Male Married Male Unemployment Duration .0086*** .0015* .002 .0014*

Constant 1.945*** 2.622*** 2.292*** 2.541***

N of obs 5159 9891 3708 11398 N of groups 921 1363 757 1573

R2 0.0126 .0005 0.0007 .0004 *p<.05, ***p<.001

75 Table 6. Fixed Effect Models for DV: Time Spent in Bad Mood Single Female Married Female Single Male Married Male Unemployment Duration .222*** .185*** .322*** .141***

Constant 4.090*** 1.895 .1534 5.398***

N ofobs 5131 9846 3695 11366 N of groups 914 1353 752 1570

R2 0.0099 .0101 0.0225 .0053 ***p<.001

76 Table 7. Random Effects GLS Regression Results DV: Time Spent in Bad Mood Unemp. Duration 0.06**

GenderMarital Status Single Female 1.132 Married Female -2.56* Single Male 0.85

GenderMarital Status*Unemp. Duration Single Female 0.03 Married Female 0.02 Single Male 0.07**

Constant 10.59**

N of obs 29806 N of groups 4554

R2 within 0.0100 R2 between 0.0196 *p<.05, **p<.Ol. Note. Married men is the omitted category.

77 Table 8. Random Effects with Longitudinal Data for Time Spent Negative Mood (Only Men) Model 1 Model 2 Model 3 Model 4 Unemployment Duration .081*** .079*** .128*** .115*** Single 3.94** 1.104 4.318* 2.646

Household Income -.780*** -.670*

Black -1.722 -2.825 Asian 6.513* 6.311* American Indian -9.455 -9.300 Pacific Islander -15.202 -14.589 Hispanic -2.201 -3.450 Age (20-24) 11.098* 10.365* Age (25-29) 12.108** 11.217* Age (30-34) 12.184* 11.166** Age (35-39) 13.835*** 13.367*** Age (40-44) 11.895* 11.117** Age (45-49) 13.761*** 13.643*** Age (50-54) 12.775*** 12.283* ** Age (55-59) 10.992** 10.891** Age (60-64) 3.464 3.337 College -5.821** -4.869** Number of children -.070 -.0616 Monthly Rent or Mortgage Pay .001* .001** Extended Study -.701 -.618

Constant 31.898 39.119 21.955*** 27.866***

0.0069 0.0158 0.0412 0.0471

N of obs 14945 14708 12244 12041 N of groups 2302 2270 1772 1751 *p<.05, **p<.ol, ***p<.000 Note. White and age (over 65) are omitted categories.

78 Table 9. Random Effects with Longitudinal Data for Time Spent Negative Mood (Only Women) Model 1 Model 2 Model 3 Model 4 Unemployment Duration .0784*** .0724*** 0.0604** 0.045* Single 4.728*** 2.901* 6.407*** 4.062*

Household Income -.494** -0.763***

Black -6.897*** -7.960*** Asian 4.408 4.617 American Indian 16.828 14.829 Pacific Islander 16.629 17.945 Hispanic 2.337 1.367 Age (20-24) 17.683** 16.747** Age (25-29) 18.669** 17.479** Age (30-34) 15.970** 15.310** Age (35-39) 18.409** 18.479** Age (40-44) 17.465** 17.692** Age (45-49) 20.573*** 20.666*** Age (50-54) 17.598** 17.653** Age (55-59) 18.713** 17.892** Age (60-64) 11.388* 11.823* College -3.341* -1.979 Number of children -.283 -.368 Monthly Rent or Mortgage Pay .0005 0.000 Extended Study .213 .519

Constant 30.152 34.711*** 16.889** 23.411***

0.0150 0.0192 0.0448 0.0502

N of obs 14861 11310 12470 12000 N of groups 2252 2 181 1842 1785 *p<.05, **p<.ol, ***p<.000 Note. White and age (over 65) are omitted categories.

79 Tale 10. Random effects Regression Results for DV: Negative Mood Only Men Only Women Unemployment Duration .35** .14

Education level High school diploma or equivalent 5.08 5.98 Some college -.12 2.89 College diploma -2.82 4.29 Some graduate school 9.28 5.61 Graduate degree 4.11 3.82

Education level* Unemployment Duration High school diploma or equivalent -.29* -.04 Some college -.22P -.04 College diploma -.210 -.08 Some graduate school -.44** -.09 Graduate degree -.39** -.07

Constant 31.77*** 27.99***

N of observations 14945 14914 N of groups 2302 2260

R2 within 0.0061 0.0015 R2 between 0.0118 0.0090 Note. Some high school is the reference category. Pp<.10, *p<.05, **p<.01, ***p<.000

80 2.9. Figures

Figure 1 Percent of Time Spent in Negative Mood 45.0000

40.0000 - ...... 35.0000 --

30.0000

25.0000

20.0000

15.0000

10.0000

5.0000

0.0000...... Some ghsh meig College Some graduate diploma or Some college dilmscolege Graduate school or less diploma school eqIui val ent degree Men37.9163 36.4392 35.1594 39.2609 33.3311 Women 34.7077 37.0642 34.5897 34.1423 35.2906 32.8995

81 82 3. Activation Programs for Unemployed Insurance Recipients: Comparison of Two Training Delivery Formats

3.1. Introduction Unemployment leads to a variety of problems for individuals such as loss of income and impairment of physical and mental health (Paul & Moser, 2009; Van Horn, 2013). Negative effects of unemployment and the ways to promote unemployed individuals' transition back to work have been hot topics for economists (Clark & Oswald, 1994; Krueger & Mueller, 2011), psychologists (Jahoda, 1981; Wanberg, 2012; Warr, 1987), and sociologists (Newman, 1999; Sharone, 2013; Smith, 2011). Previous research examined the effectiveness of various policies aimed to facilitate reemployment such as subsidized employment, job search assistance, and on-the-job training (Kluve, 2010). Meta-analytic findings showed that job search assistance programs produce more favorable impacts than other active labor market programs (Card, Kluve, & Weber, 2010). Over the past decades, a variety of job search assistance programs have been utilized successfully (e.g., Azrin, Flores, & Kaplan, 1975; Caplan, Vinokur, Price, & van Ryn, 1989). Two main goals of these intervention programs were promoting job seekers' transition from unemployment to employment and reducing negative psychological effects of being unemployed by providing job seekers with necessary skills and by boosting their motivation. Liu, Huang, and Wang (2014) recently conducted a meta-analysis and provided evidence for the effectiveness of job search interventions. The authors found the odds of finding employment were 2.67 times higher for job seekers who participated in interventions compared to those in control groups. Although research on the effectiveness of job search intervention programs is large, no study yet investigated the effect of different training formats (i.e., flexible vs. structured) on job search outcomes of Unemployment Insurance (UI) recipients. This is an important research inquiry for two reasons. First, it is crucial to find out the most effective intervention designs that help unemployed individuals get maximum benefit. Second, because resources dedicated to public employment services are scarce, it is important to figure out the optimal use of them. In this paper, we report results from a randomized field experiment where we compare the effectiveness of a mandatory training program including elective workshops with another mandatory program without elective options.

83 The organization of the paper is as follows. First, we will provide a brief review of job search interventions to point out what we know about what works. Then, we will discuss recent studies on labor market programs designed specifically for UI recipients. Next, in light of the literature on organizational training, we will argue why training format may matter. Then, we will introduce the setting and design of the current study. Finally, we will present the results and discuss their implications.

3.2. Job search Interventions Job search intervention is a training and support program designed to help job seekers obtain employment and deal with negative psychological effects of unemployment by providing job search skills and stress management techniques (Caplan, Vinokur, Price, & van Ryn, 1989; Liu et al., 2014). Since 1970s, job search intervention programs have been developed and tested experimentally by assigning job seekers into treatment and control groups which receive and do not receive support, respectively. Among many others, the Job Club project (Azrin et al., 1975; Azrin & Philip, 1979; Rife & Belcher, 1994) and the JOBS intervention program (Caplan et al., 1989; Vinokur, Price, & Schul, 1995; Vuori, Silvonen, Vinokur, & Price, 2002) are well-known examples of job search interventions which attracted attention from both popular press and scientific community. The JOBS intervention program, developed by Michigan Prevention Research Center, is a preventive intervention aimed at providing job search skills and inoculation against setbacks which might occur during unemployment. The Job Club was developed by Nathan Azrin and his colleagues as a job preparation tool which adopts a group-based behavioral approach to teach job search skills and reinforce social support among participants. Since the development of JOBS program and the Job Club, other intervention programs which either replicate some of the components of these two influential programs or offer different elements have been developed and executed. In these efforts, several theoretical perspectives such as behavioral learning theory (Azrin, Philip, Thienes-Hontos, & Besalel, 1980), theory of planned behavior (Van Hooft, Born, Taris, van der Flier, & Blonk, 2004), social cognitive theory ( Yanar, Budworth, & Latham, 2009), and coping theory (Rife, & Belcher, 1993) have been adopted by researchers (for a summary, see Liu et al., 2014). A number of qualitative reviews and a recent meta-analytic study examined the effectiveness of job search intervention programs. In their recent study, Liu and his colleagues

84 (2014) reviewed 47 experimentally evaluated job search intervention programs and found that the odds of obtaining employment were 2.67 times higher for job seekers participating in interventions compared to job seekers in the control group who did not participate in such programs. To figure out which components are effective in achieving intended outcomes of interventions, Liu and his colleagues (2014) ran moderator analyses. The authors found that the programs that included teaching job search skills, improving self-presentation, boosting self- efficacy, encouraging proactivity, promoting goal setting, and enlisting social support were more effective than interventions that did not include such components. With the help of this recent quantitative analysis, we now know which components of the content are effective however, we do not know whether training format (i.e., providing elective modules vs. structured delivery) makes a difference. This is simply due to lack of research on training formats. In this study, using a randomized field experiment design, we compare search performance of UI recipients who attended a structured program to the performance of other recipients who were provided with the option to choose which workshop to attend.

3.3. Interventions and Other Programs for UI Recipients The UI system is designed to provide temporary financial support to eligible unemployed individuals while they look for a job. To promote reemployment, some parts of UI programs are designed as compulsory. However, compulsory elements in Active Labor Market Policies (ALMPs) have puzzling effects on the search behavior of unemployed individuals (Manning, 2009; van Ours, 2007). On the one hand, job seekers might intensify their search effort to meet requirements, which ultimately leads to exit from unemployment. On the other hand, these elements might introduce extra burden on unemployed and let them quit the labor market. Although they vary across programs, some common compulsory elements in ALMPs are displaying certain amount of search effort (Venn, 2012), reporting effort to the employment agency (Klepinger, Johnson, & Joesch, 2002) and accepting suitable job offers (Abbring, Berg, & Ours, 2005). When these requirements are not met by UI recipients, benefits may be cut or terminated totally (Arni & Schiprowski, 2015). For example, in some states of the U.S., as part of activation policy, unemployed workers have to attend a training session to begin collecting

85 benefits. In some European countries, benefits might get cuts by 50% if the recipient rejects joining a program. To test UI system's effectiveness, researchers have studied the role of various factors in facilitating exit from unemployment, such as monitoring practices (McVicar, 2008, Van den Berg & Van der Klaauw, 2006), the amount of benefit (Centeno, 2004; Roed & Zhang, 2003), the duration of benefit provision (Card & Levine, 2000; Katz & Meyer, 1990), and mandating attendance to job search assistance programs (Klepinger et al., 2002; van Ours, 2007). Mandatory job search assistance programs vary and researchers have studied the effectiveness of different program types such as periodic meetings with a personal advisor (Blundell, Dias, Meghir, & Reenen, 2004), start interviews (Dolton & O'Neill, 1996), and mandatory job search workshops (Graversen & van Ours, 2008). Dolton and O'Neill (1996) studied the effect of Restart program in the U.K., which consists of an interview that is required to collect benefits and found that the program significantly reduced unemployment duration. The authors, in a following study, examined the long-run effects of the Restart and found that the program only helped men but not women in the long-term (Dolton & O'Neill, 2002). In another study in the U.K., Dorsett, Smeaton, and Speckesser (2013) studied the effects of making a voluntary program compulsory and found that mandating participation increased employment rates. Graversen and van Ours (2008) experimentally tested a mandatory activation program in Denmark, which included intensive counseling. The authors found that job finding rate in the treatment group was 30% higher than in the control group. Maibom and his friends (2012) compared mandatory individual meetings with Danish benefit recipients early in their unemployment duration to group meetings and found that clients who attended individual meetings did much better in the labor market than group meeting attendees. Although job search assistance programs were shown to be effective in increasing reemployment rates, the mandatory aspect of these programs and the validity of treatment effect have also been discussed in the literature. Malmberg-Heimonen and Vuori (2005) studied the effects of enforced participation into a job search training on mental health among Finland UI recipients. The authors found that enforced participation in the program impaired mental health of unemployed individuals. Klepinger and his colleagues (2002) studied work-search requirements in Maryland and found that a job-search training requirement reduced unemployment benefit duration but not through enhancing job search productivity, which implies

86 a threat effect rather than a treatment effect. Black, Smith, Berger, and Noel (2003) examined a mandatory training program and found that significant portion of unemployed in the treatment group found a job after receiving the notice about the program but before actually joining it. Similarly, in Sweden, an invitation to a meeting that aims at monitoring and consulting increased exit rate from unemployment among unemployed individuals (Hagglund, 2011). Although the use of mandatory workshops is common in public employment agencies across the globe, no research yet studied whether different delivery formats have differential effects on search outcomes. In this study, we examine whether introducing a room for choice into a mandatory program makes any difference in job searching performance of unemployed individuals.

