Television Advertising and Online Search

Home , Television advertisement

This article was downloaded by: [128.54.18.48] On: 26 May 2016, At: 14:01 Publisher: Institute for Operations Research and the Management Sciences (INFORMS) INFORMS is located in Maryland, USA

Management Science

Publication details, including instructions for authors and subscription information: http://pubsonline.informs.org Television Advertising and Online Search

Mingyu Joo, Kenneth C. Wilbur, Bo Cowgill, Yi Zhu

To cite this article: Mingyu Joo, Kenneth C. Wilbur, Bo Cowgill, Yi Zhu (2014) Television Advertising and Online Search. Management Science 60(1):56-73. http://dx.doi.org/10.1287/mnsc.2013.1741

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher approval, unless otherwise noted. For more information, contact [email protected]. The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or support of claims made of that product, publication, or service. Copyright © 2014, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, management science, and analytics. For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org MANAGEMENT SCIENCE Vol. 60, No. 1, January 2014, pp. 56–73 ISSN 0025-1909 (print) ISSN 1526-5501 (online) http://dx.doi.org/10.1287/mnsc.2013.1741 © 2014 INFORMS

Television Advertising and Online Search

Mingyu Joo Fisher College of Business, The Ohio State University, Columbus, Ohio 43210, [email protected] Kenneth C. Wilbur Rady School of Management, University of California, San Diego, La Jolla, California 92093, [email protected] Bo Cowgill Haas School of Business, University of California, Berkeley, Berkeley, California 94720, [email protected] Yi Zhu Carlson School of Management, University of Minnesota, Minneapolis, Minnesota 55455, [email protected]

espite a 20-year trend toward integrated marketing communications, advertisers seldom coordinate tele- Dvision and search advertising campaigns. We ﬁnd that television advertising for ﬁnancial services brands increases both the number of related Google searches and searchers’ tendency to use branded keywords in place of generic keywords. The elasticity of a brand’s total searches with respect to its TV advertising is 0.17, an effect that peaks in the morning. These results suggest that practitioners should account for cross-media effects when planning, executing, and evaluating both television and search advertising campaigns. Key words: advertising; information search; media; search engine marketing; television History: Received February 17, 2012; accepted February 23, 2013, by Pradeep Chintagunta, marketing. Published online in Articles in Advance May 23, 2013.

1. Introduction United States (Internet Advertising Bureau 2012, Tele- It is difficult to overstate the importance of television. vision Bureau of Advertising 2012). The average American household contains 2.6 peo- Two decades of research and practice on integrated ple and 2.5 televisions (Nielsen 2011, U.S. Census marketing communications have shown that deliver- Bureau 2011). Television remains the most trusted ing a consistent message through multiple consumer source of news and information by a wide margin touch points is more effective than managing dis- (Danaher and Rossiter 2011, eMarketer 2012). The parate, medium-specific campaigns. One might expect average American watches 5.1 hours of television that advertisers would coordinate their television per day, more than the 4.6 self-reported daily hours advertising and search advertising campaigns. After spent on all work-related, educational, and house- all, synchronizing “push” and “pull” tactics is an old work activities (Nielsen 2011, Bureau of Labor Statis- topic in marketing. tics 2011). The typical consumer is exposed to about Despite these apparent incentives, a review of 29 minutes of paid television advertising per day.1 Advertising Age’s top traditional and online advertising agencies (AdAge 2011) showed that coordina- Also remarkable is the growth of online search. tion of advertising campaigns across television and Americans used core search engines like Google and search media remains unusual. Client-facing agency Bing 17.7 billion times in July 2012, or about 1.8 times websites were examined to determine each agency’s per person per day (comScore 2012). In other words, services and areas of expertise. Table 1 presents some the average consumer now practices, about twice per surprising results from this survey. Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. day, an activity that barely existed 15 years ago. Per- Twenty-four of the top 25 online advertising haps the best evidence that search converts prospects agencies do not offer television advertising services to customers is provided by changes in advertis- in-house. Twenty-four of the top 25 traditional adver- ing budgets. Marketers spent $14.8 billion on search tising agencies do not offer search advertising services advertising in 2011, a substantial amount relative to in-house. Only three of these 50 agencies coordinate the $69.6 billion spent on television advertising in the online/offline advertising campaigns, and they define coordination as media budget planning or show- 1 This inference assumes 11.3 minutes of advertising per hour, as is ing Web addresses in TV ads. The scarcity of coor- typical in prime time, and an advertising avoidance rate of 50%. dinated television/search campaigns is further con-

Table 1 Top Advertising Agencies’ Service Offerings 37% of Internet users reported that a television ad had prompted them to conduct an Internet search. How- No. of top 25 No. of top 25 search traditional ever, work of this kind is typically based on searchers’ advertising advertising self-reports, includes limited controls for competing agencies agencies explanations and often focuses on a single brand. The one extant study based on passively measured In-house services for both TV/search advertising search data is that by Zigmond and Stipp (2010). They No 24 24 presented several case studies that showed peaks Yes 1 (PlattForm) 1 (Ogilvy & Mather) in Google searches corresponding to several TV ads Online/offline coordination that aired during the Vancouver Winter Olympics. No coordination 23 24 We extend this research by examining a mature prod- Provide Web address in offline 1 (iProspect) 0 uct category over a three-month period, estimating creative Provide media budget planning 1 (Covario) 0 both short- and longer-run effects of advertising, and across media by separately considering how TV advertising may Coordination offered but no 0 1 (Leo Burnett) affect both the volume of category search and the details provided probability of choosing a branded keyword. Online/offline integrated The analysis indicates that TV advertising increases performance metrics the number of product category-relevant searches and None offered 21 21 increases consumers’ tendency to use branded key- Media budget allocation only 2 (Acronym, 2 (Doner, Martin) Rosetta) words. The elasticity of a financial services brand’s ROI across all media 1 (Covario) 2 (Campbell Ewald, online searches with respect to its advertising is 0.17, Leo Burnett) an effect that is largest in the morning and smallest in Metrics for effects of TV to 1 (iProspect) 0 the late afternoon. These findings suggest that brands online might profit from coordinating their television and search advertising campaigns.

firmed by the topic’s absence in leading advertising 1.1. Relationship to Prior Literature textbooks and academic literature and by personal A burgeoning literature is finding that advertising in conversations with dozens of researchers and man- one medium may influence the results of advertising agers at Amazon, Google, Yahoo! and other firms. As in another medium (Assael 2011). Naik and Raman Enge (2012) explained, “there is a big gap between (2003) and Vakratsas and Ma (2005) gave evidence of traditional marketing thinking and the way search advertising synergies among multiple offline media, marketers typically think. For [the search marketing] and Naik and Peters (2009) found synergies between industry to reach full maturity, that gap needs to spending in online and offline media. Lewis and close, and there needs to be movement on both sides.” Reiley (2011) found via a large-scale controlled exper- The purpose of this paper is to investigate whether iment that 93% of the effects of online display adver- and how television advertising expenditures influ- tising were found in offline retailer sales. Goldfarb and ence online search behavior. The empirical analysis Tucker (2011) found a substitution pattern between combines two comprehensive databases of television online advertising and offline advertising: advertisers advertising and consumer search in the financial ser- pay more for online search keywords when offline vices product category. The Internet search data are advertising is prohibited. Lewis and Nguyen (2012) provided by Google and include well over one billion and Papadimitriou et al. (2011) ran field experi- searches for financial services keywords. The advertis- ments and found that display advertising increased ing data record the exact times and estimated expen- searches for both the advertised brand and its com- diture on 58,226 television advertisements aired by petitors. Rutz and Bucklin (2012) found a similar 15 financial services brands at a total cost of about effect: users exposed to a display advertisement on a $200 million. The effects of television advertising product search engine were subsequently more likely on online search are identified by changes in the to browse pages related to the advertised brand. The Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. hourly time series of search behavior corresponding current paper buttresses the existing media syner- to brands’ placements of television advertisements. gies literature by showing specifically how advertis- The effects of primary interest are highly robust and ing in one medium can change consumer behavior in can be reproduced through several different estima- another. tion techniques. A small number of recent papers have predicted Although the academic literature has not explored ways in which cross-media advertising effects might the effects of television advertising on online search, influence market competition. Kim and Balachander a few practitioner studies have considered it. (2010) showed how an advertiser’s cost per con- iProspect (2007) surveyed consumers and found that sumer in traditional media influences its optimal bid Joo et al.: Television Advertising and Online Search 58 Management Science 60(1), pp. 56–73, © 2014 INFORMS

