Negativity Bias and Ideology, p.1
Behaviorally Indicated Negativity Bias, Personality, and Political Ideology
Abstract
Recent work suggests that right-wing political preferences are associated with individual differences in negativity bias: a tendency to pay more attention, and give more weight, to negative versus positive stimuli in decision making. This work tends to rely on either self-report measures of personality in large, often-representative samples, or physiological and behavioral indicators of negativity bias in small and unrepresentative samples. Little work has tested hypotheses emerging from this literature with large, diverse samples using non-self-report indicators of negativity bias. In the present study, we attempt to fill this gap. We examine the relationship of negativity bias to political ideology in three large national samples using five distinct behavioral measures of negativity bias. We also examine the association of these measures to four of the most commonly utilized self-report measures of personality. Across a wide range of tests, we find little support for the negativity bias hypothesis.
WORD COUNT: 9974
Negativity Bias and Ideology, p.2
What factors determine attraction to the political left and right? Theories highlighting stable individual differences in personality now occupy an important place beside classic sociological and economic approaches focused on individual and group interests (e.g., Aarøe,
Petersen, and Arceneaux 2017, Gerber et al. 2010, Hibbing, Smith, and Alford 2014, Jost et al.
2003, Jost 2017, Mondak 2010). While diverse, much of this literature emphasizes traits related to negativity bias—a tendency to pay more attention, and give more weight, to negative (relative to positive) stimuli in decision making. In general, a focus on negative potential outcomes creates a disposition to favor existing arrangements and certain outcomes over mixed gambles
(Kahneman, Knetsch, and Thaler 1991, Tversky and Kahneman 1979). In the political realm, social order and the status quo provide predictable outcomes, while reform and change are risky prospects with the potential for either gain or loss. In turn, negativity bias creates a preference for right-wing policies that promote stability in social and economic arrangements (Jost et al. 2003).
In this view, there is an enduring political conflict between two “primal mindsets”: one focused on order and stability and one on reform and progress (Hibbing, Smith, and Alford 2014, 297).
Evidence for this theory tends to fall into one of two categories. First, many studies examine the association of political ideology to self-reported personality traits related to negativity bias and its implications, such as the need for cognitive closure, threat and disgust sensitivity, and openness to experience (Federico and Malka 2018, Jost 2017). This work often, and increasingly, relies on large samples that are reasonably representative of the broader population of interest (Aarøe, Petersen, and Arceneaux 2017, Federico and Goren 2009, Gerber et al. 2010, Johnston, Lavine, and Federico 2017, Mondak 2010). Yet such studies assume the exogeneity of traits to political ideology, as well as the unbiased reporting of such traits. If traits are instead endogenous to political preferences, or if citizens misrepresent (or systematically Negativity Bias and Ideology, p.3 misperceive) their own traits, the conclusions of such studies would be suspect. And there is reason to be concerned. Recent work suggests that citizens exaggerate the degree to which they possess traits prototypical for their political in-group (Ludeke, Tagar, and DeYoung 2016) and variation in traits over short periods of time may be driven by political variables (Boston et al.
2018). It is thus critical to supplement existing work with research designs with less inferential ambiguity.
And indeed, a growing literature examines the association of ideology to revealed indicators of negativity bias (Hibbing, Smith, and Alford 2014). Such studies include decision making under ambiguity (Shook and Fazio 2009), physiological responses to threatening stimuli
(Oxley et al. 2008, Smith et al. 2011a), and brain imaging studies focused on sensitivity to threat and uncertainty (Ahn et al. 2014, Amodio et al. 2007, Kanai et al. 2011). This work has the virtue of bypassing self-report. In many cases, there is also a strong argument for exogeneity, as indicators of negativity bias are derived from apolitical stimuli and often involve uncontrolled processes (but see Jost et al. 2014). On the downside, the procedures involved, especially in physiological studies, often require substantial resources in time and money, and the physical presence of participants, to employ. This limits the size and representativeness of samples. For example, the seminal study of Oxley et al. (2008) on threat sensitivity surveyed only 48 individuals local to the university.