3.4. Does training delivery format matter? Different forms of training delivery have been a topic of interest for both educational scientists and organizations scholars (Beier & Kanfer, 2009; Goldstein, 1991; Deci, Vallerand, Pelletier, & Ryan, 1991). Researchers in the field of education have studied the antecedents and consequences of taking elective courses by students. Certain motivational factors such as intrinsic motivation and goal orientation were found to be influential on taking more elective courses (Ferrer-Caja & Weiss, 2002; Schweinle, Turner, & Meyer, 2008). These motivational factors are reflected on the evaluations of elective courses. Students evaluate elective courses more positively than non-elective ones (Darby, 2006). On the other hand, past research has also showed that curriculum mandates especially the ones on social and economic issues yield positive outcomes. A course on finance increases savings in adulthood (Bernheim, Garrett, & Maki, 2001), a sustainability class raises environmental awareness (Wu, Huang, Kuo, & Wu, 2010), an ethics course improves moral courage (May, Luth, & Schwoerer, 2014) and a course on healthy sexual behavior decreases unintended pregnancy (Kirby, 1992). Students of organizations have also argued that mandatory training programs may be more effective than optional training if the attitudes of employees toward training is positive (Salas, Tannenbaum, Kraiger, & Smith-Jentsch, 2012; Tannenbaum & Yukl, 1992). However, mandatory trainings do not always yield intended outcomes (Kulik, Pepper, Roberson, & Parker,

87 2007). For instance, Dobbin and Kalev (2016) showed that making diversity training mandatory in companies did not lead to desired positive outcomes. The literature on motivation and learning provides us some ideas about why mandatory training may not be as effective as expected. First, some individuals may already possess skills aimed to be provided by the training (Bernardin & Russell, 1998). Second, individuals as self- determined, autonomous agents might be resistant to external control of their behaviors (Deci & Ryan, 2002). Alternatively, if they believe that the process is under their control, they can master the program (Noe, 1986). The amount of discretion given to workers to decide on participation in training influences motivation to learn and program attendance rates (Hick, 1984). The degree of choice workers have while selecting training opportunities affects their success. Commitment increases under conditions of participation and choice (Salancik, 1977). If individuals are provided with choice, they become bound by their actions and sustain their activities. Organization scholars provided evidence for positive outcomes of voluntarily joining a program. It increased participants' self-esteem and learning motivation as well as knowledge and skill acquisition (Hicks & Klimoski, 1987; Mathieu et al., 1993; Quinones, 1995; Tracey et al., 2001). Although studies on the outcomes of mandatory versus voluntary training are abundant, research on the effects of providing an option for choosing specific elements of training is very limited. Baldwin, Magjuka and Loher (1991) studied the choice of a particular topic rather than the choice to attend and compared three conditions: no choice of training, choice of training but choice is not received and choice of training with choice received. The authors found that the choice received group had higher motivation than other groups, however they did not find any difference in training success across groups. Jennings, Fox, Graves, and Shohamy (1999) conducted an experimental study with students taking a language proficiency test. The authors did not find significant difference between the test scores of students who were not given choice of topic and who were given choice among five topics. When it is comes to providing training to unemployed, the format of the program is an important factor to consider. Unemployed individuals are a group of people who simultaneously experience financial strain and psychological distress. They are at risk of becoming discouraged and leaving the labor market (EPI, 2017; Farber, 2011; Sharone, 2013). It is crucial to facilitate these individuals' transition back to work by providing assistance in the most efficient way. Moreover, to use public resources effectively, it is critical to determine the methods that yield

88 low costs and high benefits. Therefore, paying attention to training format is important when organizing programs for unemployed. Providing a set of training options to unemployed individuals and allowing them to decide which one to attend may yield positive outcomes. For adult learning, free choice of subject matter is crucial (Knowles, Holton III, & Swanson, 2014) and autonomous choice enhances internalization of the material delivered (Deci & Ryan, 2002). However, individuals might not do well in determining their needs; they might overestimate their skills (Kruger & Dunning, 1999), which may ultimately lead to bad choices.

3.5. The current study The aim of this study is to test the effectiveness of two different delivery formats of job search training provided to UI recipients. Using a randomized field experiment, we compare a job search workshop delivered in a structured way to another workshop where attendees are provided with elective options. As a part of activation policy, job search workshop is a mandatory training that UI recipients are required to attend in order to be able to collect their benefits. The mandatory nature of the workshop has generated a debate among scholars and practitioners. On the one hand, job search training programs might encourage unemployed individuals to activate their benefit acquisition and to display sufficient effort. On the other hand, these workshops may be seen as an extra burden by UI recipients and let them quit the labor market. In this paper, we study the effectiveness of mandatory job search workshops delivered in two different ways. Specifically, we compare the effectiveness of a structured workshop which combines a range of selected job search topics to another workshop where participants can choose a section on one specific topic related to job search. Although both training programs are mandatory, whether introducing a room for choice mitigates the overall compulsion effect is an open question (Van Ours, 2007). U recipients especially the ones with higher education and more work and labor market experience might have a good understanding about their skill needs. When they are provided with the choice option, these individuals might become more motivated to acquire the information delivered, which ultimately may lead to higher job search performance. However, the feelings of self-blame and self-doubt are not uncommon during the experience of unemployment (Sharone, 2013).

89 Thus, unemployed individuals may not be able to recognize the skills they possess. Inaccurate assessment of skills may lead to bad choices for training, which ultimately may lead to lower search performance. It is also possible that some individuals might overestimate their skills (Kruger & Dunning, 1999), which again may lead to bad choices and bad performance. Even though the effect of strengthening job search monitoring practices on exit from unemployment has received some attention (Gorter & Kalb, 1996; van Den Berg & van der Klaauw, 2006), the effect of varying mandatory training formats has not been studied. This study has potential to contribute to our understanding by comparing job search outcomes of UL recipients when they are presented with constrained versus unconstrained training choices.

3.6. Methods 3.6.1. Sample and Procedure The sample of this study consists of UI recipients who were clients of a state-run Career Center in Massachusetts. As part of the UL program, clients must attend a workshop consisting of two sessions at the Career Center to be able to collect benefits. This mandatory workshop is provided in two different formats. In the first session of both formats, clients are first instructed about the services of Career Center and the UI program. Then, they were taught about basics of job searching such as using the Internet, resume writing, and networking. The second session varies across workshops. In the first format (for simplicity, hereafter treatment workshop), clients attend a standard and structured session, including specific pieces from a wider range of job search topics and they are not allowed to choose a class. The components of the treatment workshop were chosen based on the previous research as to which topics are most effective for inclusion in job search training. These topics were goal setting, targeting and finding job leads, self-presentation in both writing and speaking, and essential networking steps. In the second workshop (hereafter control workshop), clients choose a class from a list of classes on different job search topics and they are instructed only on that topic in-depth. In both formats, the durations of first and second sessions are around 90 and 150 minutes, respectively. The illustration of workshop formats is provided in Figure 1. The distribution of the electives that the control group attended is provided in the Appendix. For each workshop format, a day of the week is randomly chosen to deliver the workshop. Because individuals who show up on a certain week day are not expected to differ

90 significantly from others who show up another week day, this design meets random assignment of subjects. The workshops were delivered from April 2016 to September 2016. The number of attendees for each session of workshops varied between 20 and 24. In total, 223 and 135 individuals attended both parts of the treatment and control workshops, respectively. These individuals were contacted 2, 4 and 6 months after the first session they attended to be surveyed about their job search experience and employment status. Response rates for each follow-up are provided in Table 2.

3.6.2. Study Variables An email including the survey link was sent to participants in 2, 4, and 6 month follow- ups. Individuals who did not complete the survey online were called and surveyed by phone. The surveys included questions on the following variables. Number or interviews. Participants were asked about the number of phone and the number of in-person job interviews they had. Employment status. Participants were asked whether they found a job. To the ones who found a job, an additional question was asked as to whether the job is temporary or not. Employment Speed. We operationalized employment speed as the number of days between workshop attendance date and start date of the new job. Pay. Those who found a job were asked about their job's pay rate. Job satisfaction. Job satisfaction level was measured with one question: "On a scale from 0-10, how satisfied are you with your new job overall?" Positive mood. We also asked about participants' mood with the question of "On a scale from 0-10, how positive has your mood been over the past month?" Workshop Evaluation. Participants were asked to what extent they agree with the following statement regarding the workshop they attended at Career Center: "Increased my feeling of control over my job search". Responses were given on a 5-point Likert scale (1=strongly disagree, 5=strongly agree).

3.7. Results We first compare the demographics of the participants who attended both parts of the treatment workshop to the demographics of attendees who attended both parts of the control

91 workshop. Table 1 provides group statistics and t-test values. T-test analysis showed that there is no significant difference between groups on education, race, and age. However, there are more females in the treatment group than the control group. Frequencies and percentages of individuals who reported that they found a job in 2, 4, 6- month follow-up surveys are provided in Table 2. At the end of six months period, total number of individuals who reported that they found a job was 83 in the treatment group, which represents 37% of the whole treatment group. In the control group, after six months, 34 individuals reported that they found a job, which represents 25% of the whole control group. The difference in these percentages is statistically significant (x2=5.507, p<.05). These results show that individuals who were in the treatment group which did not have an elective course option were employed at higher rates than the control group which had the elective option. Among employed individuals, the rate of temporary employment did not differ significantly across groups (x2=.458, p>.05) (see Table 3). For continuous variables, independent samples t-test was conducted. The analyses showed that groups did not significantly differ from each other regarding number of interviews, positive mood, and workshop evaluation (see Table 4). Among the employed group, we compared the hourly wage, employment speed, and job satisfaction across groups. The results are provided in Table 5. These findings showed that there is no significant difference between groups in any of the employment outcomes. Additionally, we ran regression analyses to see whether any of the demographic variables predict employment outcomes (see Table 6). The only significant effect was that age predicted employment status. Younger individuals were employed at higher rates than older individuals'. Finally, to test the idea that more educated and experienced individuals might have a better understanding about their needs and therefore make better decisions about which class to take, we tested the interaction effects of age and education with the treatment effect in predicting employment status. The results of these analyses are provided in Tables 8 and 9 and neither of the interaction terms significantly predicted employment status.

1Table 7 provides results of the regression analysis where employment status is regressed only on demographics.