on search results. They found that advertisers have monthly. Variation in advertising expenditure over incentive to coordinate search advertising with tra- time is critical to identify the effect of advertising on ditional advertising, even when doing so is costly. search behavior. Bergemann and Bonatti (2011) investigated the role 3. Category brand names should not overlap too much of targeting in the competition between offline and with commonly searched keywords. Otherwise, category online media. They predicted that the entry of highly searches cannot be separated from unrelated searches. targeted media (online) will increase the price of less For example, it would not be clear whether a search targeted media (offline) and reduce their revenues. for “Apple” is for a fruit or a computer. Many well- The possibility that advertisers may free ride on com- known brands—like Miller, Target, and Visa—present petitors’ offline advertising was explored by Sayedi this problem. et al. (2011). They showed that symmetric firms may 4. The category should not be subject to obvious simul- develop asymmetric strategies, with one firm invest- taneity concerns. For example, advertising and con- ing in offline advertising to build category demand sumer search for a movie both peak around the date while the other firm uses targeted search advertising the movie is released in theaters. It would be diffi- to free ride on competitors’ investments. Our findings cult to tease apart the effect of advertising from the help to explain how online and offline media interact effect of the movie release date or other contempo- with each other to generate such effects in a market. raneous promotions without good instrumental vari- These effects are particularly relevant to the liter- ables. Ideally, the researcher should be able to get ature on search engine marketing (Jerath and Sayedi data on exogenous time-varying factors that may shift 2011, Jerath et al. 2011, Katona and Sarvary 2010, Rutz searchers’ tendency to search for brands in the cat- and Bucklin 2011, Rutz and Trusov 2011, Wilbur and egory to separately identify the effect of television Zhu 2009, Yang and Ghose 2010, Yao and Mela 2011, advertising from other factors that vary over time. Zhu and Wilbur 2010). As competition in the search The financial services product category scores well engine setting becomes better understood, a natural on all four criteria. It is a high-involvement cate- way to extend this literature is to consider how search gory with infrequent choice, as consumers tend to marketers may use traditional “push” media such as stay with investment brokerages for long periods of television to compete before consumers even arrive at time. It is the seventhmost advertised product cat- the search engine. egory on television. Most major financial services brand names, such as Schwab and Ameritrade, gen- 2. Data and Measures erally do not overlap with commonly searched key- This section explains why we chose to study the words, as shown in Table 2. The data do not suggest 2 financial services category, introduces the measures simultaneity, as §4 below explores in depth. and data, and describes how their characteristics 2.2. Dependent Variables: Consumer influence the modeling choices presented below. Search Behavior 2.1. Research Context: Product Category Choice Marketing academics have studied consumer search A product category suitable for determining how extensively since the 1970s. The literature implies offline advertising influences online search should three primary reasons that television advertising is exhibit four characteristics. The first criterion relates likely to influence online search: objective knowledge, to whether effects of advertising on search exist, perceived knowledge, and incidental exposure. whereas the other three relate to an analyst’s ability First, television advertising may increase con- to detect those effects in market data. sumers’ objective knowledge of product or category 1. Consumers must search category brand names online. features and benefits, and this may influence con- High-involvement categories may be most appropri- sumer search. For example, Brucks (1985) found that ate since consumers are likely to actively gather infor- some objective knowledge increases consumers’ abil- mation related to brands and products within those ity to acquire additional knowledge and makes search categories. Categories with infrequent choices or high more efficient. Several studies have related these Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. prices might be most appropriate, since consumers effects directly to objective knowledge provided by cannot gather information easily through product trial. 2 Several other product categories were considered. The automotive 2. Category brands’ offline advertising must be mea- category scores well on all criteria except the third. The pharmaceu- sured with high frequency. Television advertising expen- tical category scores well on all criteria except the first, since most ditures may be observed by day and precise time, consumers (but certainly not all) get information about most drugs from their doctors rather than from search engines. For example, as are online search data. Advertising expenditure the keyword “fidelity” is searched about 60 times as frequently as data for other offline media such as radio, magazines, the keyword “Vioxx.” Movies and video games score well on all and billboards are typically only observed to vary criteria except the fourth. Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 59

Table 2 Most Common Financial Services-Related Keywords

Most common generic keywords Most common branded keywords

SEP INVESTORS AGEDWARDS ML 401 K IRA AMERITRADE MORGANSTANLEY ANNUITY LIABILITY AMERIVEST NETBENEFITS BALANCE LIQUIDITY CITI NUVEEN BENEFITS LOAN CITIBANK OPPENHEIMER BONDS LOANS CITIFINANCIAL OPPENHEIMERFUNDS BROKER MUTUAL CITIGROUP OPTIONSXPRESS BROKERS OPTIONS CITIMORTGAGE RAYMONDJAMES CHECKING PAY CLEARSTATION SCHWAB CONTRIBUTION PAYMENT CYBERTRADER SCOTTRADE CREDIT PREPAID EBTACCOUNT SHAREBUILDER DEBIT PROFIT EDWARDJONES SOUTHTRUST DIVIDEND RETIREMENT ETRADE TD DIVIDENDS SECURED EWORKPLACE TDWATERHOUSE EBT SECURITIES FIDELITY TROWE EQUITY SHARES FIDELITYINVESTMENTS VANGUARD ETF SHARING FOREX VANKAMPEN FIXED STATEMENT FXCM WACHOVIA FUND STOCK LEGGMASON WATERHOUSE FUNDS STOCKS INCOME TRADE INVEST TRADING INVESTING TRUSTS INVESTMENT UGMA INVESTMENTS YIELD

advertising. Newman and Staelin (1973) showed that found that incidental exposure enhances brand lik- advertising may enlarge the set of brands a consumer ing. Shapiro et al. (1997) found that incidental expo- can recall easily. Bettman and Park (1980) found that sure to advertising influenced the products that consumers with moderate amounts of prior product enter consumers’ consideration sets, even when the information are more likely to search for a brand than consumers are not consciously aware that they saw those with little prior information. Therefore, televi- the ads. Shapiro (1999) took this a step further, show- sion advertising might stimulate online search if it ing that incidental ad exposure led to consideration increases consumers’ stock of objective knowledge. set inclusion, even when subjects were explicitly Second, advertising may alter how much a con- instructed to avoid choosing products depicted in ads. sumer thinks she knows (“perceived knowledge”), Therefore it is possible that incidental exposure to which may influence how the consumer searches. television advertising changes the keywords a con- Numerous studies have shown perceived knowledge sumer would use to search. to differ from objective knowledge with correlations This literature indicates that information can affect ranging from 0.05 (Radecki and Jaccard 1995) to 0.65 both the likelihood of search and the means of search. (Park et al. 1994). Moorman et al. (2004) experi- We therefore distinguish between category search and mentally manipulated perceived knowledge by first keyword choice. Category search is defined as the num- testing subjects’ knowledge in a particular domain, ber of searches in a period that contained generic then randomly giving artificially low test results to or branded financial services-related keywords. Key- some knowledgeable subjects and artificially high test results to some unknowledgeable subjects. Inflated word choice is defined as the fraction of all key- test scores (high perceived knowledge) led to search words entered that are related to a particular brand strategies that were less likely to uncover discon- in the category. In earlier research, a data mining firming information. Therefore, if branded television technique was developed to identify a set of generic Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. advertising increases consumers’ perceived knowl- and branded keywords and determine which searches edge, it may increase the chance the consumer enters were relevant to the financial services product cate- branded keywords into a search engine. gory; for full details, see Joo et al. (2013). This proce- Third, incidental exposure to advertising has been dure identified several previously unknown branded found to influence consumers outside their conscious keywords and a large set of generic keywords that awareness. Incidental exposure refers to advertising were frequently used in searches that led to clicks on that is perceived but not processed. It commonly financial services brands’ websites. Table 2 provides occurs when consumers redirect their attention dur- the common branded and generic keywords in the ing television commercial breaks. Janiszewski (1993) data set. Joo et al.: Television Advertising and Online Search 60 Management Science 60(1), pp. 56–73, © 2014 INFORMS

This distinction between category search and 2.4. Control Variables: Time Effects and keyword choice has important implications for Stock Market Indices consumers as well as marketers. People who search The empirical model relies on brand dummies, time a branded keyword receive less information about fixed effects, and movements in a stock market index competitors than those who search generic keywords. to estimate baseline consumer tendencies to search. Paid advertising clicks also tend to cost less when Advertising effects on search behavior are then iden- consumers search branded keywords, because these tified by deviations from this baseline corresponding keywords’ auctions typically enroll fewer bidders to brands’ TV advertising expenditures. than generic keyword auctions. It also may be the case The time controls consist of fixed effects for each that a consumer who searches a branded keyword has week in the sample, to allow baseline category search revealed a greater willingness to purchase than one tendency to vary across weeks, and two sets of hourly who has used a generic keyword. fixed effects: one for weekdays and another for weekends.5 This allows baseline tendencies to vary accord- 2.3. Data Sources ing to times when white-collar workers are in the The analysis combines a large online search data set office or not. with a comprehensive television advertising database. A stock performance index, the Dow Jones Indus- The online data count all searches containing prod- trial Average (DJIA), is used as an exogenous variable uct category-relevant keywords received by Google to control for unobserved time-varying determinants from U.S. users in the eastern time zone between of searchers’ online actions. DJIA levels are widely October 1 and December 31, 2011.3 The search counts reported in the media; recent movements in the stock were aggregated hourly at the state level; information index may lead consumers to check their account bal- about individual users, their characteristics, or search ances by searching their financial service providers’ history was not included. Company disclosure poli- brand names. Two variables based on DJIA are cies prevent us from revealing the specific number of included: (1) the absolute positive percentage change searches for this set of keywords, but we are able to since the most recent trading day’s opening value and report that it was substantially more than one billion. (2) the absolute negative percentage change since the Television advertising expenditure data were gath- most recent trading day’s opening value. These vari- ered from Kantar Media’s “Stradegy” database. ables allow the effect to be asymmetric around zero Kantar computers monitor all paid advertisements in and proved to fit the data better than several alternate national broadcast and cable networks and all local specifications. broadcast stations in the United States. Each unique 2.5. Descriptive Statistics new advertising creative is flagged and watched by Table 3 describes the advertising and search data. a Kantar employee, who then records the advertised To comply with company disclosure policies, the brand, product, and product category. Kantar supple- search data are normalized so that the weekday ments the ad occurrence data with program-specific 4 average totals 100. Keyword search and advertising average advertising prices reported by the networks. expenditures varied considerably across brands. Four For each ad in the sample period, the data report brands (Charles Schwab, CitiGroup, E-Trade, and the brand advertised (e.g., Fidelity), the network, start TD Ameritrade) spent more than $100,000 per day time, duration, and the estimated cost. on TV advertising, whereas nine other brands spent These two time series are linked at the hour level. less than $1,000 per day. Generic financial services The choice of the hour as the unit of analysis bal- keywords are searched three times as frequently as ances the competing concerns of data sparseness and branded keywords. Category search fell by 29% from possible aggregation biases. It also requires fewer lags weekdays to weekend days, whereas brand advertis- to be estimated than if the data were combined at a ing expenditures were 85% higher on weekend days more disaggregate level (e.g., minute or second). The than on weekdays. sample period includes 2,208 hours in 92 days. There Figure 1 shows the variation in the advertising

Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. are 22 brands in the data, so the sample size in the expenditures of the three highest-spending brands keyword choice analysis is 48,576. across days in the sample. Whereas CitiGroup spent about equally on weekdays and weekends, E-Trade 3 Only searches in the eastern time zone were considered to match the data on TV ads’ times of airing. 5 Speciﬁcation tests preferred these two sets of day/hour dummies 4 Advertising ratings and demographics would be preferable to to a simpler alternative (24 hourly dummies that did not vary expenditures, but these additional data were cost prohibitive. across weekdays) and a more complex alternative (a different set In their absence, it is necessary to presume that ad prices corre- of 24 hourly dummies for each of the seven weekdays). The adver- late with program audiences. This presumption is common in the tising response parameter estimates were robust to the alternate literature and has substantial empirical support (e.g., Wilbur 2008). day/hour speciﬁcations. Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 61

Table 3 Descriptive Statistics where yt = ln4Yt5 is the log of the total number of category searches Y (including both generic and branded Monday–Friday Saturday–Sunday t daily averages averages keywords) in period t, at = ln41 + At5 is the log of one plus category television advertising expenditure Normalized Normalized 6 At in period t, is the effect of TV advertising query Adv. exp. query Adv. exp. − Brands volumea ($000) volumea ($000) done in period t on the number of category searches performed in period t, and T = 96 is the AG Edwards 0000 0 0000 0 number of hourly lags of advertising included in the Charles Schwab 0027 130 0011 298 model. The vector X contains control variables—an CitiGroup 5018 11 075 3041 11 153 t E-Trade 0019 151 0009 635 intercept, day/hour fixed effects, week fixed effects, Edward Jones 0001 38 0000 89 and the two DJIA index variables—described in §2.4; Fidelity 0066 96 0031 365 is a vector of parameters that represent the effects Forex 0038 4 0034 8 of the exogenous variables on category search, and FXCM 0001 2 0001 3 is an error term that accounts for any unobserved Legg Mason 0000 0 0000 0 t Merrill Lynch 9065 3 7052 2 determinants of the number of category searches in Morgan Stanley 0003 0 0002 0 period t. Nuveen 0001 0 0000 0 Given a volume of category searches, Equation (2) Oppenheimer 0005 0 0002 0 relates the keyword choice share for each brand OptionsXpress 0001 5 0000 0 k = 11 0 0 0 1 K in each period t to advertising and Raymond James 0000 14 0000 31 covariates: Scottrade 0019 22 0005 64 ShareBuilder 0003 0 0001 0 PT exp4 =0 ak1 t− + Xkt + kt5 T. Rowe Price 0009 32 0006 255 s = 1 (2) kt PK PT TD Ameritrade 6067 166 5015 311 1 + k0=1 exp4 =0 ak01 t− + Xk0t + k0t5 Vanguard 0034 0 0021 0 Van Kampen 0000 0 0000 0 where skt is the share of keywords entered in period t Wachovia 0026 0 0016 0 that are directly related to brand k (as defined in §2.2);

All brand 24000 11 738 17048 31 214 akt = ln41 + Akt5 is the log of one plus the television keywords advertising expenditure Akt of brand k in period t; All generic 76000 — 54017 — is the effect of TV advertising done by brand k in keywords period t − on its keyword choice share in period t; Total (brand+ 100000 71065 X is a vector of exogenous control variables that generic) kt includes an intercept, ﬁxed effects for brand, day- aNormalized query volume standardizes query totals so that the daily aver- hour, and week, and the two DJIA index variables; age of total searches (branded plus generic) on weekdays is 100. and kt is an error term that represents all unobserved determinants of brand k’s keyword choice share in and TD Ameritrade spent two to four times more period t. on weekend days. Figure 2 shows the distribution of Equation (2) can be transformed into a linear model these brands’ expenditures across hours on weekdays by taking logs and subtracting the log of generic key- and weekend days. The correlations are much higher words’ share from both sides (Berry 1994):

here, as television viewing peaks at prime time. How- T X ever, Citi’s media strategy relies on prime time to a zkt = ak1 t− + Xkt + kt1 (3) proportionally greater extent, whereas E-Trade and =0 TD Ameritrade tend to spend more money on week- where zkt = ln4skt5 − ln4s0t5 and s0t is the fraction of end afternoons. total category searches in period t for purely generic keywords (i.e., the share of searches that do not con- 3. Empirical Model tain any branded keywords). This section derives estimating equations to relate

Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. 3.2. Almon Parameterization of category search and keyword choice to television Advertising Effects advertising. The primary drawback of direct estimation of distributed-lag models such as (1) and (3) is that, for 3.1. Models Equation (1) relates category search in each day/hour period t to advertising and control variables: 6 The empirical ﬁndings are qualitatively robust to replacing this measure of advertising with raw expenditure or share of voice. T We use logs of search volume and advertising expenditures because X visual inspection of the data showed the relationship between these y = a − + X + 1 (1) t t t t two variables to be approximately log-linear. =0 Joo et al.: Television Advertising and Online Search 62 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Figure 1 Major Brands’ Ad Spending by Date

2,500 Citi E-trade 2,000 TD Ameritrade

1,500

1,000

Advertising expenditure ($000) 500

10/1/2011 10/8/2011 10/15/2011 10/22/2011 10/29/2011 11/5/2011 11/12/2011 11/19/2011 11/26/2011 12/3/2011 12/10/2011 12/17/2011 12/24/2011 12/31/2011

large T , the number of and parameters to be assumes that the effect of the th lag on current search estimated is large. Almon (1965) proposed a simpler behavior is given by a p-order polynomial in . For parameterization of distributed-lag models to address large p, this can provide a very ﬂexible representation this issue. The fully speciﬁed distributed-lag function of the effects of lagged advertising on search behavior. is replaced with an assumption that the effects of This transforms Equations (1) and (3) into lagged television advertising on search behavior can be approximated by a p–degree polynomial function of the time lag; that is, we assume p1 X yt = vt + Xt + t and (6) p1 X =0 = and (4) p =0 X2 zkt = wkt + Xkt + kt1 (7) p2 X =0 = 1 (5) =0 = PT = PT where the and terms are parameters to be where vt =0 at− and wkt =0 ak1 t− are estimated. In other words, the Almon (1965) model functions of past advertising that may be constructed

Figure 2 Major Brands’ Ad Spending by Day/Hour

20,000 Citi 18,000 E-trade 16,000 TD Ameritrade

14,000

12,000

10,000

6,000

4,000 Advertising expenditure ($000) 2,000

0 . . . . . n ...... M .M M M .M o .M .M M M M M M M .M. M M M M .M. A A A. A. A P P P.M P. P. A. A. A. A. A P. P. P. P. P 8 0 10 yno idnight y6 a ay2 ay4 y ay 6 day 4 da ay day 6 da end 6 end 8 nd 10 end 8 kday 2 k ekd k kend 2 kend 4 k k ekend noonkend 2 kend 4 k ek e e e e e e ekd ee ee e eekday 8 eekday 1end midnight e eekend 10 ekday mW Wee W W eekd We Weekd Weekd Week Wee k We We W We We We We We W e W W ee Weeke W W W Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 63

directly from the data (Gujarati 2003 provides a 3.4. Measure of Advertising Elasticities detailed explanation). The number of advertising A standard approach to presenting the size of an effectiveness parameters to be estimated has fallen advertising effect is to calculate it as an elasticity

from 2T to p1 + p2, and the original effects of the lags (e.g., Ataman et al. 2010, Lodish et al. 1995). We mea- can be recovered by plugging the estimates of and sure the percentage change in total future searches for 7 = PT into Equations (4) and (5). brand k (defined as qkt =0 Yt+ sk1 t+ , the number Davidson and MacKinnon (1993) and Gujarati of branded searches between points t and t +T , inclu- (2003) give prescriptive advice on how to choose sive) given a change in brand k’s television advertis- the number of lags (T ) and degree of polynomials ing expenditure (Akt5 at time t. From Equations (1) 8 and (2), this elasticity is (p1 and p2). The number of lags is settled first by including a relatively large number of lags and then ¡q A dropping some to check whether including fewer lags kt kt ¡A q appreciably reduces the fit of the model. Second, the kt kt T degree of the polynomial is specified for each model Akt X = Y s by starting from the lowest-order polynomial and P t+ k1 t+ qkt41 + k Akt5 =0 incrementing p until either the last parameter added T contains zero within its 95% confidence interval or Akt X + Yt+ sk1 t+ 41 − sk1 t+ 50 (10) until the covariance matrix of the parameter estimates qkt41 + Akt5 =0 becomes noninvertible because of multicollinearity. The first term on the right-hand side of Equation (10) represents the increase in searches for brand k’s key- 3.3. Serial Correlation in Search Behavior words due to the expansion in the number of cat- The measures of online search behavior are serially egory searches, holding brand k’s share of category correlated. A common remedy for this is to estimate searches constant. The second term is the increase the model by taking the first differences of Equa- in searches for brand k’s keywords accruing to its tions (6) and (7), increase in keyword choice share, holding total category searches constant. The first term is analogous to yt − yt−1 a “category expansion” effect, measuring the increase X in brand searches accruing to an increase in cate- = p4vpt − vp1 t−15 + 4Xt − Xt−15 + t − t−11 (8) gory search total, holding the brand’s keyword choice p=0 share constant. The second term is similar to a “busi- zkt − zk1 t−1 ness stealing” effect, showing the increase in the X brand’s searches due to its change in keyword choice = p4wkpt − wkp1 t−15 + 4Xkt − Xk1 t−15 share, holding total category search constant. The rel- p=0 ative size of each effect will influence the managerial

+ kt − k1 t−10 (9) implications of the results.