The upshot is that the literature is densely populated with studies supporting the negativity bias hypothesis, but there is a dearth of research with both of two critical characteristics: (1) large, diverse samples, and (2) revealed, rather than self-reported, traits. The primary aim of the present study is to fill this gap. We examine the negativity bias hypothesis using five distinct behavioral measures in three national samples with over 4,000 respondents. Negativity Bias and Ideology, p.4
Relying on very different operationalizations of negativity bias, we increase the generalizability of our study, and ensure that any conclusions drawn are not specific to a single measurement approach. The large size of our sample also allows for more precise estimates of the relationship of negativity bias to ideology. Whereas small-sample studies tend to focus on statistical significance at the expense of effect size, we can more clearly rule out important subsets of the parameter space.
We also contribute to the literature in three other ways. First, we provide the first examination of the relationship between political ideology and loss aversion, which is perhaps the most intuitive conceptualization of negativity bias, and one that ties the literature to behavioral decision theory. We apply three methods for eliciting individual differences in loss aversion, two for decisions under risk and one for ambiguity (Abdellaoui et al. 2016, Tanaka,
Camerer, and Nguyen 2010, Toubia et al. 2013). Second, we provide the first large-sample examination of the association between behavioral indicators of negativity bias and the most commonly studied self-report personality traits in the literature on ideology: need for closure, openness to experience, conservation versus openness to change, and authoritarian childrearing values. Strong relationships between behavioral measures of negativity bias and self-report measures of personality would provide evidence for a possible pathway through which basic attentional and evaluative dispositions operate, namely, core values and traits (Smith et al.
2011b). Finally, we explore the relationship of negativity bias to multiple dimensions of left- right political preferences, including political identity and policy attitudes in both the social and economic domains. We also examine the moderating role of political engagement for these relationships. Recent work suggests variation in the relationship of traits to ideology across different manifestations of left and right and across engagement (Feldman and Johnston 2014, Negativity Bias and Ideology, p.5
Hibbing, Smith, and Alford 2014, Johnston and Wronski 2015, Malka et al. 2014, Terrizzi Jr,
Shook, and Ventis 2010), and we extend this line of research here.
Background Theory and Research
Contemporary theorizing on personality and political ideology begins with the authors of
The Authoritarian Personality (Adorno and Frenkel-Brunswik 1950, Frenkel‐Brunswik 1949), who argued that right-wing preferences are rooted in psychological insecurity and cognitive rigidity. While this work has been heavily criticized (Altemeyer 1981, Brown 2004, Duckitt
1989), the core thesis has been continuously renewed since its publication, albeit with non-
Freudian theoretical foundations (Altemeyer 1988, Duckitt 2001, Feldman and Stenner 1997,
Kruglanski et al. 2006, Wilson 1973). The last two decades have witnessed a surge in interest in these ideas, in large part due to the work of Jost et al. (2003). In their wide-ranging review of the literature, Jost and colleagues theorize that there are at least two aspects of political conflict that transcend time and place: orientations toward change and inequality. They argue that right-wing politics is defined by resistance to the former and acceptance of the latter. In a meta-analysis of over eighty studies, they find these broad orientations are associated with both sensitivity to threat and the aversion to uncertainty and ambiguity that it entails (see Jost 2017 for an updated meta-analysis). As Jost and Hunyady (2005, 262) argue, “There is a good match between needs to reduce uncertainty and threat and system justification, because preserving the status quo allows one to maintain what is familiar and known while rejecting the uncertain prospect of social change. For many people, the devil they know seems less threatening and more legitimate than the devil they don’t.” In this view, threat sensitivity creates a preference for sure-things, and this need is best met by right-wing values, policies, and candidates that preserve status quo Negativity Bias and Ideology, p.6 values and socioeconomic inequalities (e.g., Jost et al. 2017, Landau et al. 2004, Thórisdóttir and
Jost 2011).