92 3.8. Discussion This paper adds to scarce empirical evidence on the effect of training format on job search outcomes. Previously, unemployment researchers compared group training to individual assistance, however no research yet studied the effectiveness of introducing choice into mandatory training programs. The findings of this study show that allowing UI recipients to choose one from a range of different courses does not increase their employability. In contrast, unemployed individuals who were presented with the choice option had lower rates of employment in six-month time window compared to others who were not presented with this option. The results of this study may be considered as controversial. One might expect that providing course choice to clients would improve employment outcomes. There are two possible explanations for this unexpected finding. One is that clients might be overestimating their skills and ultimately failing to make the optimal choice. Kruger and Dunning (1999) argue that low skilled individuals suffer a dual burden: they not only make wrong choices but also fail to recognize them. This overestimation may let job seekers to fail in making the optimal choice for themselves. Another possible explanation might be related to the nature of job searching. Job search is conceptualized as a dynamic, multi-phased process, involving specific behaviors to identify opportunities and pursue some of those alternatives (Van Hooft, Wanberg, & van Hoye, 2012). To find a job, individuals need to perform reasonably well in all stages. A person who is good at using the Internet to identify opportunities but is not very good at writing cover letters might fail to receive an interview invitation. Another individual who is good at job identification and application stages but lack interview skills may fail to get an offer. Thus, in the case of job search, rather than a focused workshop on one specific topic, a workshop which encompasses a range of aspects of job search, in condensed form yields better results. In addition to its employment effects, implementing the workshop without elective courses would be more cost-efficient. Because administrative cost (i.e., arranging clients' choices) and personnel costs (i.e., expert instructors on elective topics) would be higher in the workshop with electives, the structured workshop without electives might be cheaper for public agencies which have to use their resources effectively. It is also important to note that there is

93 more attrition in the control group than the treatment group with regard to the attendance to the second part of workshops. Because deciding on which elective workshop to choose require more cognitive effort and initiative on behalf of the client than just attending a structured workshop, we think that more attrition has occurred in the control group. This is an important observation that should be taken into account when designing workshops with elective options. Future research with larger samples can study how different training formats influence different segments of the unemployed population (i.e., female, older workers). It is important to figure out what works for whom when it comes to designing labor market policies (O'Leary, Eberts, & Hollenbeck, 2011). One effective training design for future research to test is first conducting needs assessment (Goldstein & Ford, 2002) and then assigning clients to relevant course rather than leaving the choice to them. It is important to improve the effectiveness of public labor market institutions, especially the ones which directly get in touch with unemployed individuals and serve as clearing houses between job seekers and employers. These institutions have to be strong since the cost of unemployment for individuals and society can be severe (Newman, 1999; Sharone, 2013). The primary institution that should serve unemployed individuals is the Unemployment Insurance system and consideration of behavioral factors is important while designing this system.

94 3.9. References Abbring, J. H., Berg, G. J., & Ours, J. C. (2005). The effect of unemployment insurance sanctions on the transition rate from unemployment to employment. The Economic Journal, 115(505), 602-630. Arni, P. P., & Schiprowski, A. (2015). The effects of binding and non-binding job search requirements. IZA DiscussionPaper, No. 8951. Azrin, N. H., Flores, T., & Kaplan, S. J. (1975). Job-finding club: A group-assisted program for obtaining employment. Behaviour Research and Therapy 13(1), 17-27. Azrin, N. H., & Philip, R. A. (1979). The job club method for the job handicapped: A comparative outcome study. Rehabilitation Counseling Bulletin 23(2), 144-155. Azrin, N. H., Philip, R. A., Thienes-Hontos, P., & Besalel, V. A. (1980). Comparative evaluation of the Job Club program with welfare recipients. Journal of Vocational Behavior, 16(2), 133-145. Baldwin, T. T., Magjuka, R. J., & Loher, B. T. (1991). The perils of participation: Effects of choice of training on trainee motivation and learning. PersonnelPsychology, 44(1), 51- 65. Beier, M. E. & Kanfer, R. (2009). Motivation in training and development: A phase perspective. In S.W. J. Kozlowski & E. Salas (Eds.), Learning, Training, and Development in Organizations, 65-97. NY: Psychology Press. Bernardin, H. J., & Russell, J. E. A. (1998). Human resource management. Boston: Irwin/McGraw Hill. Bernheim, B. D., Garrett, D. M., & Maki, D. M. (2001). Education and saving: The long-term effects of high school financial curriculum mandates. Journal of Public Economics, 80(3), 435-465. Black, D. A., Smith, J. A., Berger, M. C., & Noel, B. J. (2003). Is the threat of reemployment services more effective than the services themselves? Evidence from random assignment in the UI system. The American Economic Review, 93(4), 1313-1327. Blundell, R., Dias, M. C., Meghir, C., & Reenen, J. (2004). Evaluating the employment impact of a mandatory job search program. Journal of the EuropeanEconomic Association, 2(4), 569-606. Caplan, R. D., Vinokur, A. D., Price, R. H., & Van Ryn, M. (1989). Job seeking, reemployment, and mental health: A randomized field experiment in coping with job loss. Journalof Applied Psychology, 74(5), 759. Card, D., Kluve, J., & Weber, A. (2010). Active labour market policy evaluations: A meta- analysis. The Economic Journal, 120(548), 452-477. Card, D.E., & Levine, P.B. (2000). Extended benefits and the duration of UI spells: Evidence from the New Jersey extended benefit program. Journal of Public Economics, 78, 107-138. Centeno, M. (2004). The match quality gains from unemployment insurance. Journalof Human Resources, 39(3), 839-863. Clark, A. E., & Oswald, A. J. (1994). Unhappiness and unemployment. The Economic Journal, 104(424), 648-659. Darby, J. A. (2006). The effects of the elective or required status of courses on student evaluations. Journalof Vocational Education and Training 58(1), 19-29. Deci, E. L., & Richard M. R. (2002). Handbook of self-determination research. University Rochester Press.

95 Deci, E. L., Vallerand, R. J., Pelletier, L. G., & Ryan, R. M. (1991). Motivation and education: The self-determination perspective. EducationalPsychologist, 26(3-4), 325-346. Dobbin, F., & Kalev, A. (2016). Why Diversity Programs Fail and what works better." HarvardBusiness Review 94, (7-8), 52-60. Dolton, P., & O'Neill, D. (1996). Unemployment duration and the restart effect: Some experimental evidence. The Economic Journal, 106(435), 387-400. Dolton, P. & O'Neill, D. (2002) The long-run effects of unemployment monitoring and Work search programmes: Experimental evidence from the United Kingdom. Journalof Labor Economics, 2, 381-403. Dorsett, R., Smeaton, D., & Speckesser, S. (2013). The effect of making a voluntary labour market programme compulsory: Evidence from a UK experiment. FiscalStudies, 34(4), 467-489. EPI (2017). Missing Workers. Retrieved from http://www.epi.org/publication/missing-workers/ Farber, H. S. (2011). Job Loss in the Great Recession: Historical Perspective from the Displaced Workers Survey, 1984-2010. NBER Working Paper, 17040. Ferrer-Caja, E., & Weiss, M. R. (2002). Cross-validation of a model of intrinsic motivation with students enrolled in high school elective courses. The Journalof ExperimentalEducation, 71(1), 41-65. Goldstein, I. L. (1991). Training in Work Organizations. In J.D. Dunnette & L.M. Hough (Eds.). Handbook of Industrialand OrganizationalPsychology, 507-620. Palo Alto, CA: Consulting Psychologists Press. Goldstein, I. L., & Ford, J. K. (2002). Trainingin organizations:Needs assessment, development, and evaluation. Pacific Grove, CA: Brooks-Cole. Gorter, C., & Kalb, G. R. (1996). Estimating the effect of counseling and monitoring the unemployed using a job search model. Journalof Human Resources, 31(3), 590-6 10. Graversen, B. K., & Van Ours, J. C. (2008). How to help unemployed find jobs quickly: Experimental evidence from a mandatory activation program. Journal of Public Economics, 92(10), 2020-2035. Guerrero, S., & Sire, B. (2001). Motivation to train from the workers' perspective: Example of French companies. InternationalJournal of Human Resource Management, 12(6), 988- 1004. Hagglund, P., (2011). Are there pre-programme effects of Swedish active labour market policies? Evidence from three randomized experiments. Economic Letters, 112(1), 91-93. Hicks, W. D., & Klimoski, R. J. (1987). Entry into training programs and its effects on training outcomes: A field experiment. Academy of ManagementJournal, 30(3), 542-552. Jahoda, M. (1981). Work, employment, and unemployment: Values, theories, and approaches in social research. American Psychologist, 36(2), 184. Jennings, M., Fox, J., Graves, B., & Shohamy, E. (1999). The test-takers' choice: An investigation of the effect of topic on language-test performance. Language Testing, 16(4), 426-456. Katz, L. F., & Meyer, B. D. (1990). The impact of the potential duration of unemployment benefits on the duration of unemployment. Journalof Public Economics, 41(1), 45-72. Kirby, D. (1992). School-based programs to reduce sexual risk-taking behaviors. Journal of School Health, 62(7), 280-287. Klepinger, D. H., Johnson, T. R., & Joesch, J. M. (2002). Effects of unemployment insurance work-search requirements: The Maryland experiment. ILR Review, 56(1), 3-22.

96 Kluve, J. (2010). The effectiveness of European active labor market programs. Labour economics, 17(6), 904-918. Knowles, M. S., Holton III, E. F., & Swanson, R. A. (2014). The adult learner: The definitive classic in adult education and human resourcedevelopment. Routledge. Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journalof Personalityand Social Psychology, 77, 1121-1134. Krueger, A. & Mueller, A. (2011). Job search, emotional well-being, and job finding in a period of mass unemployment: Evidence from high frequency longitudinal data. Brookings Papers on Economic Activity. Kulik, C. T., Pepper, M. B., Roberson, L., & Parker, S. K. (2007). The rich get richer: Predicting participation in voluntary diversity training. Journal of OrganizationalBehavior, 28(6), 753-769. Liu, S., Huang, J. L., & Wang, M. (2014). Effectiveness of job search interventions: A meta- analytic review. PsychologicalBulletin, 140(4), 1009. Maibom P. J., Rosholm, M., & Svarer, M. (2012). Experimental evidence on the effects of early meetings and activation. IZA Discussion PaperSeries, 6970. Malmberg-Heimonen, I., & Vuori, J. (2005). Activation or discouragement-the effect of enforced participation on the success of job-search training. European Journal of Social Work, 8(4), 451-467. Manning, A. (2009). You can't always get what you want: The impact of the UK Jobseeker's Allowance. LabourEconomics, 16(3), 239-250. May, D. R., Luth, M. T., & Schwoerer, C. E. (2014). The influence of business ethics education on moral efficacy, moral meaningfulness, and moral courage: A quasi-experimental study. Journalof Business Ethics, 124(1), 67-80. Mathieu, J. E., Martineau, J. W., & Tannenbaum, S. I. (1993). Individual and situational influences on the development of. PersonnelPsychology, 46(1), 125. McVicar, D. (2008). Job search monitoring intensity, unemployment exit and job entry: Quasi- experimental evidence from the UK. Labour Economics, 15, 1451-1468. Newman, K. S. (1999). FallingFrom Grace: Downward Mobility In The Age Of Affluence. New York, NY: Free Press. Noe, R. A. (1986). Trainees' attributes and attitudes: Neglected influences on training effectiveness. Academy of Management Review, 11(4), 736-749. O'Leary, C. J., Eberts, R. W., & Hollenbeck, K. (2011). What works for whom in public employment policy? Kalamazoo: W.E. Upjohn Institute for Employment Research. Paul, K. I., & Moser, K. (2009). Unemployment impairs mental health: Meta-analyses. Journal of Vocational behavior, 74(3), 264-282. Quinones, M. A. (1995). Pretraining context effects: Training assignment as feedback. Journalof appliedpsychology, 80(2), 226. Rife, J. C., & Belcher, J. R. (1994). Assisting unemployed older workers to become reemployed: An experimental evaluation. Research on Social Work Practice,4(1), 3-13. Roed, K., & Zhang, T. (2003). Does unemployment compensation affect unemployment duration?. The Economic Journal, 113(484), 190-206. Salancik, G. R. (1977). Commitment and the control of organizational behavior and belief. In B. M. Staw & G. R. Salancik (Eds.), New directions in organizational behavior: 1-54. Chicago: St. Clair Press.