Analytically, Equations (8) and (9) estimate the 4. Identifying Assumptions same parameters as direct estimation of Equa- In the present application, the temporal ordering tions (6) and (7), but the first differences estimator is that advertisements must be purchased prior to air- more efficient than fixed effects estimator when the ing rules out the possibility that consumer search errors are serially correlated (Wooldridge 2010). The data (y) directly cause TV advertising placements (x). appendix provides further information about estima- However, there are two ways in which television tion, including modeling choices, serial correlation advertising (x) may depend indirectly on online tests, and standard errors. search behavior (y). One possible source of endogeneity is that brands Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. 7 A popular alternative to the Almon (1965) parameterization is anticipate when consumers will search and pur- the Koyck model, which would assume that an infinite number of chase television advertising at times that will max- lags enters Equations (1) and (2) and that the effects of those lags imally influence that search. To investigate this, we decrease monotonically and exponentially. Clarke (1976) discusses held a series of several dozen informal conversations the application of the Koyck model to advertising response, and with managers and researchers working in market- Gujarati (2003) compares the Koyck and Almon models. The Almon model is more flexible and can replicate a Koyck decay pattern ing and advertising at three leading Internet compa- when the number of lags is large and the degree of the polynomial nies (Google, Yahoo!, and Amazon) and two leading is two or larger. financial services brands in the sample. We began 8 For further discussion, see Thomas (1977). each conversation by asking whether brands typically Joo et al.: Television Advertising and Online Search 64 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Figure 3 Category Search and Ad Spending by Date

8,000 200 Category advertising expenditures 7,000 Daily aggregate category searches

6,000

5,000

4,000 100

3,000

2,000 (average daily searches =100) Advertising expenditure ($000)

1,000 Normalized number of category searches

0 0 1 1 1 1 1 1 11 11 11 11 11 0 0 0 0 011 0 /2 /20 /2 /2011 6 3 4/2011 2 10/1/20 10/8/2011 11/5/20 12/ /31 10/15/2 10/22/2 10/29/2 11/12/2 11/19/2011 11/2 12/10 12/17/2011 12/ 12

coordinate their television advertising and search search varies reasonably little over weeks in the sam- engine marketing campaigns. Our conversation part- ple. In addition, the two variables do not seem to ners were unanimous that they did not know of any correspond in their extremes. Advertising peaked on ﬁrms who were using any online search data to plan November 20, but there was no apparent movement their television advertising campaigns. in search volume, at least in the daily aggregates. We also investigated the available data to check Search activity peaked on December 5, but advertising for evidence that brands were anticipating spikes in expenditures lay below their sample average on this online search activity. If television advertising expen- date. Search volume stayed high in the several weeks ditures were planned on the basis of expected search following December 5, whereas advertising expendi- volume, one would expect advertising expenditures tures remained fairly low. The correlation between to increase immediately before or during periods daily advertising and daily search volume was −0028. of intense category search. Figures 3 and 4 present Figure 4 corroborates this pattern among hours data showing that such covariation is not obvious. within weekdays and weekend days. The majority Although category search and television advertising of advertising expenditures occur during prime time show extensive variation across days within the week, and weekend afternoons when category search is

Figure 4 Category Search and Ad Spending by Day/Hour

25,000 500 Category advertising expenditures Hourly aggregate category searches 20,000 400

15,000 300

Advertising expenditure ($000) 5,000 100 Normalized number of category searches

0 0 ...... n . ht .M .M. .M. .M. .M M M M M M .M. M .M. M M o M. M M. .M. ig A A A A A P. P. P. P. P. ight A A. A A. A. o P.M. P. P. P. P n 4 8 noon 2 8 n d 10 y y y4 y6 y 10 y d4 d8 d10 ay da a d10 end 2 end 6 n kday 2 kday kday 6 k d dmid kend 2 en e e ekda eekda eekda eekda k n eken day mi e eekdayekd e ee e k We W Wee W e Wee W W W W ek Wee WeekenWeekendWe 6 WeekendWeek n WeekendWeek 4 Weekend 8 W W Week Weeke Wee We Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 65

significantly lower than its peak during standard tures on category search. TV advertising has a small business hours. The available data do not allow us to but positive effect on category search. Of the 96 lag falsify the hypothesis that brands planned advertis- effects, 76 are positive and statistically significantly ing expenditures based on expectations of search data; different from zero. None may be proven to be less there is no clear evidence to support that hypothesis. than zero at a regular confidence level. However the The other possible source of endogeneity is that effects are quite small. A unit increase in log cate- television advertising expenditures may be correlated gory advertising expenditures will lift the number of with unobserved variables that also influence online queries for category-related keywords by about 0.5% search behavior. For example, if a financial services per hour over a period of about three days. This brand pulses its television advertising with advertis- nonimmediate, sustained, subtle effect may indicate ing expenditures in other media (such as Internet dis- that category advertising acts as a “reminder” that play advertising), and those other media influence increases the likelihood that a consumer will think online search behavior, then one would expect the about and search category-relevant keywords. Over- estimates reported in §5 to be inflated. all, the results indicate that category search rises after Under the identifying assumptions that (1) in- financial services advertising, showing that TV adver- tertemporal variation in category television advertis- tising has category expansion effects. ing is exogenous with respect to category searches Figure 6 shows the 95% confidence intervals for and (2) intertemporal variation in brand television the effects of lagged brand advertising expenditures advertising is exogenous with respect to brands’ key- on searchers’ tendency to choose branded keywords word choice shares, the estimates may be interpreted instead of generic keywords. TV advertising increases as causal. We believe these assumptions to be credi- branded keyword choice. Sixteen of the first 18 lagged ble based on our understanding of the industry and effects are positive and statistically significantly dif- our inspection of the data. However, the possibil- ferent from zero, indicating elevated levels of branded ity remains that unobserved variables may lead to keyword choice corresponding to the placement of inflated estimates of the effects. This alternate hypoth- that brand’s commercials on television. The esti- esis is not falsifiable using the available data. mates indicate an inverted-“U” pattern peaking in the 6–11 hour range after TV advertising goes on 5. Findings the air. All of the confidence intervals after the 17th This section presents results of first-difference regres- lag include zero. Along with the category expansion sions of Equations (8) and (9). effects of advertising, brand advertising also increases the tendency to use branded keywords, but this effect 5.1. Effects of Television Advertising on Category is much shorter lived. Search and Keyword Choice How big are these effects? The elasticity of a Figure 5 displays the 95% confidence intervals for brand’s total searches with respect to television adver- the effects of lagged category advertising expendi- tising may be calculated according to Equation (10).

Figure 5 Effects of Television Advertising on Category Search

0.012 Est. 0.010 95% confidence 0.008 interval

0.006

0.004

–0.002

–0.004 2-hour lag 4-hour lag 6-hour lag 8-hour lag 10-hour lag 12-hour lag 14-hour lag 16-hour lag 18-hour lag 20-hour lag 22-hour lag 24-hour lag 26-hour lag 28-hour lag 30-hour lag 32-hour lag 34-hour lag 36-hour lag 38-hour lag 40-hour lag 42-hour lag 44-hour lag 46-hour lag 48-hour lag 50-hour lag 52-hour lag 54-hour lag 56-hour lag 58-hour lag 60-hour lag 62-hour lag 64-hour lag 66-hour lag 68-hour lag 70-hour lag 72-hour lag 74-hour lag 76-hour lag 78-hour lag 80-hour lag 82-hour lag 84-hour lag 86-hour lag 88-hour lag 90-hour lag 92-hour lag 94-hour lag Current adv.