This thesis is isomorphic with the claim from behavioral decision theory that greater sensitivity to negative relative to positive outcomes creates a preference for the status quo and certain prospects over mixed gambles, even when the latter substantially exceeds the former in expected value (Kahneman, Knetsch, and Thaler 1991, Köbberling and Wakker 2005, Tversky and Kahneman 1992). When losses loom larger than gains, people will be drawn to options that avoid downside risks, unless the gains are very large and/or very certain. This “negativity bias” hypothesis, as applied to politics, has been stated most explicitly by Hibbing, Smith, and Alford
(2014). They review a growing body of research that suggests right-wing preferences are associated with several psychological and physiological characteristics that evince greater attention and responsiveness to negative than positive stimuli. Most notably, recent work has found associations of political ideology with physiological responsiveness to negative and disgusting stimuli (Ahn et al. 2014, Oxley et al. 2008, Smith et al. 2011a), greater grey matter volume in areas of the brain associated with the fear response (Kanai et al. 2011), attentional biases to negative stimuli (Dodd et al. 2012, McLean et al. 2014, Mills et al. 2014), and a tendency to avoid negative potential outcomes in a decision making task under ambiguity (Shook and Fazio 2009) (for an extensive review, see Federico and Malka 2018).
While acknowledging extensive empirical support for the basic claim of a link between negativity bias and right-wing views, recent work has extended the literature in several ways.
First, several studies demonstrate that the relationship between related traits and political ideology is conditional on political engagement and expertise (Federico, Fisher, and Deason
2011, Federico and Goren 2009, Johnston, Lavine, and Federico 2017, Jost, Federico, and Napier Negativity Bias and Ideology, p.7
2009). Citizens who pay more attention to politics, and know more about it, show stronger relationships between self-reported traits and political attitudes. This is because they are better able to identify the positions that match their dispositions and are more likely to adopt the positions of elite actors who share their broader worldview.
Second, scholars have argued that relationships between traits and ideology vary across ideological dimensions (Federico and Malka 2018, Feldman and Johnston 2014, Johnston,
Lavine, and Federico 2017, Malka et al. 2014). Despite growing constraint over time
(Abramowitz 2010, Freeze and Montgomery 2016), mass belief systems are multidimensional in the sense that they are well-described by models positing distinct, correlated dimensions of values and policy preferences, with social and economic dimensions most prominent (Baldassarri and Gelman 2008, Feldman and Johnston 2014, Layman and Carsey 2002, Lupton, Myers, and
Thornton 2015, Treier and Hillygus 2009). As these two dimensions are correlated, it is possible that negativity bias may be the foundation that links them through time (Jost, Federico, and
Napier 2009). However, it is also possible that negativity bias influences social and economic preferences in distinct ways. Indeed, recent work suggests this is the case. In unconditional models, traits related to negativity bias (e.g., need for closure) are more strongly associated with social than economic preferences (e.g., Feldman and Johnston 2014, Terrizzi, Shook, and
McDaniel 2013, see Federico and Malka 2018 for a review) and occasionally promote left-wing economic views (Arikan and Sekercioglu 2019, Crowson 2009, Johnston 2013). Additionally, in models that condition on political engagement, these traits promote right-wing views among the engaged, but often left-wing views among the unengaged (Jedinger and Burger 2019, Johnston
2018, Johnston, Lavine, and Federico 2017, Malka et al. 2014). Negativity Bias and Ideology, p.8
Finally, scholars have proposed that basic physiological and cognitive differences in negativity bias operate through more proximate psychological constructs such as personality traits and basic values (Smith et al. 2011b, 372). For example, Caprara and Zimbardo (2004) propose a model in which citizens identify with elites who share their core personality dispositions—what they refer to as “dispositional heuristics.” In this view, value priorities (e.g., security; Schwartz 1992) and self-conceptions in terms of traits (e.g., openness to experience;
McCrae and Costa 2003) serve as guides that help citizens navigate a complex political space and choose among diverse alternatives.