97 Salas, E., Tannenbaum, S. I., Kraiger, K., & Smith-Jentsch, K. A. (2012). The science of training and development in organizations: What matters in practice. Psychological Science in the Public Interest, 13(2), 74-101. Schweinle, A., Turner, J. C., & Meyer, D. K. (2008). Understanding young adolescents' optimal experiences in academic settings. The Journalof Experimental Education, 77(2), 125- 146. Sharone, 0. (2013). Flawed System/Flawed Self: Job Searching and Unemployment Experiences. Chicago, IL: University of Chicago Press. Sharone, 0., & Vasquez, A. (2016). Sociology as a Strategy of Support for Long-Term Unemployed Workers. The American Sociologist, 1-20. Smith, V. (2002). Crossing the great divide: Worker risk and opportunity in the new economy. Cornell University Press. Tracey, J. B., Hinkin, T. R., Tannenbaum, S., & Mathieu, J. E. (2001). The influence of individual characteristics and the work environment on varying levels of training outcomes. Human Resource Development Quarterly, 12(1), 5. Tannenbaum, S. I., & Yukl, G. (1992). Training and development in work organizations. Annual review ofpsychology, 43(1), 399-441. Van den Berg, G. J., & Van der Klaauw, B. (2006). Counseling and monitoring of unemployed workers: Theory and evidence from a controlled social experiment. International Economic Review, 47(3), 895-936. Van Hooft, E. A., Born, M. P., Taris, T. W., Flier, H. V. D., & Blonk, R. W. (2004). Predictors of job search behavior among employed and unemployed people. PersonnelPsychology, 57(1), 25-59. Van Hooft, E. A., Wanberg, C. R., & Van Hoye, G. (2013). Moving beyond job search quantity: Towards a conceptualization and self-regulatory framework of job search quality. OrganizationalPsychology Review, 3(1), 3-40. Van Horn, C. E. (2014). Working scared (or not at all): The lost decade, great recession, and restoringthe shatteredAmerican dream. London: Rowman & Littlefield. van Ours, J. C. (2007). Compulsion in active labour market programmes. National Institute Economic Review, 202(1), 67-78. Venn, D. (2012). Eligibility criteria for unemployment benefits: Quantitative indicators for OECD and EU countries. OECD Social, Employment and Migration Working Paper, 131. Vinokur, A. D., Price, R. H., & Schul, Y. (1995). Impact of the JOBS intervention on unemployed workers varying in risk for depression. American Journalof Community Psychology, 23(1), 39-74. Vuori, J., Silvonen, J., Vinokur, A. D., & Price, R. H. (2002). The Ty6h6n job search program in Finland: Benefits for the unemployed with risk of depression or discouragement. Journal of OccupationalHealth Psychology, 7(1), 5-19. Wanberg, C. R. (2012). The individual experience of unemployment. Annual Review of Psychology, 63, 369-396. Warr, P. (1987). Work, unemployment, and mental health. Oxford: Clarendon Press. Wu, Y. C. J., Huang, S., Kuo, L., & Wu, W. H. (2010). Management education for sustainability: A web-based content analysis. Academy of Management Learning& Education, 9(3), 520-531. Yanar, B., Budworth, M. H., & Latham, G. P. (2009). The effect of verbal self-guidance training

98 for overcoming employment barriers: A study of Turkish women. Applied Psychology, 58(4), 586-601.

99 3.10. Tables

Table 1. Group Statistics Mean SD t-value Female Treatment .52 .501 2.226* Control .40 .491 Age Treatment 45.43 13.134 .997 Control 43.99 12.492 Education Treatment 4.83 1.305 1.842 Control 5.08 1.109 White Treatment .67 .472 .734 Control .71 .457 *p<.0 5

100 Table 2. Response Rates and Employment Status Treatment Group (N=223) Control Group (N=135) # of 2-month survey respondents 104 44 (response rate) (47%) (33%) Employed 29 9 Unemployed 75 35

# of 4-month survey respondents 69 26 (response rate) (36%) (21%) Employed 25 8 Unemployed 44 18

# of 6-month survey respondents 76 39 (response rate) (45%) (33%) Employed 29 17 Unemployed 47 22

Total # of respondents in 3 waves 165 76 (response rate) (74%) (56%) Employed 83 (37%) 34 (25%) Unemployed 82 42

101 Table 3. Employment in Temporary Job Control Treatment Yes 7 (22%) 23 (29%) No 25 (78%) 57 (71%) Total 32 (100%) 80 (100%)

102 Table 4. Means and T-test Values for Job Search Outcomes 2-months 4-months 6-months # of phone Elective course 6.81 4.22 4.89 interviews group Non-elective 5.68 3.95 3.48 course group t value .835 .264 1.986

# of in-person Elective course 4.10 2.64 3.03 interviews group Non-elective 3.24 2.67 2.67 course group t value 1.218 .061 .675

Mood Elective course 6.71 6.96 6.39 group Non-elective 7.03 6.73 6.39 course group t value 1.092 .433 .002

Workshop Elective course 4.03 Evaluation group Non-elective 4.10 course group t value .435

103 Table 5. Means and T-test Values for Employment Outcomes Mean N Employment Elective course 91.67 30 Speed group Non-elective 88.58 77 course group t value .277

Hourly pay Elective course 38.43 23 group Non-elective 32.92 65 course group t value 1.333

Job satisfaction Elective course 7.93 29 group Non-elective 8.15 67 course group t value .551

104 Table 6. Regression results for Employment Outcomes Employment Status Reemployment Speed Job satisfaction Exp(B) B B Female 1.009 -.074 -.032 Age 957** .225* -.031 Education 1.004 -.069 .046 White 1.200 -.108 -.009 Treatment dummy 1.302 -.044 .060

Cox & Snell R Square .075 R Square .063 .008 **p<.001; *p<.05

105 Table 7. Regression results for Employment Outcomes Employment Status Exp(B) Female 1.032 Age .957** Education .999 White 1.196

Cox & Snell R Square .072 **p<.001; *p<.05

106 Table 8. Logistic Regression Results for Age Interaction Employment Status Exp(B) Age .954** Treatment*Age 1.005

Cox & Snell R Square .073 **p<.001

Table 9. Logistic Regression Results for Education Interaction Employment Status Exp(B) Education .979 Treatment*Education 1.036

Cox & Snell R Square .002

107 3.11 Figures

Figure 1. Workshop Delivery Formats

Job Search Workshop Job Search Workshop Format 1 (Treatment) Format 2 (Control)

Career Center Career Center Workshop Workshop Part A Part A (-u90 min.) ('-9O min.)

+ Resume Writing Career Center Elective * NetwOrking + w th Confidence Workshop Workshop Interview Part B * Career Exploration Part B (~S m ) SSalary Negotiation (-150 min.)

._ -- -...... , _ __..

108 3.12 Appendix

Appendix A. Distribution of Elective Workshops Control Group Attended Elective Workshop Frequency Percent Salary Negotiation 28 20.7 Interviewing with Confidence 20 14.8 Career Exploration 17 12.6 Linkedln Basics: Effective Profiles 16 11.9 Networking 10 7.4 What's age got to do with it? 10 7.4 Resume Workshop 9 6.7 Basic Online Job Applications 6 4.4 Entrepreneur Think Tank 5 3.7 Dislocated Worker Workshop 4 3 Industry Briefing 3 2.2 Job Search Essentials 3 2.2 Coping with Job Loss 2 1.5 Career Services Speaker Series 1 0.7 DLW Job Search Team 1 0.7 Total 135 100.0

109 110 4. Why Is Regulatory Compliance Difficult? Variable Performance and the Insulation of Economically Resourceful Actors

4.1. Introduction Environmental, health and safety (EHS) concerns at the workplace have captured the attention of scholars from different disciplines for several decades. Although research findings show that legal sanctions and governmental regulations improve EHS conditions in organizations, non-compliance with rules and regulations still persists (Silbey, 2013). To reduce non-compliance, regulatory agencies have encouraged organizations to adopt self-regulation, conceived as a strategy that falls between free market and public regulation (Ayres & Braithwaite, 1992; Coglianese & Nash, 2001). Various forms of self-regulation such as codes of conduct, compliance offices, and grievance procedures have been proposed to encourage organizations to comply not only with legal requirements but also with industry standards (for reviews, see King & Lenox, 2000; Short & Toffell, 2010). One of the self- regulation methods proposed to increase organizational compliance with EHS standards is a Management System (MS). Although EHS Management Systems were welcomed by leaders of organizations, health and safety practitioners, and academic scholars, little is understood about their effectiveness. In this paper, we examine an EHS Management System (EHS-MS) established in a large research university in the United States, and address the following questions: Is the system effective in controlling safety hazards in academic laboratories? Do scientific units (i.e., labs) vary in their reaction to the system? If such variance exists, what are the factors that affect differential reaction to inspections of laboratory safety practices? Using longitudinal quantitative data from laboratory inspections, we investigate whether the university's EHS-MS achieves compliance with safety rules and regulations. The analyses show that over a seven year period, the EHS-MS did not decrease the total number of safety violations observed in semi-annual inspections. However, a close examination of this unexpected outcome shows that an overall increase in the number of violations is driven by certain labs with high financial resources. Furthermore, to understand the reasons that hinder the success of the system, we analyze qualitative data from interviews with important stakeholders and agents of the system such as

111 safety inspectors and EHS personnel. The interview data provide important insights about the factors that contribute to differential compliance performance across academic labs. This study has potential to contribute to existing understandings of safety culture. By extending the empirical domain, we examine the effectiveness of an EHS-MS that is welcomed enthusiastically in academic circles, but not yet tested sufficiently. Moreover, this study sheds light on why some actors follow system requirements while others do not. Previous literature on compliance with regulations has not focused on the patterns of this variation, nor has it focused on individual variation. By understanding factors that predict adoption/compliance, internal interventions can be developed to incentivize non-compliant actors to change their behaviors. In the next section, we provide a brief review of the literature on self-regulation and EHS management systems. We argue that the lack of quantitative review of these systems' effectiveness limits our understanding of whether and under what conditions these systems work. We then describe the research setting and discuss why establishing EHS management systems in universities creates an especially difficult challenge, similar to what we would expect under similar conditions in other networked and knowledge-based organizations (Bailey, Leonardi, & Chong, 2010; Carlile, 2002; 2004) . We follow with the presentation of quantitative findings, providing evidence to demonstrate how the system has worked over time. We show what characteristics differentiate non-compliant labs from compliant ones. We then present qualitative data to identify the factors that are seen by organizational actors as barriers to the success of the system. Finally, we conclude with a discussion of the implications of this study for our understanding of differential reactions to EHS management systems and their effectiveness in academic settings as well as self-regulation in complex, hazardous organizations more broadly.

4.2. Regulatory Compliance through Self-Regulation Compliance with legal regulations has been the focus of attention among economists, sociologists, political scientists, and scholars of regulation and governance. Regulations, while enacted in response to multiple interests, are publicly justified by their capacity to channel behavior of regulated entities in ways that enhance social welfare (Coglianese, 2016). Empirical evidence shows that compliance with legal regulations and rules more often than not improves the lives of employees, protects the environment, and makes the world a safer place (for reviews, see Schneiberg & Bartley, 2008; Silbey, 2009).