“s-hour lag” indicates the effect of TV advertising expenditure in period t-s on category searches in period t (s) Joo et al.: Television Advertising and Online Search 66 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Figure 6 Effects of Television Advertising on Keyword Choice

1.2 Est. 1.0 95% confidence 0.8 interval 0.6

0.4

0.2

0.0

–0.2

–0.4

–0.6

–0.8 2-hour lag 4-hour lag 6-hour lag 8-hour lag 10-hour lag 12-hour lag 14-hour lag 16-hour lag 18-hour lag 20-hour lag 22-hour lag 24-hour lag 26-hour lag 28-hour lag 30-hour lag 32-hour lag 34-hour lag 36-hour lag 38-hour lag 40-hour lag 42-hour lag 44-hour lag 46-hour lag 48-hour lag 50-hour lag 52-hour lag 54-hour lag 56-hour lag 58-hour lag 60-hour lag 62-hour lag 64-hour lag 66-hour lag 68-hour lag 70-hour lag 72-hour lag 74-hour lag 76-hour lag 78-hour lag 80-hour lag 82-hour lag 84-hour lag 86-hour lag 88-hour lag 90-hour lag 92-hour lag 94-hour lag Current adv.

“s-hour lag” indicates the effect of TV advertising expenditure in period t-s on keyword choice in period t (s)

Given a 10% increase in a brand’s television advertis- and long-run (0.12) advertising elasticities found by ing expenditures, the model predicts a 1.7% increase Ataman et al. (2010). Perhaps its magnitude can be in its number of branded searches over the follow- explained by the fact that consumer search is an action ing 96 hours. This elasticity is statistically signiﬁcantly that takes place at the top of the funnel, where adver- different from zero at the 99% conﬁdence level, with a tising is thought to be relatively more effective; pre- standard error of 0.2%. This effect magnitude is larger vious studies more frequently measured advertising’s than both the short-term market share elasticity of effect on sales, which requires consumers to go deeper advertising (0.05) and the long-term market share elas- into the purchase funnel and convert. ticity of advertising (0.10) reported by Lodish et al. Figure 7 shows how the elasticity estimate breaks (1995). Similarly, it exceeds both the short-run (0.01) down across sources of brand searches (the “category

Figure 7 Elasticities by Day/Hour

0.22 “Business stealing” effect 0.21 “Category expansion” effect 0.20

0.19

0.18

0.17

0.16

0.15

0.13

0.12 P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. Weekday noon Weekend noon Weekday 1 Weekend 1 Weekday 2 Weekday 3 Weekday 4 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 2 Weekend 3 Weekend 4 Weekend 5 Weekend 6 Weekend 7 Weekend 8 Weekend 9 Weekend 5 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 6 Weekend 7 Weekend 8 Weekend 9 Weekday 10 Weekday 11 Weekend 10 Weekend 11 Weekday 10 Weekday 11 Weekend 10 Weekend 11 Weekday midnight Weekend midnight Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 67

expansion” and “business stealing” effects discussed advertising exposure occurs during prime time and in §3.4) and by hours on weekdays and weekend that its effects have mostly dissipated within the fol- days. The effects of TV advertising on online search lowing 16 hours or so (as shown by Figure 6). Both are greatest in the morning and decrease throughout explanations probably help to explain the pattern of the day, before rising again after 5 p.m. and through- elasticity effects in Figure 7. out the evening. This pattern seems to be consistent with exposure to advertising during prime time and a 5.2. Effects of Control Variables gradual wearout prior to the next day’s evening tele- Here we present some of the effects of control vari- vision. The effects are slightly higher on weekdays ables on category search and keyword choice. Fig- than on weekends. ures 8 and 9 display the 95% confidence intervals of The dark area of the figure represents contribu- the effects of day/hour dummies on category search tion of category search expansion to the percent- and keyword choice, respectively. Category search age increase in searches for the brand’s keywords. rose rapidly between 6 a.m. and 10 a.m. on weekdays p.m. On average, business stealing effects are about two and then laid approximately flat until about 10 Its pattern on weekend days was similar. Keyword orders of magnitude larger than category expansion choice increased quickly from 5 a.m. until 10 a.m., but effects. The graph shows that television advertising started to fall after 5 p.m. On weekend days it laid does not generate large positive spillovers, suggesting mostly flat between 10 a.m. and 10 p.m. that brands will not benefit much by free riding on Stock market index parameter estimates (in Table 4) competitors’ advertising expenditures. cannot be distinguished from zero. Analyses of secondary data are typically unable to discern precise behavioral mechanisms, but we can 5.3. Polynomial Parameters and speculate on the causes. There are two overlapping Alternative Specifications explanations for the pattern in Figure 7 in which elas- The procedure described in §3.2 led us to set T = 96, ticities peak in the morning and fall throughout busi- p1 = 2, and p2 = 6. We found that estimating higher- ness hours. First, it is possible that the part of the order polynomials did not change the statistical sig- audience whose search for financial services brands nificance of the lower-order polynomial parameters. is influenced by television advertising is more likely Table 5 reports the parameter estimates (p and p) to search in the morning. For example, it might be underlying the effects of lagged TV advertising in that working professionals are both more likely to be Figures 5 and 6. Individual polynomial parameters exposed to television advertising for financial services do not correspond to marginal effects, but they are brands, and more likely to enter Google searches in included for completeness. the morning, than the average person. Another pos- We previously estimated distributed-lag models sible explanation is that these effects could be due without the Almon parameterization. The unre- solely to consumers’ memories and the duration of stricted distributed-lag models found qualitatively the effect of advertising on search. It may be that most similar effects (positive effects of TV advertising on

Figure 8 Day/Hour Fixed Effect Estimates in Category Search Model

1.5

1.0

0.5

0.0

Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. –0.5 Est. 95% –1.0 confidence interval –1.5 P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. Weekday noon Weekend noon Weekday 1 Weekday 2 Weekday 3 Weekday 4 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 1 Weekend 2 Weekend 3 Weekend 4 Weekend 5 Weekend 6 Weekend 7 Weekend 8 Weekend 9 Weekday 1 Weekday 3 Weekday 4 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 1 Weekend 2 Weekend 3 Weekend 4 Weekend 5 Weekend 6 Weekend 7 Weekend 9 Weekday 2 Weekend 8 Weekday 10 Weekday 11 Weekend 10 Weekend 11 Weekday 11 Weekend 10 Weekend 11 Weekday 10 Weekend midnight Joo et al.: Television Advertising and Online Search 68 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Figure 9 Day/Hour Fixed Effect Estimates in Keyword Choice Model

60 50 40 30 20 10 0 –10 Est. –20 95% confidence –30 interval –40 P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. P.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. A.M. Weekday noon Weekend noon Weekday 1 Weekday 2 Weekday 3 Weekday 4 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 1 Weekend 2 Weekend 3 Weekend 4 Weekend 5 Weekend 6 Weekend 7 Weekend 8 Weekend 9 Weekday 2 Weekday 1 Weekday 3 Weekday 4 Weekday 5 Weekday 6 Weekday 7 Weekday 8 Weekday 9 Weekend 1 Weekend 2 Weekend 3 Weekend 4 Weekend 5 Weekend 6 Weekend 7 Weekend 8 Weekend 9 Weekday 10 Weekday 11 Weekend 10 Weekend 11 Weekday 11 Weekend 10 Weekend 11 Weekday 10 Weekend midnight

both category search and use of branded keywords), 6. Discussion but the lagged advertising parameters showed a This paper has investigated two large data sets to cyclical pattern with a 24-hour frequency and little investigate how television advertising expenditures apparent decay over time. The fluctuations seemed to are related to online search. It found that television match the pattern of autocorrelation in the advertis- advertising increases the number of product category- ing expenditure data, so the Almon parameterization relevant searches and increases the advertised brand’s was thought to be a reasonable way to simplify the share of keywords searched. For financial services, this model specification. latter effect dominates; the primary effect of a brand’s advertising expenditure is to “steal” query share from Table 4 Stock Index Effects rivals more so than to increase the number of searches Parameter Newey–West in the product category. The elasticity of a brand’s estimate std. err. searches with respect to its advertising is 0.17. The most important implication of these results is Category search the need for advertisers to consider how television Absolute positive % DJIA change −00012 00008 Absolute negative % DJIA change 00006 00007 advertising may impact their search advertising cam- Adjusted R2 00944 paigns. Search engine marketing is typically run as a Keyword choice standalone activity that seeks to maximize its incre- Absolute positive % DJIA change 1057 6032 mental profits. The importance of cross-channel strat- Absolute negative % DJIA change −5008 6040 egy in marketing has been documented previously, Adjusted R2 00005 but managers are still relying on experience and gut intuition to decide how to divide advertising budgets Table 5 Advertising Polynomial Parameter Estimates across media (Pfeiffer and Zinnbauer 2010). The effects of television advertising on online Parameter Newey–West search may differ across product categories, con- estimate std. err. sumers, or time. The easiest way for a brand to Category search investigate these effects is to use A/B tests to mea- 0 −2.0E−04 1.5E−03 sure changes in online behavior corresponding to the ∗∗ 1 2.4E−04 1.1E−04 ∗∗ occurrence of television advertisement. It would also Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. 2 −2.4E−06 1.1E−06 Keyword choice be possible to use the number of people exposed to a TV advertisement within a local market to explain 0 −1.1E−01 1.0E−01 ∗∗ 1 2.8E−01 3.7E−02 market-specific measures of online activity. Lewis and ∗∗ 2 −2.8E−02 4.2E−03 Reiley (2013) explore these issues in greater depth. ∗∗ 3 1.0E−03 1.8E−04 Knowledge of these effects may lead a mar- 1.8E−05 3.6E−06∗∗ 4 − keter to alter its pulsing, creatives, budgets, and 1.4E−07 3.3E−08∗∗ 5 return-on-investment metrics for both television and −4.5E−10 1.1E−10∗∗ 6 search advertising. It is likely advisable to facilitate ∗∗Significant at the 99% confidence level. or require coordination between the two agencies Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 69

executing the advertising campaigns in each medium. research support. Kenneth Wilbur was employed by the For example, Web search metrics could be used as an Duke University Fuqua School of Business and Yi Zhu was input into the TV advertising campaign dashboard to a doctoral student at University of Southern California Mar- help guide realtime creative decisions.9 When televi- shall School of Business while much of this research was sion advertising increases branded keyword choice, conducted. Both members of the first and second pairs of authors contributed equally and are listed in alpha- it will reduce the number of expensive clicks on betical order. The authors gratefully acknowledge finan- generic keyword searches and increase the number of cial support from Marketing Science Institute/Wharton clicks paid for on cheaper branded keywords. (Costs Interactive Media Initiative multichannel research award per click are typically higher for generic category key- [Award 4-1645]. words because of greater competition in the keyword auction.) It also might increase the conversion rate of Appendix. Further Empirical Details both branded and generic search keywords. The mar- This appendix gives further details about the estimation and keter who remains ignorant of these effects may risk results. underspending on television advertising and over- Functional Forms 10 spending on search advertising. Several functional forms were considered for use in Equa- This study has a number of limitations. It consid- tion (1). Figure A.1 plots the log of category search against ered only one product category. Television advertising for new brands or for evolving categories would Figure A.1 Log Category Search vs. Category Advertising Expenditure likely show stronger category expansion and lower business stealing effects than the ones found here. The available data excluded direct website traffic, additional search engines, social networks, paid search advertisements, and click-through and conversion information. The effects were based on time series identification and did not come from a single-source panel in which advertising exposures and search behavior are observed within a single household. We hope that some of the issues raised in this paper will stimulate field experiments. Though it is well known that exposure to online advertising affects online purchase behavior (e.g., Manchanda et al. 2006), more information is needed about how adver- Log of category search volume (hidden) tising creative elements interact with consumers’ stages in the purchase funnel to determine product 0 500 1,000 1,500 choices. Although Internet commerce enables marketers to observe many aspects of the consumer Advertising expenditure ($000) choice process, further research is needed to deter- Figure A.2 Log Category Search vs. Log Category Advertising mine how to optimize all types of advertising to max- Expenditure imize advertising effectiveness.

Acknowledgments The authors thank Pradeep Chintagunta, the associate edi- tor, and three anonymous reviewers for their valuable and insightful comments. They also thank Randall Lewis, Preston McAfee, Hal Varian, and numerous seminar audiences for helpful comments. Bo Cowgill thanks Google for

9 Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. With DVRs found in nearly 45% of U.S. households (Televi- sion Bureau of Advertising 2012) and live audiences falling by as much as 30% during commercial breaks (Schweidel and Kent 2010), search data may offer an important means to test television advertising effectiveness. As Swasy and Rethans (1986, p. 33) suggested,

“advertisers should begin to incorporate a measure of curiosity Log of category search volume (hidden) generation and question solicitation in their new-product ad con- cept and ad-testing procedures.” 10 As Wiesel et al. (2011, p. 609) noted, “it is unwise to credit a 4681012 14 marketing activity only for orders in ‘its’ channel, a practice typical Log advertising expenditure in companies with different managers for different channels.” Joo et al.: Television Advertising and Online Search 70 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Table A.1 Week Fixed Effects to Control for Seasonality hypothesis of no serial correlation with a z-statistic of 41.84 (p < 0005). When first differences are taken and Equation (8) Category search Keyword choice is estimated instead, the z-statistic is −1045 (p > 0005), which Parameter Newey–West Parameter Newey–West fails to reject the null of no serial correlation. Therefore, estimate std. err. estimate std. err. estimation results based on first differences are preferred (Wooldridge 2010). Week 3 (Oct.) −00005 00019 −404 2204 Week 4 (Oct.) 00003 00039 −4909 4302 Standard Errors Week 5 (Oct.) 00015 00058 −7407 5507 The first-difference estimator controls completely for time- Week 6 (Oct./Nov.) 00049 00077 −5805 6107 invariant unobserved heterogeneity, but it remains possi- Week 7 (Nov.) 00040 00097 −8303 7308 ble that the standard errors are biased by serial correlation Week 8 (Nov.) 00080 00116 −12804 9104 Week 9 (Nov.) 00134 00134 −11201 10200 (Bertrand et al. 2004). Therefore all estimates are reported Week 10 (Nov./Dec.) 00132 00154 −7505 11401 using the standard errors suggested by Newey and West Week 11 (Dec.) 00187 00173 −10003 12401 (1987), which are robust to heteroskedasticity and serial cor- Week 12 (Dec.) 00220 00193 −8309 13403 relation of unknown form. Clustered standard errors and Week 13 (Dec.) 00309 00212 −4707 14300 nonparametric “bootstrap” standard errors were similar in Week 14 (Dec.) 00387 00232 −7203 15300 magnitude to the Newey–West standard errors and did not change the qualitative conclusions. Wooldridge (2010) provides a complete discussion. the level of category advertising expenditures. It is difficult to discern the shape of the relationship because there are Week Fixed Effects many time periods with zero category advertising dollars. Week fixed effects are presented in Table A.1 as controls Figure A.2 shows a log–log plot. Based on this, a log-linear for seasonality in category search tendency and keyword specification represents the data better. choice. Category search was lowest in the third week of the sample, 0.5% lower than in the baseline second week. Serial Correlation Tests It peaked in the fourteenth week of the sample, 38.7% The standard Arellano and Bond (1991) test for serial cor- higher than the baseline week. The baseline tendency to relation was applied to Equation (1) and rejected the null choose branded keywords also varied over time.

Table A.2 Lagged Advertising Parameter Estimates

Category search Keyword choice Category search Keyword choice

Newey–West Newey–West Newey–West Newey–West Est. std. err. Est. std. err. Est. std. err. Est. std. err.

Current advertising −000002 000015 −00113 00101 Lagged advertising (48 hrs) 000059 000021∗∗ 00163 00178 Lagged advertising (1 hr) 000000 000014 00140 00093 Lagged advertising (49 hrs) 000059 000021∗∗ 00187 00178 Lagged advertising (2 hrs) 000003 000014 00343 00094∗∗ Lagged advertising (50 hrs) 000059 000021∗∗ 00210 00177 Lagged advertising (3 hrs) 000005 000013 00502 00100∗∗ Lagged advertising (51 hrs) 000059 000021∗∗ 00231 00177 Lagged advertising (4 hrs) 000007 000013 00622 00107∗∗ Lagged advertising (52 hrs) 000059 000021∗∗ 00249 00176 Lagged advertising (5 hrs) 000010 000012 00708 00113∗∗ Lagged advertising (53 hrs) 000059 000021∗∗ 00265 00175 Lagged advertising (6 hrs) 000012 000012 00764 00119∗∗ Lagged advertising (54 hrs) 000059 000021∗∗ 00279 00173 Lagged advertising (7 hrs) 000014 000012 00794 00123∗∗ Lagged advertising (55 hrs) 000059 000021∗∗ 00290 00172 Lagged advertising (8 hrs) 000016 000012 00803 00127∗∗ Lagged advertising (56 hrs) 000059 000021∗∗ 00299 00171 Lagged advertising (9 hrs) 000018 000012 00792 00130∗∗ Lagged advertising (57 hrs) 000058 000020∗∗ 00305 00169 Lagged advertising (10 hrs) 000020 000012 00767 00133∗∗ Lagged advertising (58 hrs) 000058 000020∗∗ 00309 00168 Lagged advertising (11 hrs) 000022 000012 00728 00136∗∗ Lagged advertising (59 hrs) 000058 000020∗∗ 00311 00166 Lagged advertising (12 hrs) 000024 000012 00680 00139∗∗ Lagged advertising (60 hrs) 000057 000020∗∗ 00310 00165 Lagged advertising (13 hrs) 000026 000013∗∗ 00624 00143∗∗ Lagged advertising (61 hrs) 000057 000020∗∗ 00307 00164 Lagged advertising (14 hrs) 000027 000013∗∗ 00563 00146∗∗ Lagged advertising (62 hrs) 000056 000019∗∗ 00301 00163 Lagged advertising (15 hrs) 000029 000013∗∗ 00498 00150∗∗ Lagged advertising (63 hrs) 000055 000019∗∗ 00294 00163 Lagged advertising (16 hrs) 000031 000013∗∗ 00431 00153∗∗ Lagged advertising (64 hrs) 000055 000019∗∗ 00286 00162 Lagged advertising (17 hrs) 000032 000014∗∗ 00364 00157∗∗ Lagged advertising (65 hrs) 000054 000018∗∗ 00276 00162 Lagged advertising (18 hrs) 000034 000014∗∗ 00297 00160 Lagged advertising (66 hrs) 000053 000018∗∗ 00265 00162 Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. Lagged advertising (19 hrs) 000036 000015∗∗ 00233 00164 Lagged advertising (67 hrs) 000053 000018∗∗ 00253 00161 Lagged advertising (20 hrs) 000037 000015∗∗ 00171 00167 Lagged advertising (68 hrs) 000052 000017∗∗ 00240 00161 Lagged advertising (21 hrs) 000039 000015∗∗ 00112 00169 Lagged advertising (69 hrs) 000051 000017∗∗ 00227 00161 Lagged advertising (22 hrs) 000040 000016∗∗ 00058 00171 Lagged advertising (70 hrs) 000050 000017∗∗ 00213 00161 Lagged advertising (23 hrs) 000041 000016∗∗ 00009 00173 Lagged advertising (71 hrs) 000049 000016∗∗ 00200 00161 Lagged advertising (24 hrs) 000043 000016∗∗ −00035 00174 Lagged advertising (72 hrs) 000048 000016∗∗ 00188 00160 Lagged advertising (25 hrs) 000044 000017∗∗ −00074 00175 Lagged advertising (73 hrs) 000047 000016∗∗ 00176 00159 Lagged advertising (26 hrs) 000045 000017∗∗ −00107 00175 Lagged advertising (74 hrs) 000046 000015∗∗ 00164 00158 Lagged advertising (27 hrs) 000046 000017∗∗ −00134 00175 Lagged advertising (75 hrs) 000044 000015∗∗ 00154 00157 Lagged advertising (28 hrs) 000047 000018∗∗ −00156 00175 Lagged advertising (76 hrs) 000043 000014∗∗ 00145 00156 Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 71

Table A.2 (Continued)

Category search Keyword choice Category search Keyword choice

Newey–West Newey–West Newey–West Newey–West Est. std. err. Est. std. err. Est. std. err. Est. std. err.

Lagged advertising (29 hrs) 000048 000018∗∗ −00172 00175 Lagged advertising (77 hrs) 000042 000014∗∗ 00138 00154 Lagged advertising (30 hrs) 000049 000018∗∗ −00183 00174 Lagged advertising (78 hrs) 000041 000014∗∗ 00131 00152 Lagged advertising (31 hrs) 000050 000019∗∗ −00188 00174 Lagged advertising (79 hrs) 000039 000013∗∗ 00126 00150 Lagged advertising (32 hrs) 000051 000019∗∗ −00189 00173 Lagged advertising (80 hrs) 000038 000013∗∗ 00123 00147 Lagged advertising (33 hrs) 000052 000019∗∗ −00185 00173 Lagged advertising (81 hrs) 000036 000013∗∗ 00121 00145 Lagged advertising (34 hrs) 000053 000020∗∗ −00177 00173 Lagged advertising (82 hrs) 000035 000012∗∗ 00119 00143 Lagged advertising (35 hrs) 000054 000020∗∗ −00164 00173 Lagged advertising (83 hrs) 000033 000012∗∗ 00119 00140 Lagged advertising (36 hrs) 000054 000020∗∗ −00149 00173 Lagged advertising (84 hrs) 000032 000012∗∗ 00118 00137 Lagged advertising (37 hrs) 000055 000020∗∗ −00130 00173 Lagged advertising (85 hrs) 000030 000012∗∗ 00118 00135 Lagged advertising (38 hrs) 000056 000020∗∗ −00108 00173 Lagged advertising (86 hrs) 000028 000012∗∗ 00116 00131 Lagged advertising (39 hrs) 000056 000021∗∗ −00084 00174 Lagged advertising (87 hrs) 000027 000012∗∗ 00113 00128 Lagged advertising (40 hrs) 000057 000021∗∗ −00059 00174 Lagged advertising (88 hrs) 000025 000012∗∗ 00108 001233 Lagged advertising (41 hrs) 000057 000021∗∗ −00031 00175 Lagged advertising (89 hrs) 000023 000012 00099 00118 Lagged advertising (42 hrs) 000058 000021∗∗ −00003 00176 Lagged advertising (90 hrs) 000021 000012 00086 00112 Lagged advertising (43 hrs) 000058 000021∗∗ 00025 00176 Lagged advertising (91 hrs) 000019 000012 00066 00105 Lagged advertising (44 hrs) 000058 000021∗∗ 00054 00177 Lagged advertising (92 hrs) 000017 000013 00039 00097 Lagged advertising (45 hrs) 000059 000021∗∗ 00082 00178 Lagged advertising (93 hrs) 000015 000013 00002 00090 Lagged advertising (46 hrs) 000059 000021∗∗ 00110 00178 Lagged advertising (94 hrs) 000013 000014 −00046 00087 Lagged advertising (47 hrs) 000059 000021∗∗ 00137 00178 Lagged advertising (95 hrs) 000011 000014 −00107 00093

∗∗Signiﬁcant at the 99% conﬁdence level.

Table A.3 (Continued) Table A.3 Day/Hour Fixed Effect Parameter Estimates Category search Keyword choice Category search Keyword choice Newey–West Newey–West Newey–West Newey–West Est. std. err. Est. std. err. Est. std. err. Est. std. err. Weekend noon 00560 00025∗∗ 310015 70330∗∗ Weekday 1 a.m. −00371 00005∗∗ −90094 30514∗∗ Weekend 1 p.m. 00572 00025∗∗ 270231 70337∗∗ Weekday 2 a.m. −00650 00009∗∗ −190643 30582∗∗ Weekend 2 p.m. 00595 00026∗∗ 360510 70385∗∗ ∗∗ ∗∗ Weekday 3 a.m. −00851 00011 −170311 30634 Weekend 3 p.m. 00592 00027∗∗ 320692 70288∗∗ ∗∗ ∗∗ Weekday 4 a.m. −00901 00013 −190306 30799 Weekend 4 p.m. 00596 00030∗∗ 280062 70417∗∗ Weekday 5 a.m. −00812 00014∗∗ −20742 30690 Weekend 5 p.m. 00591 00031∗∗ 290132 70360∗∗ Weekday 6 a.m. −00485 00016∗∗ 40160 30629 Weekend 6 p.m. 00582 00032∗∗ 270634 70413∗∗ Weekday 7 a.m. −00055 00017∗∗ 240220 40010∗∗ Weekend 7 00632 00035∗∗ 350117 70854∗∗ Weekday 8 a.m. 00461 00018∗∗ 320410 30950∗∗ p.m. ∗∗ ∗∗ Weekday 9 a.m. 00804 00019∗∗ 440488 40011∗∗ Weekend 8 p.m. 00689 00039 300201 80052 ∗∗ ∗∗ Weekday 10 a.m. 00937 00018∗∗ 440481 30966∗∗ Weekend 9 p.m. 00691 00032 330279 80306 Weekday 11 a.m. 00962 00018∗∗ 430821 30967∗∗ Weekend 10 p.m. 00567 00025∗∗ 250718 70813∗∗ Weekday noon 00913 00019∗∗ 440880 30954∗∗ Weekend 11 p.m. 00348 00017∗∗ 180908 70940∗∗ Weekday 1 p.m. 00939 00020∗∗ 430972 30924∗∗ ∗∗ Weekday 2 p.m. 00955 00022∗∗ 430811 30971∗∗ Significant at the 99% confidence level. Weekday 3 p.m. 00950 00022∗∗ 440021 30962∗∗ Weekday 4 p.m. 00912 00022∗∗ 430842 30974∗∗ ∗∗ ∗∗ Weekday 5 p.m. 00752 00020 380060 40040 For completeness, Tables A.2 and A.3 present the param- Weekday 6 p.m. 00669 00018∗∗ 260419 30945∗∗ Weekday 7 p.m. 00698 00017∗∗ 260860 40021∗∗ eter estimates and standard errors used to construct Fig- Weekday 8 p.m. 00740 00016∗∗ 290770 30886∗∗ ures 5, 6, 8, and 9. Weekday 9 p.m. 00724 00014∗∗ 210931 30751∗∗ Weekday 10 p.m. 00595 00010∗∗ 220363 30874∗∗ Weekday 11 p.m. 00338 00007∗∗ 120034 30338∗∗ References ∗∗ Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. Weekend midnight 00067 00012 20379 70567 ∗∗ AdAge (2011) Agency report 2011 index. Accessed April 24, 2013, Weekend 1 a.m. −00233 00026 −60906 80147 http://adage.com/article/datacenter-agencies/agency-report Weekend 2 00531 00028∗∗ 150228 70772 a.m. − − -2011-index/226900/. Weekend 3 a.m. −00757 00027∗∗ −180421 70679∗∗ Weekend 4 a.m. −00847 00032∗∗ −140926 70733 Almon S (1965) The distributed lag between capital appropriations Weekend 5 a.m. −00880 00028∗∗ −130538 70693 and expenditures. Econometrica 41(4):775–788. Weekend 6 a.m. −00681 00030∗∗ −10379 70450 Arellano M, Bond S (1991) Some tests of specification for panel Weekend 7 a.m. −00298 00033∗∗ 200490 70245∗∗ data: Monte carlo evidence and an application to employment Weekend 8 a.m. 00109 00032∗∗ 160891 70203∗∗ equations. Rev. Econom. Stud. 58(2):277–297. Weekend 9 a.m. 00369 00029∗∗ 270072 70354∗∗ Assael H (2011) From silos to synergy: A fifty-year review of cross- Weekend 10 a.m. 00493 00028∗∗ 280180 70300∗∗ media research shows synergy has yet to achieve its full poten- Weekend 11 a.m. 00553 00026∗∗ 260779 70244∗∗ tial. J. Advertising Res. 51(1):42–48. Joo et al.: Television Advertising and Online Search 72 Management Science 60(1), pp. 56–73, © 2014 INFORMS

Ataman MB, van Heerde HJ, Mela CF (2010) The long-term effect Lewis RA, Reiley DH (2011) Does retail advertising work? of marketing strategy on brand performance. J. Marketing Res. Measuring the effects of advertising on sales via a con- 47(5):866–882. trolled experiment at Yahoo! Unpublished manuscript, Yahoo! Bergemann D, Bonatti A (2011) Targeting in advertising markets: Accessed April 24, 2013, http://www.davidreiley.com/papers/ Implications for offline vs. online media. RAND J. Econom. DoesRetailAdvertisingWork.pdf. 42(3):417–443. Lewis RA, Reiley DH (2013) Down-to-the-minute effects of Super Berry ST (1994) Estimating discrete choice models of product dif- Bowl advertising on online search behavior. 14th ACM Conf. ferentiation. RAND J. Econom. 25(2):242–262. Electronic Commerce 9(4):Article 61. Bertrand M, Duflo E, Mullainathan S (2004) How much should Lodish LM, Abraham M, Kalmenson S, Livelsberger J, Lubetkin B, we trust differences-in-differences estimates? Quart. J. Econom. Richardson B, Stevens ME (1995) How TV advertising works: 119(1):249–275. A meta-analysis of 389 real world split cable TV advertising Bettman J, Park CW (1980) Effects of prior knowledge and expe- experiments. J. Marketing Res. 32(2):125–139. rience and phase of the choice process on consumer decision Manchanda P, Dube JP, Goh KY, Chintagunta PK (2006) The effect processes: A protocol analysis. J. Consumer Res. 7(3):234–248. of banner advertising on Internet purchasing. J. Marketing Res. Brucks M (1985) The effects of product class knowledge on infor- 43(1):98–108. mation search behavior. J. Consumer Res. 12(1):1–16. Moorman C, Diehl K, Brinberg D, Kidwell B (2004) Subjective Bureau of Labor Statistics (2011) American time use survey, Table 1. knowledge, search locations, and consumer choice. J. Consumer Accessed April 24, 2013, http://www.bls.gov/news.release/ Res. 31(3):673–680. atus.t01.htm. Naik PA, Peters K (2009) A hierarchical marketing communications Clarke DG (1976) Econometric measurement of the duration of model of online and offline media synergies. J. Interactive Mar- advertising effect on sales. J. Marketing Res. 13(4):345–357. keting 23(4):288–299. comScore (2012) comScore releases July 2012 U.S. search engine Naik PA, Raman K (2003) Understanding the impact of synergy in rankings. Accessed April 24, 2013, http://www.comscore multimedia communications. J. Marketing Res. 13(4):375–388. .com/Press_Events/Press_Releases/2012/8/comScore_Releases Newey WK, West KD (1987) A simple, positive semi-definite, _July_2012_U.S._Search_Engine_Rankings. heteroskedasticity and autocorrelation consistent covariance Danaher PJ, Rossiter JR (2011) Comparing perceptions of marketing matrix. Econometrica 55(3):703–708. channels. Eur. J. Marketing 45(1/2):6–42. Newman JW, Staelin R (1973) Information sources for durable Davidson R, MacKinnon JG (1993) Estimation and Inference in Econo- goods. J. Advertising Res. 13(2):19–29. metrics (Oxford University Press, New York). Nielsen (2011) State of the Media 2010: Audiences and devices. eMarketer (2012) Traditional media still most trusted sources of Accessed April 24, 2013, http://blog.nielsen.com/nielsenwire/ info. Accessed April 24, 2013, http://www.emarketer.com/ wp-content/uploads/2011/01/nielsen-media-fact-sheet-jan Article.aspx?R=1009182. -11.pdf. Enge E (2012) Integrated marketing: Why search needs a large Papadimitriou P, Garcia-Molina H, Krishnamurthy P, Lewis R, seat at the table. Search Engine Watch (April 8), http:// Reiley D (2011) Display advertising impact: Search lift and searchenginewatch.com/article/2166526/Integrated-Marketing social influence. Proc. 17th ACM SIGKDD Internat. Conf. Knowl- -Why-Search-Needs-a-Large-Seat-at-the-Table. edge Discovery and Data Mining (ACM, New York), 1019–1027. Goldfarb A, Tucker CE (2011) Search engine advertising: Chan- Park CW, Mothersbaugh DL, Feick L (1994) Consumer knowledge nel substitution when pricing ads to context. Management Sci. assessment. J. Consumer Res. 21(1):71–82. 57(3):458–470. Pfeiffer M, Zinnbauer M (2010) Can old media enhance new media? Gujarati DN (2003) Basic Econometrics, 4th ed. (McGraw-Hill/ How traditional advertising pays off for an online social net- Irwin). work. J. Advertising Res. 50(1):42–49. Internet Advertising Bureau (2012) IAB Internet advertising rev- enue report. Accessed April 24, 2013, http://www.iab.net/ Radecki CM, Jaccard J (1995) Perceptions of knowledge, actual AdRevenueReport. knowledge and information search behavior. J. Experiment. Soc. Psych. 31(2):107–138. iProspect (2007) Offline channel influence on online search behavior study. Accessed March 1, 2011, http://www.iprospect.com/ Rutz OJ, Bucklin RE (2011) From generic to branded: A model premiumPDFs/researchstudy_2007_offlinechannelinfluence.pdf. of spillover in paid search advertising. J. Marketing Res. 48(1):87–102. Janiszewski C (1993) Preattentive mere exposure effects. J. Consumer Res. 20(3):376–392. Rutz OJ, Bucklin RE (2012) Does banner advertising affect brows- Jerath K, Sayedi A (2011) Exclusive display in sponsored ing for brands? Clickstream choice model says yes, for some. search advertising. Unpublished manuscript, Carnegie Mellon Quant. Marketing Econom. 10(2):231–257. University, Pittsburgh. Rutz OJ, Trusov M (2011) Zooming in on paid search ads—A Jerath K, Ma L, Park YH, Srinivasan K (2011) A “position paradox” consumer-level model calibrated on aggregated data. Market- in sponsored search auctions. Marketing Sci. 30(4):612–627. ing Sci. 30(5):789–800. Joo M, Wilbur KC, Zhu Y (2013) Effects of television adver- Sayedi A, Jerath K, Srinivasan K (2011) Competitive poaching in tising on Internet search. Unpublished manuscript, Ohio sponsored search advertising and strategic impact on tradi- Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved. State University, Columbus. Accessed April 24, 2013, http:// tional advertising. Unpublished manuscript, Carnegie Mellon papers.ssrn.com/sol3/papers.cfm?abstract_id=1720713. University, Pittsburgh. Katona Z, Sarvary M (2010) The race for sponsored links: Bidding Schweidel DA, Kent RJ (2010) Predictors of the gap between pro- patterns for search advertising. Marketing Sci. 29(2):199–215. gram and commercial audiences: An investigation using live Kim A, Balachander S (2010) Coordinating traditional media tuning data. J. Marketing 74(3):18–33. advertising and search advertising. Unpublished manuscript, Shapiro S (1999) When an ad’s influence is beyond our conscious Purdue University, West Lafayette, IN. control: Perceptual and conceptual fluency effects caused by Lewis R, Nguyen D (2012) Display advertising’s impact on incidental ad exposure. J. Consumer Res. 26(1):16–36. online branded search. Unpublished manuscript, Yahoo! Shapiro S, MacInnis DJ, Heckler SE (1997) The effects of incidental Accessed April 24, 2013, http://conference.nber.org/confer/ ad exposure on the formation of consideration sets. J. Consumer 2012/SI2012/PRIT/Lewis_Nguyen.pdf. Res. 24(1):94–104. Joo et al.: Television Advertising and Online Search Management Science 60(1), pp. 56–73, © 2014 INFORMS 73

Swasy JL, Rethans AJ (1986) Knowledge effects on curiosity and Wilbur KC (2008) A two-sided, empirical model of television adver- new product advertising. J. Advertising 15(4):28–34. tising and viewing markets. Marketing Sci. 27(3):356–378. Television Bureau of Advertising (2012) TV basics. Accessed April Wilbur KC, Zhu Y (2009) Click fraud. Marketing Sci. 28(2): 24, 2013, http://www.tvb.org/media/file/TV_Basics.pdf. 293–308. Thomas JJ (1977) Some problems in the use of Almon’s tech- Wooldridge JM (2010) Econometric Analysis of Cross Section and Panel Data, 2nd ed. (MIT Press, Cambridge, MA). nique in the estimation of distributed lags. Empirical Econom. 2(3):175–193. Yang S, Ghose A (2010) Analyzing the relationship between organic and sponsored search advertising: Positive, negative, or zero U.S. Census Bureau (2011) U.S. QuickFacts. Accessed April 24, 2013, interdependence? Marketing Sci. 29(4):602–623. http://quickfacts.census.gov/qfd/states/00000.html. Yao S, Mela CF (2011) A dynamic model of sponsored search adver- Vakratsas D, Ma Z (2005) A look at the long-run effectiveness of tising. Marketing Sci. 30(5):447–468. multimedia advertising and its implications for budget alloca- Zhu Y, Wilbur KC (2010) Hybrid advertising auctions. Marketing tion decisions. J. Advertising Res. 45(2):241–254. Sci. 30(2):249–273. Wiesel T, Pauwels K, Arts J (2011) Marketing’s profit impact: Quan- Zigmond D, Stipp H (2010) Assessing a new advertising effect: tifying online and offline funnel progression. Marketing Sci. Measurement of the impact of television commercials on Inter- 30(4):604–611. net search queries. J. Advertising Res. 50(2):162–168. Downloaded from informs.org by [128.54.18.48] on 26 May 2016, at 14:01 . For personal use only, all rights reserved.