Hypotheses
Few studies directly test these claims with revealed measures of negativity bias in large, diverse samples. Moreover, there have been no attempts to test intermediate mechanisms linking negativity bias to ideology through personality dispositions, such as values and traits. The present study is intended to fill this gap in the literature. We thus explore the following hypotheses:
1. Indicators of negativity bias will be associated with right-wing preferences in general, but
will be more strongly associated with general left-right identity and social preferences
relative to economic preferences.
2. Indicators of negativity bias will be associated with self-reported personality traits
commonly used to predict right-wing ideology, including: (low) openness to experience,
the prioritization of conservation values over openness to change values, authoritarian
childrearing values, and the need for non-specific cognitive closure.
3. Political engagement will moderate the relationship between indicators of negativity bias
and political ideology. Negativity Bias and Ideology, p.9
The remainder of the paper proceeds as follows. We first describe our data sources and five distinct measures used to operationalize negativity bias. We then turn to our primary results, which are followed by a series of robustness checks. Finally, we discuss the implications of our results for the broader literature.
Data and Methods
Our data consist of three U.S. samples collected during three time periods, one in January of 2015, one in August of 2018, and one in October/November of 2019. The 2015 sample was collected by ClearVoice research through a contract with Qualtrics Panels (see supplemental appendix for further information about all samples). Steps were taken to ensure data quality: respondents were terminated from the survey if they failed either of two attention checks1 or if they failed to complete at least 90% of all survey items in the first half of the survey. The 2018 and 2019 samples were collected through Lucid’s Fulcrum Exchange (now Marketplace) with a restriction of the sample to only non-mobile-device users. We also terminated respondents who failed either of two attention checks in the first part of the survey.2 Respondents completed one
(and only one) of five different measures of negativity bias, which are described in the following
1 The first attention check was a survey item that appeared to be a standard question about self- perceived knowledge in politics but embedded an instruction to select the response option “No knowledge at all.” The second was a multiple-choice item concerning the position currently held by Barack Obama.
2 In the first check, respondents had to accurately select two photos of roads (from a set of six) that contained a stop sign. The second, a multiple-choice question, asked respondents to identify the current President of the United States. Negativity Bias and Ideology, p.10 section. Specifically, all respondents to the 2015 sample completed measure 3; all respondents to the 2019 sample completed measure 4; and respondents to the 2018 sample were randomly assigned to either measure 1, 2, or 5. After describing these measures in detail, we turn to our dependent variables, which are identical across the 2018 and 2019 samples and very similar to the 2015 sample.
Measure 1: Cognitive Accessibility of Threatening Concepts
Our first measure of negativity bias was randomly assigned to a subset of the 2018 Lucid sample and extends previous work on the psychological roots of authoritarianism. This work uses a lexical decision task (LDT) to measure negativity bias (Lavine et al. 2002). Respondents first completed a survey of basic demographics, political interest and knowledge, then the LDT, and then a survey of personality and political attitude and identity measures.
In an LDT, respondents are exposed to a series of trials. In each trial they are asked to decide, as quickly as possible while being accurate, whether a string of letters is a legal English word or a non-word. Speed and accuracy in the legal word trials are, theoretically, measures of the relative accessibility of the concepts represented by the words. That is, concepts which are highly accessible are more quickly activated by exposure to their textual representation and are thus more quickly judged to be words compared to less accessible concepts (Higgins 1996,
Lodge and Taber 2005, Meyer and Schvaneveldt 1971). We use the difference in the average response time to positive relative to negative words as a measure of the relative accessibility of negative concepts, and thus of negativity bias.