112 Organizations comply with regulatory requirements in response to beliefs in the legitimacy of regulations, competitive pressures, threats of legal sanctions, or informal sanctions such as negative publicity and other social costs (Gunningham, Kagan, & Thornton, 2003). However, noncompliance persists, and the investigation of the antecedents and consequences of non-compliant behavior receives continuing attention from scholars. Previous research has studied variation in compliance (and non-compliance) across organizations in the same industry (e.g., Delmas & Toffel, 2008; Edelman & Suchman, 1997; Kellogg, 2009), and across groups in the same organization (e.g., Gray & Silbey, 2014; Heimer & Staffen, 1998). However, it is important to note that no research has yet looked at variation across individuals in one actor group. This is an important gap in our knowledge because it is not unusual to observe high variation in attitudes and behaviors across individuals, especially in the case of high status groups where individuals enjoy high levels of autonomy (Gray & Silbey, 2014; Evans, 2014). One proposed way to increase organizational compliance is self-regulation. Self- regulation refers to rules that are imposed by an organization on itself (Coglianese & Mendelson, 2010). Conceived as falling between government-imposed regulation and laissez-faire policies (Ayres & Braithwaite, 1992; Coglianese & Nash, 2001; Rees, 1997), self-regulation can shape behavior through locally developed mechanisms such as transferring norms or diffusing best practices (Nash & Ehrenfeld, 1997). However, it has been argued that in order for self-regulation to be effective, the institutional structure of the organization or industry should facilitate the use of informal coercive, normative and mimetic means (DiMaggio & Powell, 1991; Gunningham, 1995; King & Lenox, 2000). The use of these means can encourage actors to mimic compliance efforts and abandon non-compliant behaviors, which ultimately makes self-regulation successful. Although the success of self-regulatory systems has been studied in industrial settings (e.g., Khanna & Damon, 1999; Ebenshade, 2004), we lack information about their effectiveness in organizations of high status, relatively autonomous actors, such as professional or academic settings, including scientific laboratories.

4.2.1. A Form of Self-regulation: EHS Management Systems Management Systems consist of digitized technologies that bundle various software packages to enable the centralized collection and archiving of data for managing diverse informational and procedural transactions of large complex organizations. "What had been

113 dispersed and often idiosyncratic forms of observing, inspecting, auditing and reporting about organizational behavior have been brought together under the umbrella of enterprise resource systems, sometimes called management systems" (Huising & Silbey 2016; p. 810). Ostensibly, these systems can be easily searched and analyzed for continuous observation and improvement toward organizational goals. "If used to their technological capacity, management systems move from regulating spaces to surveilling organizational members, groups and practices, seeking to control all forms and aspects of risk, including the behavior of scientists themselves." (Huising & Silbey 2016; p. 810). As a form of self-regulation, management systems are proposed to improve EHS performance standards in organizations. While public regulation imposes rules and regulations from outside, the regulatory structure of a management system emerges within organization (Coglianese & Nash 2001). In management-based regulation, regulated entities engage in internal planning and rulemaking to achieve specific, publicly sanctioned goals (Coglianese & Lazer, 2003). More specifically, management systems are a form of governance that locates the design and implementation of regulation within the organization (Howard-Grenville, Nash, & Coglianese, 2008). While management systems make every actor in the organization responsible for compliance through decentralization, the information about compliance performance is collected and analyzed centrally to standardize the functioning of the system (Husing & Silbey, 2013). One of the distinguishing characteristics of management systems is the ability to address organizational needs by self-reflection using information channels and feedback loops. Although the specifics of EHS management systems such as goals, style of monitoring, and sanctions vary across organizations, basic characteristics of the system like developing an EHS plan, assigning roles and responsibilities, monitoring performance and correcting problems are common among organizations (Coglianese & Nash, 2001). Environmental management systems, following the protocols of the International Organization for Standardization under ISO 14001 and promoted by the Environmental Protection Agency (EPA) have become a popular means for managing compliance in organizations (Kagan, Gunningham, & Thornton, 2003). It has been argued that, especially for achieving occupational health and safety, management systems are more powerful than command and control regulations (Gunningham, 1999). For a number of reasons, management systems yield better compliance. First, because local and internal strategies are adopted, it

114 reduces costs associated with adaptation of inappropriate elements, processes, or criteria. Second, because members with expert local knowledge are involved in the creation of the system, they invest in the system's effectiveness, identifying locally appropriate strategies as well as criteria and processes. Although the adoption of EHS Management Systems by organizations has increased greatly in recent decades (Darnall & Edwards, 2006), lack of implementation data has impeded empirical research investigating the effectiveness of these systems. Audit records are often not made public nor available to scholars due to organizations' concerns for trade secrets as well as reputation. In this study, using longitudinal data of system outputs, we investigate whether an EHS-MS established in a large research university is effective in increasing compliance and reducing the number of safety violations.

4.2.2. EHS Management Systems in Academic Laboratories EHS management systems have been seen by the EPA as a potential means of identifying and fixing noncompliance in academic laboratories by creating systematic self-observation and response (Huising & Silbey, 2011). However, for self-regulated management systems to be effective, the underlying institutional structure should be procedurally coordinated and normatively consistent. For instance, previous research on compliance with labor standards has shown that private voluntary regulation is effective when organizations design their work organization and human resource management policies in a way that complement traditional monitoring efforts (Locke & Romis, 2010; Locke, Qin, & Brause, 2007). Academic laboratories have been observed to be intractable governance sites, with extensive faculty discretion and autonomy (Huising & Silbey, 2013). These prerogatives make scientific laboratories relatively closed to outside interventions. Yet, academic laboratories are risky places. Researchers lose eyes, limbs and sometimes life itself in laboratory accidents. In addition, non-compliance with safety rules may lead to certain penalties such as cessation of operations, exclusion of individuals from labs, or monetary fines. Thus, securing safety of the students, technicians, as well as faculty has become a critical goal, with an explicit mandate to administrators and support staff to improve the safety culture in academic institutions (National Research Council, 2014). However, the institutional structure of academia makes the improvement of safety culture in these places a challenge.

115 Universities, on the one hand, are bureaucratic organizations with top-down managerial controls; on the other hand they are composed of decentralized autonomous units. Certain characteristics of academia such as professional status of faculty members, norms of academic freedom, privileges of scientific autonomy, the opacity of scientific work to outsiders, insufficient information flow from middle level managers to upper levels of administration, and a perennially transient workforce make legal interventions difficult, resulting in a loose coupling of policy and practice (Huising & Silbey, 2013; Silbey & Ewick 2003; Weick, 1976). Although academic institutions are self-governed - rather than totally managed from the top down - the institutional structure is not conducive to the introduction nor embrace of (self) management- based regulation. Gunningham (1995) argues that because explicit external sanctions are not present in self-regulation, alternative informal means should be available to make the system work. Again, these means can be coercive, normative, or mimetic forces. Use of informal coercive forces including management or peer pressure on bad actors through publicity is not easily feasible in the case of scientists who occupy higher-status positions than EHS managers and who regularly defer to each other's independence. Scientists' authority based on their knowledge and expertise in their respective field as well as based on large research funds they manage give them power in their transactions with other organizational members, ultimately providing shield against external pressures (Weber, 1947; Wrong, 1995). Although normative and mimetic forces were often thought to be enhance compliance through diffusion of norms and best practices, scientists cultivate their unique identities and guard the boundaries of their own labs. Labs are detached from each other and each lab has a unique culture. Even though the outcomes of science are universal, the practices that generate the outcomes are often local and idiosyncratic (Knorr & Cetina, 1999). Between scientists, there is not much communication on topics other than research. Thus, the nature of social relations among scientists and the organization of universities as loosely coupled systems make transfer of norms or practices difficult. In this paper, we study an EHS-MS established in a large American university (hereafter Eastern University) to see how the system has affected safety violations in research laboratories. We examine whether scientists vary in their reaction to the management system, and also investigate the factors that lead to differential reactions to the system.

116 It is important to note that research on EHS management systems in academic labs is scarce. Huising and Silbey (2011) studied the creation of an EHS-MS at a research university, observing how front-line managers achieve compliance through relational regulation. In a subsequent paper, they studied challenges faced by safety managers in making faculty members responsible and accountable (Husing & Silbey, 2013). However, no research to date has examined the effectiveness of EHS management systems in academia using actual records of inspections. In this paper, beyond the examination of system's effectiveness, using both quantitative and qualitative data, we study factors that lead to variation in actors' compliance performance.

4.3. The Current Study: Academic Laboratories at Eastern University 4.3.1. Setting Eastern University is a major research university in the eastern United States with over 400 laboratories. In the past, Eastern had delegated compliance responsibilities to groups of professionals who operated relatively independently from direct management oversight, working in accordance with the occupational norms of their distinct fields such as industrial hygiene, chemical and biological waste management, and radiation protection. During the 1990s, the U.S. Environmental Protection Agency (EPA) inspected Eastern's laboratories as part of a higher education initiative. Although the inspectors found no major violations, they found over 3000 minor violations. Most significantly, the EPA concluded that Eastern could not account for how it managed to prevent spills and emissions. Thus, the EPA required the creation of a compliance management system to create consistent practices across the university. Eastern agreed as part of a consent decree to create an EHS management system, thereby limiting fines and avoiding public embarrassment. The major intended change initiated through the management system was the reassignment of compliance responsibility from centralized administrative staff professionals to the academic departments. Departmental administrators and researchers became responsible for ensuring that their research practices complied with city, state, and federal EHS regulations. To assist the researchers and academic departments with their new responsibility, the position of an EHS coordinator was created. Each academic department hired a coordinator whose job was to oversee laboratory compliance within the department, ensuring that researchers integrated concern for safety and the environment into their research protocols and practices. The

117 coordinators would be supported by the central staff of professionals (e.g. radiation, biohazards, air quality specialists) who would provide in-depth expertise, training, and coordination across all university departments.

4.3.2. Data: The number of safety violations Although the in-house development of Eastern's EHS-MS began in 2001, the system started running in its complete form in 2006. The data exploited in this study include inspection findings recorded by department coordinators from 2006 fall to 2012 spring (12 inspections in total). The observation of EHS practices at Eastern University and the actual record of EHS violations in this time window provide a unique opportunity to isolate some of the factors that might interfere with inspection outcomes. In the time period observed, each laboratory had been inspected by the same coordinator twice a year and all coordinators used the same inspection protocol and recording template. These two important characteristics of the setting keep coordinators and inspection protocols constant throughout the data while ruling out the possibility that inspection outcomes were affected by changes in inspectors or inspection templates. After each inspection, feedback about findings (observed discrepancies between regulations and performance) was provided to the Principal Investigator (PI) (faculty member) of each lab, with instructions for the responsible PI to fix the observed EHS problems. The data in this study include inspection records from 304 labs in nine departments. Although the total number of labs at Eastern is higher, we eliminated teaching labs, art studios, and those with very few hazardous substances. In the time period we study, across 12 inspection cycles, a total of 9340 violations were recorded. The number of violations by department is provided in Table 1. The main variable of interest in this study is the number of EHS violations in labs, which had been recorded by coordinators during semi-annual inspections. These violations include instances of any type of non-compliance with EHS regulations from minor issues like the untidiness of the lab to major ones such as the mismanagement of chemical waste or failure to wear personal protective equipment (e.g., glasses, gloves, lab coats). The observed violations in the lab had been recorded by inspectors in open text formats. These open-text violations were coded into twelve categories. The categories and sample violations for each category are provided in the Appendix A.

118 4.4. Results 4.4.1. DescriptiveAnalysis of Violations: Change in the Number of Violations over Time Figure 1 presents the change in the number of total violations in university labs over the years studied. The graph shows that, from 2006 to 2012, the total number of violations at Eastern had increased over time. In contrast to the intended goal, the management system failed to decrease the number of inspection findings or in other words safety violations. To have a better understanding of this unexpected trend, we examine over time change in percentiles. A closer examination shows that the number of violations at the median do not show an increase over time. However, the number of violations in 7 5th Igth and 9 5 th percentiles show an upward trend (see Figure 2). These results indicate a widening gap over time between the median and the higher percentiles. It is essential to note that the labs in each percentile in Figure 2 might not necessarily be the same labs across years. Therefore, in further examination, we flagged labs based on their percentile rank in 2006 and followed same labs over time. As seen in Figure 3, the bulk of labs do not increase their rate of violation over years. However, those few labs for whom the highest number of violations were observed and recorded in 2006 (9 5th percentile) do tend to violate more over time. Up to this point, by looking at the number of violations in the whole university, we show that although the majority of labs do not increase their safety violations over time, a small group of high violators show increasing numbers of violations. This finding might be due to the variation across inspectors since departments have different inspectors and inspectors might have varying background, monitoring styles or learning capabilities. Because each department had the same inspector in the time period observed, we look at within-department compliance performances to see whether the pattern observed at the university level is also observed at the department level. This within-department observation allow us to keep inspectors constant. The departmental-level analysis shows an increase in the number of violations over time within departments as well, not simply across departments or across coordinator/inspectors. The departmental-level analysis also shows that a few high violators in each department have been driving the total number of violations of the department. Shapiro-Wilk statistics provided in Table 2 indicate that the distribution of findings in seven out of nine departments significantly differ from normal distribution.

119 In sum, each department's high violators showed increased violations over the six year time period. However, for the majority of labs, the number of violations did not significantly change. Thus, the pattern observed at the university level is repeated at the department level, which is that the total number of findings are driven by a few labs in each department. It is important to note that because each department is inspected by the same inspector throughout the time window studied, the findings are not affected by inspector taste or differential learning (Esbenshade, 2004). In addition, as mentioned earlier, because the same materials were used by coordinator/inspectors in the observed window, variation in monitoring protocols also do not explain this variation (Jenkins, 2001; O'Rourke, 2003).

4.4.2. InferentialAnalysis of Violations To differentiate high safety violator labs from low violator ones, we investigate the predictors of the number of violations. For this analysis, we collected data on a group of possible predictor variables listed below from organizational archives and a variety of public sources.

PI level predictorvariables: Demographics. In the regression analyses, we use a host of PI level demographic variables including age, gender, race (White vs. non-White), and country of origin (U.S. vs. non- U.S.). Taking into account these variables in this study is especially important because past research showed that in environments where collaboration between different actor groups is required (i.e, ensuring safety in an organization), demographics of actors matter. More specifically, in a study of high-status actors (i.e., scientists), it is important to consider cross- cutting demographics because the interaction of different status characteristics such as high occupational status and low demographic status (i.e., female scientist) might lead to varying levels of positive experiences with the low-status actor group (i.e., inspectors) (DiBenigno & Kellogg, 2014). Employment and academic variables. We take into account the duration of the scientist at Eastern (in years) as well as whether the PI has tenure. Since tenured faculty enjoy greater autonomy and privilege, their reactions toward interventions might be different from non-tenured faculty (Gray & Silbey, 2014). In addition, we control for the quality of PIs' Ph.D. degree

120 granting institution, using rankings from the Times Higher Education World Universities Rankings. Research output. To take into account research output of each PI in the time window observed, we determined the number of publications for each year from 2006 to 2013. Because publications are often outcomes of previous year(s), we treated it as a lagged variable by associating the number of violations in year y with the number of publications in year y+1 in the analyses. Funding. The amount of yearly research budget for each PI, from 2006 to 2010, was collected from university archives. The distribution of research funding based on source is provided in the Appendix B. Lab and department. We control for lab size operationalized as the number of people in the lab. We also control for the department that the lab belongs to because disciplinary organizations may shape local practices (Silbey, 2016; Whitley, 2000). Descriptive statistics of the study variables are provided in Table 3 and Table 4 presents correlations among variables. For inferential analyses, we run random and fixed effects models. Panel data analysis with random effects model is provided in Table 5. The random effects model assumes that the variation across units is random and uncorrelated with the predictors included in the model. In the first random effects model, we regress the number of violations on demographic variables (age, gender, race, and country of origin). None of the demographic variables significantly predicted the number of violations. In the second model, we add academic and employment- related variables (time spent at Eastern, tenure status, and Ph.D. granting institution ranking). The results indicate that tenured PIs have significantly more violations than non-tenured ones. In Model 3, we add into the regression equation the faculty member's volume of research funding, measured in dollars. The results show that holding demographic, employment, and academic variables constant, research funding significantly predicts the number of violations observed during inspections. In laboratories where more people work and where more work is done, the number of violations recorded might be higher. To take this into consideration, in models 4 and 5, we add the number of publications by year and the number of individuals in the lab, respectively. Even

121 after controlling for these variables, the research budget still remains as the only significant predictor of the number of recorded inspection violations. Although we included a host of control variables in the random effects model, we cannot totally rule out the existence of unobserved heterogeneity across individual labs. To deal with this, we run PI fixed effects analysis which allows us to rule out heterogeneity across labs (see Table 6). Three time-variant variables that we use in fixed effects analyses are the number of violations, the yearly research budget in dollars, and the number of publications published in a given year. To see whether research budget affects compliance performance, in the first fixed effects model, we regress the number of violations on the budget amount. The results of this model show that the relationship between budget amount and violation performance is marginally significant. In the second model, we add number of publications, and the relationship between budget amount and violation performance still remains marginally significant. These results suggest that after controlling for number of publications in the following year, as the amount of research funding increases in a given year, the amount of violation intensifies. As a robustness check, to see whether there is a differential violation performance across labs based on their percentile in terms of the amount of research budget, we introduced dummies for each lab's percentile in a given year and run fixed effects regression analysis, controlling for publication performance. The analysis show that labs in 7 5 th percentile and above violate at significantly higher rates than labs below median in terms of research budget. As seen in Table 7, when the percentile increases, the predictive power increases as well.

4.5. Qualitative Data Analysis The quantitative analyses of safety violation records at Eastern University provided important findings. First, the system is not effective in reducing the number of observed violations and thus does not achieve a higher rate of compliance with EHS rules and regulations. Second, there is a differential reaction to the system. A significant amount of variance in compliance across labs within departments is observed. Particular individuals in each department are outliers, consistently performing with a higher than average rate of violations. Finally, inferential analyses showed that in labs where PIs are tenured and have more abundant financial resources, the number of violations tend to be higher and increase with time. Now, to better

122 There is one very important thing, and that has to do with PIs. The cooperation and the understanding of the importance of the EHS process by the PIs is fundamental. I think that's what makes this thing work. Of course, there are many ways to make sure that the PIs actually understand the importance of this. But until they do, then this thing does not work. But as soon as the PIs understand the importance of this thing, then with all this infrastructure that we just discussed - you know, the CEHSs, - all these inspections, all these things - then everything comes into place. Because you could inspect all you want, but until the people that are responsible for it respond to the findings that you have, nothing is going to happen, right?

PI involvement has two important implications. First, at the cultural level, it shapes norms about safety within the lab. Second, at the practical level, it makes the system work efficiently. Practically, more involved PIs keep track of their students' training records, take concrete steps to fix problems, and engage in communication with coordinators. They are the ones who get the ship in order. At the cultural level, because PIs often serve, simultaneously, as mentor, coach, role model, and friend (Kram, 1988), their beliefs and practices about safety shape lab members' sensitivity toward compliance through indirect learning (Merton, 1957) and socialization in the lab (Huising & Silbey, 2013). In labs where PIs care about environmental sustainability, lab members are more aware of EHS rules and regulations. The first and foremost indicator of how serious the PI is about safety is whether he took his own training. Interviewees uniformly stated that if the PI himself is not fulfilling his obligation by taking the necessary training, it is impossible to go further. James, a coordinator with seven years of experience in the EHS field, mentioned how difficult it sometimes is to make PIs take their training: There's always PIs who are out of compliance with training (...) Some of the PIs, they're very difficult to get to - to go - do their training because I mean, you don't meet with them a lot of the times, and the rep [safety representative is a standard job in the lab, often assigned to a graduate student] is only meeting with them and saying, "OK, can you do the training?" And, you know, if he doesn't want to do it, it just doesn't get done. For the system to work effectively, PIs should be, and according to the coordinators and CEHS must be, on top of things. They are expected to be attentive to important outputs of the

124 understand the factors that lead to differential safety performance across labs, we turn to qualitative data consist of interviews with department based EHS coordinators and central administrative EHS staff at Eastern University. In the summer of 2007, 18 coordinators and 14 EHS staff were interviewed about the functioning of the EHS-MS. In these semi-structured interviews, interviewees were asked about their general impression about the workings of the system, their relationships with PIs and other lab members, and their opinions about the barriers to producing compliance in labs, as well as their insights about how the system could be improved. The interviews lasted an hour on average and were transcribed verbatim. The interviews with department coordinators and central EHS staff are especially useful for the purpose of this study because they provide valuable information on the working processes of the management system. Coordinators maintain compliance in laboratories by inspecting them semi-annually, regularly interacting with PIs, students, post-docs and other lab members such as technicians and lab managers to promote compliant practices by identifying appropriate materials and procedures while also correcting observed violations. Department level coordinators work at the interface between the system and the scientist, and are therefore in a privileged position of observing both internal dynamics of labs and bugs of the entire organizational management system. Central EHS staff (hereafter CEHS), on the other side, are in a position to observe the system across departments and see the larger organizational context. The organizational chart of the EHS-MS at Eastern is provided in Appendix C. The analysis of interviews involved several deep readings of the transcripts and focused on identification of major themes that explain variation in compliance across labs. Interestingly, non-human components of the management system -manuals, checklists, and digital tools including software and database- almost never came up in interviews as barriers to the success of the system. Rather, interviewees attributed success or failure of the system to human factors related to PI and lab group members (students, post-docs, technicians, and lab managers).

Principal Investigator's key role. Almost without exception, interviewees agreed that PIs' involvement with the system is the main factor that contributes to the success or failure in achieving compliance. Sam, who is a coordinator and has five years of prior EHS-related experience, summarizes the fundamental role of PIs in making the system work:

123 system such as training records of students and EHS inspection results. However, teaching and research work are experienced by both the PIs and the staff as impediments to PIs' consistent involvement. Lisa, a coordinator with five years of experience in the field of EHS, says that making just the PIs compliant, aside from the lab members, is a full-time job unto itself. She points out that, ironically, their core academic obligations become one of the factors that prevent PIs from fulfilling their safety obligations. They are experienced as trade-offs: These two PIs have their own labs in Building C, and because they supervise students, they need to take the training all their students have. But they themselves have not. And so last year, I sent and held the hand of one of the PIs, and went through every question in person, live, so that he could take the - the training. The other one finally took the training last year, but now is out of compliance since February, and I've been bugging him since February, but he said, "Well, see me at the end of the semester," because he's teaching classes. And I caught him at the end of the semester, and then I caught him again and he's still too busy.

Diane recognizes that even if some PIs care about maintaining safety in their labs, they may not be able to do so because of other priorities: "They want to do what's right, they want to do things safely, but it's one more thing they have to think about, and when they're doing research, it's not the priority." Thus, there appear to be two barriers to having greater PI involvement in the system: deliberate avoidance and what the faculty perceive and staff interpret as excessive overwork. According to interviewees, avoidance can stem from a range of factors from lack of confidence or respect for the system itself to the usual range of personality differences. Sometimes unique identities of PIs and diverse lab cultures can lead to isolation from the rest of the organization, which ultimately contributes to avoidance. Celine, who is a coordinator with three and half years of experience, points to this issue: They kind of view themselves as on their own and separate from the university. So some of these - the requirements of the management system -- they happily just brush off. And it's at all levels, you know, within the lab. I think the students are - would be happy to, you know, get involved, but because the PIs aren't really as supportive, it makes it difficult to kind of do their own thing.

125 The coordinators and CEHSs uniformly agreed that in order to produce compliance in labs, lab members should see safety part of their job. Echoing many of her peers, Alison says: "It's the same kind of - you know, people will have their specific jobs, and adding something on to that, you're not going to get their full attention to it unless they feel like it's part of their job."

Hierarchy and status. Despite the privileges of academic scientists, they are not entirely autonomous actors. They work within departments and schools within the university, where traditional ranks and hierarchy matter, albeit within a structure where extraordinary discretion still lies with the lowest level faculty. The presence of academic hierarchy along with EHS administrative hierarchy is an important aspect that makes system function (Huising & Silbey, 2013). Alison acknowledges this: Why didn't they - you know, why weren't the PIs - really getting something from the top down? (...) And it really should be coming from their bosses in conjunction with EHS saying, "This is what's going on; this is - you will do this. This is - as a PI, this is your responsibility." (...) they should be told, "This is the EHS program, the management program, and this is your responsibility.

Gary, a CEHS, notes that the academic hierarchy can be invoked to encourage PIs to complete safety training. Thus, it is not the system per se - that is, the available online training, or the data records - but the authority of the department chair that enables the system: Well, I think the drawback - like most departments from what I understand - is just getting PIs into the training system and taking the training. But what's good about the program is that, again, the department head directly contacts the PIs and tries to get them to take it, so.... And he even - he even does steps to try and force them to take it.

Echoing Gary, Tim says that department chairs sometimes threaten PIs with even more senior levels of hierarchy, such as deans: Actually, the people who don't do their training most frequently are the PIs. There's almost an annual communication with them saying, you know, "They're going to make

126 you meet with the dean soon because you're not doing this, so could you get it together and do it?" And they'll - you know, at that point, they say, "Yeah, OK - I'll do it by Wednesday." And they do it, you know? It's just - so you have to - they're not - they're not instantly responsive. One of the factors that lead to insulation of PIs is their perception about the status difference between themselves and coordinators. James reflects on his experience: I did a pre-inspection, and I went through and I visited the laboratory, and I wrote down whatever findings that I saw, and I sent an e-mail to the PI under my name. And because I'm not a faculty member, and I'm - I mean, it was, "Who is this guy sending me an e- mail saying that he found some issues in the laboratory?" And it really hit the fan. It is not rare for coordinators to encounter with PIs who take feedback on inspections as an attack on them personally as well as their status generally. As the quote from Mario below illustrates, in some cases, coordinators' work is undervalued and questioned: There was one instance in the past where I did an inspection of a faculty member's lab and he basically told me that, well he said, these are not issues in my lab. So either I had made them up or...something. So he sent me this pretty long e-mail that was a little bit vicious I would say.

Workload and competing priorities are also challenges for lab members in making sure that their daily practices comply with safety requirements. Student EHS representatives who conduct weekly lab inspections and interact with inspectors when needed sometimes do not have either motivation or time to fulfill their roles properly. Inspectors suggest ways to make student reps more engaged in the system. Gary notes: I know they're (students) doing their research projects and thesis and ... they're pulled a lot of different ways, and ... when more duties come up, it seems to fall on them. So maybe there's a way to ... give them some sort of - I'm not going to say "compensation," but something... that they'll ... some sort of reward that they would want to do it. And maybe it's just a function of some of them being appointed as opposed to volunteering. (...) I'm just going to say they're probably committed first... to their research, their projects. I'm not sure that being a rep is their number one commitment.

127 Interviewees also mentioned that labs with a high rate of turnover tend to have higher rates of violation. Especially labs where many short-term, undergraduate research assistants are employed turn out to be problematic. Jack: Most people don't want to get hurt and are not looking to skirt the rules, but they are - they have jobs to do other than that, so.... So sometimes it's hard to get - and - and the students are probably the worst villains in that sense, (...) students come and go on a regular basis, and you - you know, I've - I know that there's some of them down there that I've never seen. They've come and gone before I even been there to see them. So there's no way for me to manage that from this ... situation.

4.6. Discussion Universities can be very risky environments. Regulators and university administrators hope that conducting inspections and taking remedial steps based on inspection findings may help prevent tragic outcomes. However, the findings of this study show that despite regular audits and feedback to PIs and lab members, the number of safety violations increased over time. A closer examination of findings revealed differential reaction of PIs to audits and a widening of the gap between compliant and non-compliant researchers. The quantitative analyses indicate that the magnitude of funding, and thus material resources available to a PI - a signal of economic resourcefulness of the academic lab is an important predictor of violations. In addition, tenure, a formal marker of status in academia, also significantly predicted the number of violations. Tenured scientists violated EHS regulations more frequently more than their non- tenured peers, and scientists with more abundant research funds likewise violated EHS regulations at higher rates relative to those with less funding. Interviews with central system actors, inspectors and CEHS personnel provided important insights about the obstacles to the system's success. Interviewees see PIs' involvement crucial to make the system work. They point out two factors for PIs' lax involvement: avoidance and workload. While some scientists simply do not care about making the systems effective, others, who might care in principle, devote their time to other activities - research, and raising research funds. We can document the higher levels of publication, but even those measures fall away when we look at funding levels. Even though Eastern University is a "true believer" in

128 environmental sustainability and safety, status as a source of power is ultimately accorded to those researchers -scientists and engineers - who garner the most external recognition through research funding, as well as publications. We note that university hierarchy, when invoked, can influence some recalcitrant actors, but apparently not the most non-compliant who are also the most well-funded laboratories. Thus, some status related factors create challenges for achieving compliance in its research labs. Status differences can create problems when individuals or teams from different occupations work together (Dibenigno & Kellog, 2014; Huising & Silbey, 2011; Ranganathan, 2013). This difficulty is intensified when the task is knowledge related (Bechky, 2003; Bailey, Leonardi, & Chong, 2010; Carlile, 2004; Huising & Silbey, 2013). In the case of Eastern University, in interviewees' accounts we again see indicators of scientists' resistance to the system (ledema et al., 2004; Waring, 2005) and faculty's privilege to ignore legal procedures (Huising & Silbey, 2013). Yet, the intensification of violations over time by high violator PIs' is an important finding that needs further consideration. First, we know that in the absence of external pressure, self-regulatory systems must draw upon a reservoir of informal, mimetic and local normative means to secure compliance (Baumeister & Heatherton, 1996). Yet, Weick (1976) suggests that, for loosely coupled systems, it is difficult to repair the defective element due to logical and physical separateness of elements throughout the organization. Perrow (2011) suggests that such loosely coupled systems might avoid tragic accidents by redundancies as well as the very unlinked processes. Here it is important to stress that Eastern University has, thus far, had a very good record with regard to environmental spills and emissions, as well as serious accidents. How can this be explained? Huising and Silbey (2011) suggest that while the coordinators regularly inspect and record the violations, they also work - off the books one might say - to correct problems when they see them. Organizational scholars have been interested in compliance with legal regulations for decades. Among many others, one line of analysis has explored whether and why groups or organizations react differently to regulations. Students of organizations have studied uneven compliance with regulations at different levels of analysis. For example, earlier generation of research in this area has studied variation in terms of uneven compliance across organizations within one institutional context (e.g., Edelman & Suchman, 1997; Kellogg, 2009). Later

129 generations studied variation across actor groups within the same organization (e.g., Gray & Silbey, 2014; Heimer & Staffen, 1998). However, up to date, the question of how different individuals within the same actor group in the same organization react to a constant regulatory environment has not yet been studied. This study pulled the unit of analysis one level down and studied individual differences in reaction to EHS audits in an elite group, namely scientists in a large research university. With specific attention to regulatory compliance, the findings are in line with previous research that showed that actor's variable autonomy, expertise, and frequency of interaction influence interpretations of regulations and regulators and orientations toward compliance (Gray & Silbey, 2014). These extensive observations of differential compliance with EHS rules and regulations in one research university force us to rethink the models of responsive regulation that emphasize the importance of tuning regulatory processes to the differing motivations of regulated actors (Ayres & Braithwaite, 1992). Since academic settings with their unique characteristics such as autonomy of scientists and academic freedom might create room for non- compliance for actors, enacting an adapted model responsive regulation might be one of the effective strategies to promote compliance in research universities.

130 4.7. References Ayres, I., & Braithwaite, J. (1992). Responsive regulation:Transcending the deregulation debate. New York: Oxford University Press. Bailey, D. E., Leonardi, P. M., & Chong, J. (2010). Minding the gaps: Understanding technology interdependence and coordination in knowledge work. OrganizationScience, 21(3), 713- 730. Bechky, B. A. (2003). Sharing meaning across occupational communities: The transformation of understanding on a production floor. OrganizationScience, 14(3), 312-330. Baumeister, R. F., & Heatherton, T. F. (1996). Self-regulation failure: An overview. PsychologicalInquiry, 7, 1-15. Carlile, P. R. (2002). A pragmatic view of knowledge and boundaries: Boundary objects in new product development. OrganizationScience, 13(4), 442-455. Carlile, P. R. (2004). Transferring, translating, and transforming: An integrative framework for managing knowledge across boundaries. OrganizationScience, 15(5), 555-568. Cetina, K. K. (2009). Epistemic cultures: How the sciences make knowledge. Cambridge, MA: Harvard University Press. Coglianese, C. (2016). Performance-based regulation: concepts and challenges. In F. Bignami & D. Zaring (Eds.), Comparative Law and Regulation: Understandingthe Global Regulatory Process, 403-430. Cheltenham: Edward Elgar. Coglianese, C., & Lazer, D. (2003). Management-based regulation: Prescribing private management to achieve public goals. Law & Society Review, 37(4), 691-730. Coglianese, C., & Mendelson, E. (2010). Meta-regulation and self-regulation. In R. Baldwin, M. Cave, & M. Lodge (Eds.), The Oxford Handbook of Regulation, 146-168. New York: Oxford University Press. Coglianese, C., & Nash, J. (2001). Regulating from the inside: Can environmental management systems achieve policy goals? Washington, DC: Resources for the Future Press. Darnall, N., & Edwards, D. (2006). Predicting the cost of environmental management system adoption: The role of capabilities, resources and ownership structure. Strategic ManagementJournal, 27(4), 301-320. Delmas, M. A., & Toffel, M. W. (2008). Organizational responses to environmental demands: Opening the black box. Strategic ManagementJournal, 29(10), 1027-1055. DiBenigno, J., & Kellogg, K. C. (2014). Administrative Science. Administrative Science Quarterly,59(3), 375-408. DiMaggio, P. J., & Powell, W. W. (1991). The new institutionalismin organizational analysis. Chicago, IL: University of Chicago Press. Edelman, L. B., & Suchman, M. C. (1997). The legal environments of organizations. Annual Review of Sociology, 23(1), 479-515. Esbenshade, J. L. (2004). Monitoring Sweatshops: Workers, Consumers, and the GlobalApparel Industry, Philadelphia, PA: Temple University Press. Evans, J. (2014). Resisting or Governing Risk? Professional Struggles and the Regulation of Safe Science. Academy of Management Proceedings, Vol. 2014, No. 1. Ewick, P. & Silbey, S. (1998). The common place of law: Stories ofpopular legal consciousness. Chicago: University of Chicago. Gray, G., & Silbey, S. (2011). The other side of the compliance relationship. In C. Parker & V. Nielson (Eds.), Explaining Compliance: Understanding and Explaining Organizational Responses to Regulation, 123-138. Northampton, MA: Edward Elgar Publishing.

131 Gray, G. C., & Silbey, S. S. (2014). Governing inside the organization: Interpreting regulation and compliance. American Journalof Sociology, 120(1), 96-145. Gunningham, N. (1995). Environment, self-regulation, and the chemical industry: Assessing responsible care. Law & Policy, 17(1), 57-109. Gunningham, N. (1999). Integrating management systems and occupational health and safety regulation. Journalof Law and Society, 26(2), 192-214. Gunningham, N., Kagan, R. A., & Thornton, D. (2003). Shades of green: Business, regulation, and environment. Stanford: Stanford University Press. Harzing, A.W. (2007). Publish or Perish, available from http://www.harzing.com/pop.htm Heimer, C. A., & Staffen, L. R. (1998). For the sake of the children: The social organization of responsibility in the hospital and the home. Chicago, IL: University of Chicago Press. Hirsch, J. E. (2005). An index to quantify an individual's scientific research output. Proceedings of the Nationalacademy of Sciences of the United States ofAmerica, 16569-16572. Howard-Grenville, J., Nash, J., & Coglianese, C. (2008). Constructing the license to operate: Internal factors and their influence on corporate environmental decisions. Law & Policy, 30(1), 73-107. Huising, R., & Silbey, S. S. (2011). Governing the gap: Forging safe science through relational regulation. Regulation & Governance, 5(1), 14-42. Huising, R., & Silbey, S. S. (2013). Constructing consequences for noncompliance: The case of academic laboratories. The ANNALS of the American Academy of Politicaland Social Science, 649(1), 157-177. ledema, R., Degeling, J., Braithwaite, L., & White, L. (2004). It's an interesting conversation I'm hearing: The doctor as manager. OrganizationalStudies 25, 15-34. Jenkins, R. (2001). Corporate codes of conduct: Self-regulation in a global economy. United Nations Research Institute for Social Development, Geneva. Kagan, R. A., Gunningham, N., & Thornton, D. (2003). Explaining corporate environmental performance: How does regulation matter? Law & Society Review, 37(1), 51-90. Kellogg, K. C. (2009). Operating room: Relational spaces and microinstitutional change in surgery. American Journal of Sociology, 115(3), 657-711. Khanna, M., & Damon, L. A. (1999). EPA's voluntary 33/50 program: Impact on toxic releases and economic performance of firms. Journal of environmental economics and management, 37(1), 1-25. King, A. A., & Lenox, M. J. (2000). Industry self-regulation without sanctions: The chemical industry's responsible care program. Academy of ManagementJournal, 43(4), 698-716. Kram, K. E. (1988). Mentoring at Work: Developmental Relationships in OrganizationalLife. University Press of America, Lanham, MD. Locke, R. M., & Romis, M. (2010). The promise and perils of private voluntary regulation: Labor standards and work organization in two Mexican garment factories. Review of InternationalPolitical Economy, 17(1), 45-74. Locke, R. M., Qin, F., & Brause, A. (2007). Does monitoring improve labor standards? Lessons from Nike. ILR Review, 61(1), 3-31. Luria, G. (2015). Safety Climate and Supervisory-Based Interventions. In S. Clarke, T. M. Probst, F. Guldenmund & J. Passmore (Eds.), The Wiley Blackwell Handbook of the Psychology of OccupationalSafety and Workplace Health, 357-377. Chichester, UK: John Wiley & Sons. Merton, R. K. (1957). Some preliminaries to a sociology of medical education. In R. K. Merton,

132 G. G. Reader, & P. L. Kendall (Eds.), The Student Physician, 3-80. Cambridge: Harvard University Press. National Research Council. (2014). Safe Science: Promoting a Culture of Safety in Academic Chemical Research. National Academies Press. Nash, J., & Ehrenfeld, J. (1997). Codes of environmental management practice: Assessing their potential as a tool for change. In R. H. Socolow (Ed.), Annual Review of Energy and Environment, 487-535. Palo Alto, CA: Annual Reviews. O'Rourke, D. (2003). Outsourcing regulation: Analyzing nongovernmental systems of labor standards and monitoring, Policy Studies Journal, 31(1), 1-30. Perrow, C. (2011). Normal accidents: Living with high risk technologies. Princeton University Press. Ranganathan, A. (2013). Professionalization and market closure: The case of plumbing in India. ILR Review, 66(4), 902-932. Rees, J. (1997). The development of communitarian regulation in the chemical industry. Law and Policy, 19, 477. Schneiberg, M., & Bartley, T. (2008). Organizations, regulation, and economic behavior: Regulatory dynamics and forms from the nineteenth to twenty-first century. Annual Review of Law and Social Science, 4, 31-61. Short, J. L., & Toffel, M. W. (2010). Making self-regulation more than merely symbolic: The critical role of the legal environment. Administrative Science Quarterly, 55(3), 361-396. Silbey, S. S. (2009). Taming Prometheus: Talk about safety and culture. Annual Review of Sociology, 35, 341-369. Silbey, S. S. (2013). Organizational challenges to regulatory enforcement and compliance: A new common sense about regulation. The ANNALS of the American Academy of Political and Social Science, 649(1), 6-20. Silbey, S. S., & Ewick, P. (2003). The architecture of authority: The place of law in the space of science. In Sarat, A., Douglas, L., & Umphrey, M. (Eds.), The Place ofLaw, 76-108. Ann Arbor: University of Michigan Press. Vidovic, M., & Khanna, N. (2007). Can voluntary pollution prevention programs fulfill their promises? Further evidence from the EPA's 33/50 Program. Journalof Environmental Economics and Management, 53(2), 180-195. Waring, J. (2005). Beyond blame: Cultural barriers to medical incident reporting. Social Science Medicine, 60, 1927-1935. Weber, M. (1947). The theory of social and economic organization. New York: Oxford University Press. Weick, K. E. (1976). Educational organizations as loosely coupled systems. Administrative Science Quarterly, 21(1), 1-19. Whitley, R. (2000). The intellectual and social organization of the sciences. Oxford: Clarendon Press. Wrong, D. H. (1980). Power: Its forms, bases, and uses. New Brunswick, N.J.: Transaction.

133 4.8. Tables

Table 1. Total number of findings by department

Department Total Chemical Engineering 3162 (33.8%) Chemistry 1160 (12.4%) Biological Engineering 1201 (12.8%) Oncology 514 (5.5%) Biology 513 (5.4%) Neuroscience 921 (9.8%) Mechanical Engineering 705 (7.5%) Materials Engineering 652 (6.9%) Electrical Engineering 512 (5.4%) Total 9340 (100.0%)

134 Table 2. Descriptive and Shapiro-Wilk statistics of violations by department in a given year Department Mean Median Std. Deviation Min Max Shapiro-Wilk test Chemical Engineering 91.35 68.00 72.309 5 257 .861** Chemistry 34.46 35.50 25.295 1 98 .942 Bioengineering 30.64 29.00 17.879 5 66 .937 Oncology 4.92 2.00 7.138 1 31 .554*** Biology 15.11 12.00 15.428 1 84 .621*** Neuroscience 30.10 23.00 28.860 1 100 .851** Mechanical Engineering 17.74 10.00 18.063 2 70 .787*** Materials Science & Engineering 14.03 13.50 10.041 1 47 .916* Electrical Engineering 19.12 17.00 14.081 3 58 .894*

135 Table 3. Descriptives Mean SD

Age 55 13.86

Female 0.20 0.40

White 0.77 0.41

Tenure 0.80 0.39

US origin 0.62 0.48

Duration at Eastern 21.73 13.07

PhD institute rank 30.25 62.14

Funding 1,168,633 1,302,085

Lab size 18.03 22.88

Number of pubs. 8.27 7.42

136 Table 4. Correlations among variables 1 2 3 4 5 6 7 8 9 1.Age 2.Female -0.12** 3.White 0.29*** -0.06* 4.US origin 0.08* 0.02 0.21*** 5.Tenured 0.46*** -0.13*** 0.20*** -0.04 6.Duration 0.90*** -0.14*** 0.25*** 0.06* 0.46*** 7.PhD rank 0.11** 0.01 0.00 -0.23*** 0.01 0.05 8.Log-budget 0.14** -0.10* 0.17*** 0.07 0.41*** 0.11* -0.00 9.# of pubs. 0.01 -0.02 -0.00 -0.01 0.24*** 0.02 -0.00 0.39*** 10.Lab size -0.04 0.01 0.01 0.04 0.05 -0.01 -0.02 0.29*** 0.42*** *p<.05; **p<.Ol; ***p

137 Table 5. Random Effects Model (1) (2) (3) (4) (5)

Age -.003 -.08 -.04 -.03 -.03 (.028) (.068) (.073) (.073) (.089) Female 1.22 1.53 1.61 1.52 1.48 (.916) (.934) (.969) (.968) (1.07) White .80 .53 .95 1.01 .79 (.962) (.985) (1.02) (1.02) (1.11) Country origin (US) -.68 -.22 -.10 -0.15 .07 (.797) (.839) (.860) (.859) (.969)

Tenured 3.09* 1.14 .64 -.79 (1.268) (1.86) (1.89) (2.48) Duration at Eastern .05 .020 .02 .01 (.068) (.073) (.073) (.086) PhD institute rank .00 .00 .00 .00 (.006) (.00) (.006) (.007)

Research budget .76** .61 * .62* (.268) (.283) (.313) Number of pubs. .08 .07 (.051) (.06) Lab size .00 (.021)

Department dummies + + + + +

Year dummies

R2 0.26 0.27 0.33 0.34 0.35

N of observations 1062 1057 570 568 509

N of groups 240 236 175 173 153

*p<.05 ; **p<.01 Standard errors are in parentheses.

138

+ + + + + Table 6. Fixed Effects Model: (DV: Number of Violations) (1) (2) Research budget .80P .820 (.435) (.439) Number of pubs. -.01 (.079)

Year dummies + +

N of obs. 577 572

N of groups 181 176 Pp< 0 .1 Standard errors are in parentheses.

139 Table 7. Fixed Effects Model: (DV: Number of Violations) (1) (2)

Budget percentile 9 5 th 4.65* 4.74* (1.91) (1.93)

Budget percentile 9 0th 3.94* 4.00* (1.82) (1.83)

Budget percentile 7 5th 3.13* 3.17* (1.23) (1.24)

Budget percentile 5 0th 1.35 1.38 (.928) (.934) Number of pubs. -.02 (.079)

Year dummies + +

N of obs. 576 571

N of groups 180 175 *p<. 0 5 Standard errors are in parentheses. Below median is the omitted category.

140 4.9. Figures

Figure 1. Total inspection findings per year in Eastern University 3500 2911 3000

2500

2000 1708 1500 1398 1.37

1000 490 500

0 Y2006 Y2007 Y2008 Y2009 Y2010 Y2011 Y2012

141 Figure 2. Number of findings by percentile over time 37.30 40.00 35.00 30.0023.30 25.00 .20.65 20.00 20.00 16.004. > 15 00 - 12.00 12.30 10.00 .00 5.25

0.0() Y2006 Y2007 Y2008 Y2009 Y2010 Y2011 Y2012 Year

- 40th percentile Median - 75th percentile - 90th percentile 95th percentile

142 Figure 3. Number of Findings over Time, grouped by 2006 scores 34.13

30

25

20 19.1 15.07 15 3

10 677.07

Year2007 Year2008 Year2009 Year201(0 Year2011 Year2012 Years

- 50th2006 - 75th2006 - 9th2006 - 95th2006

143 4.10. Appendix Appendix A. Type and percentage of findings recorded Finding Percent Sample Finding Category Housekeeping 24.6 The appearance of the laboratory/shop was not neat, orderly and clean. SAA 17.3 Non-waste materials were kept in an SAA (satellite accumulation area for chemicals). Chemical 15.1 Chemical containers were not properly labeled. storage Doc/label 12.9 Current emergency response information was not posted in all required areas. Equipment 8.3 Area was not equipped with sufficient drench showers and eyewash stations. PPE 5.2 Laboratory personnel were not wearing eye protection, lab coats, gloves and other appropriate PPE. Biowaste 3.8 Biowaste was not being appropriately managed. Training 3.6 Required EHS training was not up to date. Hoods 3.5 Fume hood housekeeping was poor and/or had excessive clutter. Biosafety 2.8 Needles and/or syringe stocks were not secured. Radiation 1.5 Radioactive Material Inventory and/or Use Logs were not up to date. Spills .9 There was evidence of spills not properly cleaned up.

144 Appendix B. The source of research funding Source Percentage Dept of Health & Human Services 35% Dept of Defense 18% Dept of Energy 13% Industry 12% National Science Foundation 9% Foundations and Other Non-Profit 4% State, Local & Foreign Governments 3% Internal 2% NASA 2% Other Federal G vernment 1%

145 Appendix C

Organizational Structure

Pvr...... o.s.

ticoic * ImdirrtuI ygiso

[StueanI OrFPE? Do4

146