Respondents completed a total of 60 trials: 10 each for positive, negative, and neutral words, and 30 for non-words. Words for each trial were randomly sampled without replacement.
Our final measure was calculated similarly to the D-score in research on the implicit association Negativity Bias and Ideology, p.11 test (Greenwald, Nosek, and Banaji 2003). We exclude all responses below 200 milliseconds or above 5000 milliseconds3 and all respondents with less than 80% correct responses.4 We then calculate, for each respondent, the difference between average response time to positive words and average response time to negative words and divide this difference by the respondent- specific standard deviation of latencies to both types of targets combined. The larger the value, the greater the variance-adjusted relative response time to positive words and thus the higher the relative accessibility of negative words. We detail the procedure and calculations in the supplemental appendix. To incentivize effort and reduce measurement error, we informed respondents that the top ten performers in the LDT study would be eligible to receive a $10.00
Amazon.com gift card (winners were paid after all data was collected).
Measure 2: Attentional Biases to Negative Stimuli
Our second measure was randomly assigned to a subset of the 2018 Lucid sample and operationalizes negativity bias as attentional biases to negative relative to positive stimuli.
Specifically, we adapt recent work by McLean et al. (2014) using the flanker task. On each of 30 trials, respondents are presented with a series of three images arrayed horizontally. The middle image is the target and the flanking images to the left and right are distractors. The goal of each trial is to identify, as quickly as possible while being accurate, whether the target image is
3 About 6% of respondents who participated in this task had one or more trials outside of this window, but no one had more than three problematic trials. Less than 1% of all trials were excluded on this basis.
4 About 5% of the sample that participated in this task was excluded on this basis. The median accuracy across respondents was 95%. Negativity Bias and Ideology, p.12 positive or negative. McLean et al. use angry and happy faces; however, we substitute three negative and three positive images drawn from the International Affective Picture System database (IAPS) and the Geneva Affective Picture Database (GAPED). The images in these databases are normed for both valence and arousal and IAPS pictures have been used in previous work on negativity bias and ideology (Ahn et al. 2014). The specific images used in our study were chosen for being extreme in terms of valence. We used three very positive images (flowers, a human baby, and a baby seal) and three very negative images (an aimed gun, a dirty toilet, and a spider). We provide a table with information on all images used in the supplemental appendix.
McLean et al. (2014) test the negativity bias hypothesis by first defining a generic flanker effect and then examining this effect across ideology for negative and positive targets separately.
In their definition, a flanker effect is the difference in average response times comparing trials where all three images are of the same valence (e.g., all three are negative) to trials where the flanking images are conflicting in valence with the target (e.g., a negative image flanked by two positive images). They argue that conflict between target and flanker valence slows response time. However, if negative images strongly capture attention, this flanking effect should be small
(or even reversed) for negative targets and large for positive targets. That is, negativity bias implies a comparative measure such that target-incongruent flankers are more distracting for positive relative to negative targets. A straightforward measure is thus the difference in the flanking effects comparing negative to positive targets. We again exclude all responses below
200 milliseconds or above 5000 milliseconds5 and all respondents with less than 80% correct
5 About 7% of respondents who participated in this task had one or more trials outside this window. Negativity Bias and Ideology, p.13 responses.6 We calculate this measure using a D-score-like procedure, as with measure 1. This is detailed in the supplemental appendix. We again incentivized performance by informing respondents that the top ten performers would be eligible to receive a $10.00 Amazon.com gift card.
Measure 3: Loss Aversion in Decision Making under Risk
Our 2015 Qualtrics study operationalized negativity bias as the loss aversion parameter in prospect theory for decision making under risk (that is, with known outcome probabilities).
Conceptually, loss aversion means that losses loom larger than nominally equivalent gains.
Within cumulative prospect theory (CPT; Tversky and Kahneman 1992), loss aversion is formalized as a weighting parameter (�) on the value function for negative potential outcomes: