Personality and Individual Differences xxx (xxxx) xxxx
Contents lists available at ScienceDirect
Personality and Individual Differences journal homepage: www.elsevier.com/locate/paid
Individual-level differences in negativity biases in news selection Sarah Bachledaa, Fabian G. Neunerb, Stuart Sorokaa,c, , Lauren Guggenheimc, Patrick Fournierd, Elin Naurine ⁎
a
Department of Communication and Media, University of Michigan, United States School of Politics and Global Studies, Arizona State University, United States Center for Political Studies, University of Michigan, United States d Department of Political Science, Université de Montréal, Canada e Department of Political Science, University of Gothenburg, Sweden b c
ARTICLE INFO
ABSTRACT
Keywords: Political communication Personality differences News consumption Negativity bias
Literatures across the social sciences highlight the tendency for humans to be more attentive to negative information than to positive information. We focus here on negativity biases in news selection (NBNS) and suggest that this bias varies across individuals and contexts. We introduce a survey-based measure of NBNS which is used to explore the correlates of negative news bias in surveys in the U.S., Canada, and Sweden. We find that some respondents are more prone to NBNS than others. There is evidence of contextual effects, but panel data suggests that some of the individual-level differences persist over time. NBNS likely reflects some combination of longterm personality differences and short-term situational factors, and is systematically related to a number of economic and political attitudes.
There is a considerable body of work illustrating negativity biases in both individual- and aggregate-level political behavior. People weigh negative traits more heavily than positive traits when assessing political candidates (Lau, 1982; Klein, 1991; Holbrook et al., 2001), and economic downturns have stronger effects on public perceptions of the economy than upturns (Soroka, 2006), for instance. The prominence and strength of negativity comes as no surprise to scholars of political communication, of course – there are vast bodies of research chronicling the tendency for media coverage to be negative (e.g., Patterson, 1994), and recent findings making clear the tendency for news consumers to be attracted to negative information (e.g., Trussler and Soroka, 2014). These findings are related to a broader pattern of behavior, extending well beyond politics, for individuals to weigh negative information more heavily than positive information. That said, drawing on past work (cited below), we suspect that the general inclination towards negativity is neither universal nor uniform. This paper accordingly explores individual-level heterogeneity in negativity biases, focusing on news selection, to illustrate the likely significance of negativity biases as an enduring source of variation in individuals’ responses to the political world. Why should we be concerned about individual-level differences in negativity biases? A growing body of research highlights the ways in which heterogeneity in news consumption matters for media effects as well as political attitude
⁎
formation (Stroud, 2007; Stroud, 2010; Knobloch-Westerwick, 2012; Garrett and Stroud, 2014). This is especially important in today's media environment, in which selectivity in consumption – even at the story level – is increasingly prevalent (Garrett, 2009). In short, if preferences for or selection of negative information vary depending on the person or context, then different people will be affected by news in different ways at different times. Given the growing significance of this heterogeneity, we believe there is a need for a measure of negativity biases focused directly on this variation in news consumption. Of course, we do not view the quantity examined below as being relevant only to the news. We expect negativity biases in news selection to at least partly reflect a negativity bias that is generalizable beyond this domain. To be clear: we suspect that there is a domain-general bias rather than just a domain-specific bias in information processing. As we shall see, we cannot easily adjudicate between these possibilities in our data. That said, even if negativity biases in news selection (NBNS) are both domain-specific and context-dependent, we view them as being of special relevance for understanding media effects and political attitudes. Much of what we know about the political and economic worlds is learned from, and shaped by, media content. Understanding the nature of negativity biases in the context of news selection is thus of real importance in politics. We will first review the potential significance of negativity biases
Corresponding author. E-mail address:
[email protected] (S. Soroka).
https://doi.org/10.1016/j.paid.2019.109675 Received 30 April 2019; Received in revised form 2 October 2019; Accepted 21 October 2019 0191-8869/ © 2019 Elsevier Ltd. All rights reserved.
Please cite this article as: Sarah Bachleda, et al., Personality and Individual Differences, https://doi.org/10.1016/j.paid.2019.109675
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
generally, and negativity biases in news selection in particular, as a partly durable individual-level characteristic, relevant not just to news consumption but to a wide range of attitudes. We then use a surveybased measure of respondents’ inclination to select negative over positive news content. This method serves to highlight the potential significance of this inclination on how individuals perceive the world around them. Using survey data from the U.S., Canada and Sweden, we find evidence that NBNS correlates with stable characteristics such as gender, political ideology, and Big 5 personality traits. We examine the durability and change in NBNS amongst panel respondents using surveys administered three years apart and demonstrate the relationship between NBNS and a range of political and economic attitudes crosssectionally and over time. We take these results as evidence of the potential value of a measure of NBNS, as a stable domain-specific trait, relevant for a range of political-attitudinal outcomes and thus worthy of further testing. We conclude with a discussion of the ways in which individual-level variation in negativity biases might be used to answer important questions in political communication and related fields.
and skin conductance responses to negative stimuli (Grabe and Kamhawi, 2006; Rohrmann, Hopp and Quirin, 2008). Relatedly, while in general women show less bias towards selecting negative information, they are more attentive to it when exposed (Soroka et al., 2016). Political ideology is also correlated with responsiveness to negative information. Dodd and colleagues (2012) find that participants who hold politically conservative attitudes have stronger electrodermal responses to negative images than positive images. This may be due to a heightened awareness for identifying potential threat (e.g., Hibbing et al., 2014), or a heightened sensitivity to instability or insecurity (Buck, 2014). The underlying mechanism of differences in negativity bias may be less about political ideology per se, and more about predispositions or personality. Work in psychology supports this notion, finding that extroversion vs. introversion, approach vs. avoidance temperament, and emotionality impact people's sensitivity to positive and negative stimuli (e.g., Pietri et al., 2013, Buck, 2014). It may similarly be the case that the Big 5 personality traits are correlated with negativity biases. And although no prior work examines this possibility, past work suggests some expectations: (a) insofar as openness is associated with liberalism (e.g., Mondak, 2010), and liberalism is associated with lower negativity biases (e.g., Hibbing et al., 2014), openness may be negatively correlated with NBNS; (b) the opposite may be true of conscientiousness, which is associated with conservatism (Mondak, 2010) and thus by implication higher negativity biases, and (c) based on recent work that finds lower negative partisan affect for those higher in agreeableness and extraversion (Webster, 2018), each of these traits may be negatively associated with NBNS. Regardless, what is most critical for our purposes is that reactivity to negative information may be a partly stable characteristic, relevant to how individuals select and process political information and develop political attitudes and preferences. Just as past work has highlighted the importance of individual differences such as personality (e.g., Feldman, 2013; Mondak et al., 2010), needs (e.g., Jost et al., 2003), and personal values (e.g., Caprara et al., 2006) on political outcomes, we expect negativity biases – regarding news selection in particular – to matter for a range of politically relevant attitudes and behaviors. That said, while we expect negativity biases to be a stable individual-level characteristic, we also suspect that they may shift over time and context, conditioned in part by the nature of individuals’ current information environments. A small body of work points in this direction. There is, for instance, long-standing literature on impression formation that highlights the role of contrast effects (e.g., Helson, 1964) and/or novelty (Fiske, 1980) in information processing. Simply put, individuals are more likely to be attentive to and/or respond to information when that information is at odds with expectations. Insofar as expectations change based on incoming information, we should expect attentiveness to negative information to decrease ceteris paribus when expectations are more negative. This is the crux of the model put forward by Lamberson and Soroka (2018), in which attentiveness to negative information is driven by “outlyingnesss” rather than negativity. There is, of course, no reason that variation in negativity biases must be either durable or contextual, and it almost certainly is both. Existing research on bias towards negative information points to an interaction between individual-level and contextual factors where interest in negative information is concerned, including Lilienfeld and Latzman's (2014) finding that although conservatives are more responsive to negative information on average, both conservatives and liberals respond to negative information when it poses a threat to their partisan identity; or Federico, Johnston and Lavine's (2014) finding that evidence of negativity biases will be conditional on political engagement. A measure of negativity bias, particularly where news selection is concerned, may thus capture both “current evaluative/contextual” and “underlying/pre-dispositional” components. Our intention here is not
1. Durable versus context-driven individual-level variation in negativity biases We use the term ‘negativity biases’ to refer to a tendency of humans to prioritize and/or react more strongly to a piece of negative information in contrast with an equivalently positive piece of information. This inclination has been especially well documented in psychology (for reviews, see, e.g., Baumeister et al., 2001; Rozin and Royzmann, 2001) and behavioral economics (e.g., Kahneman and Tversky, 1979).1 Research on negativity biases in politics has often examined how these biases function on average. These studies have found that people tend to learn more from negative political information than from positive information (e.g., Bradley, Angelini and Lee, 2007), and have stronger, more sustained reactions to negative news (e.g., Soroka and McAdams, 2015). Recent research nevertheless highlights both individual and contextual variation in attentiveness to negative information. Ito and Cacioppo (2005) highlight the asymmetry between two countervailing tendencies, a “positivity offset” at low levels of input, and a “negativity bias” as input increases; and they find individual-level differences in both relevant to attitude formation. Work by Fazio and colleagues (Pietri, Fazio, and Shook, 2013; Rocklage and Fazio, 2014) has also identified, through a series of lab-experimental games, individual-level differences in the weighting of negative and positive information. Further research by Ito and Cacioppo (2005) and Norris et al. (2011) suggests that these individual differences remain consistent over time, or at least they remain relatively constant during exposure to different negative stimuli. In sum, there is evidence that individuals appear to have at least partly durable predispositions towards negative versus positive information. What are the correlates of these durable predispositions? Gender appears to be one source of difference in negativity biases. Women and men tend to process information in different ways, attending to different aspects of a message (e.g., Darely and Smith, 1995). Techniques such as fMRI and steady-state probe topography suggest that negative stimuli activate different parts of the brain in women and men (Wrase et al., 2003; Kemp et al., 2004; see Stevens and Hamann, 2012). There is also evidence that men and women have different heart rate 1 We draw here on past definitions of “negativity,” where it is defined either analytically or affectively: “Analytical negativity refers to comparatively objective measures of negativity (i.e., many people dying, or increasing unemployment), while affective negativity is more subjective (i.e., a policy that we may disagree with for ideological reasons)” (Lamberson and Soroka, 2018: 2-3).
2
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
to adjudicate between these various possibilities, but rather to focus on the latter component, and in so doing make a case for considering negativity biases in news selection as a potentially durable driver of political attitudes. Our objective is to examine the way in which differences in preference for negative news varies between individuals, and thus could have a meaningful impact on the effect of media consumption on political attitudes. We focus below on several hypotheses:
the Swedish Citizen Panel, fielded in 2015 and 2018. (Full details on all surveys are provided in the Appendix.) Our use of these three surveys was based on several considerations. In the case of the Swedish Citizen Panel, there are advantages of having access to a range of questions from waves long before we measured NBNS and of being able to go back to the same panel respondents more than once. In the case of the surveys in the U.S. and Canada, both provide useful comparative tests of the measure; and because they are fielded as part of campaign-period surveys, they also include some of the political and economic variables that will be important to the analyses below. As for the NBNS measure itself, we use a simple battery of five questions. The first of these questions is: “Imagine that you are going to read a news story in order to learn something interesting, important or useful about the [economy/ environment/ health care/ politics/ foreign affairs]. You have four headlines from which to make one selection. Which of the following would you read?” The respondent is then given four headlines, and they select one. After that, they receive the question again, but for a different topic. There are, as noted above, five topics in total. The ordering of topics is randomized, and so too are the headlines within each topic question. The headline groupings always include two positive headlines and two negative headlines. The resulting measure of NBNS, ranging from 0 to 5, is a count variable, capturing the number of questions for which the respondent selects a negative headline. We believe our approach to measuring negativity bias has high external validity. Asking respondents to review a list of headlines and select which one they would be most interested in reading is a method that mirrors news selection in today's media environment. It has become very common for news and social media websites, and their adjoining mobile applications, to structure their landing page as a list of headlines or short posts. A user's first opportunity to engage with digital content usually involves scrolling through these brief blurbs of information, from which they can opt-in to read more via clicking (or sometimes “expanding”). Thus, measuring respondents’ draw to read a story with a negative versus positive headline likely mimics a media selection process they experience in everyday life. This is absolutely true for digital media; and is roughly similar even for hard-copy newspapers where headline selection is also an important part of news consumption. The purpose of using a battery of five questions, each on a different topic, is to build a measure which captures a general tendency to select negative headlines – that is, a measure for which the particular interests that respondents may have in one domain or another play a very small part in the overall score. The headlines themselves are shown in Table 1, which also shows the proportion of respondents selecting each headline across each of our three studies. Note that due to time constraints, we were able to include only three questions in the CES; in both the U.S. and Sweden, however, we were able to include the full five-question battery. One early concern might be that our results are driven by one particularly interesting or uninteresting headline. It does not appear as though this is the case in any domain, in any country. There is no question receiving less than roughly 9% of responses and there is no question receiving more than roughly 57% of responses – and most are well within these boundaries. In every instance, then, there is reason to believe that results are not driven by the strengths or weaknesses of any one headline. The purpose of randomization is to factor out the impact of (a) the ordering of questions, where headline selection in later questions may be affected by headline selection in prior questions, and (b) the ordering of headlines, where we might expect respondents to select the first headline more than subsequent headlines. Our tests suggest that neither of these is a major problem, however. Diagnostic results suggest no link between headline ordering and headline selection; see Appendix Fig. 1. Even so, we randomize both questions and headlines across all three surveys. We also confirm that the tone of the headlines is correct – i.e., that the headlines intended to be negative or positive were indeed perceived as negative or positive – through a separate coding exercise in
H1. Durability of individual-level variation in negativity biases will be reflected by correlations with demographic, attitudinal and personality measures. Based on past work, we expect the following: H1a. NBNS will be lower for women than for men. H1b. NBNS will be positively correlated with conservative political ideology. H1c. NBNS will be correlated with the Big 5 personality traits; specifically, NBNS will be negatively correlated with openness, agreeableness and extroversion, and positively correlated with conscientiousness. H2. Durability of individual-level variation in negativity biases will be further reflected by correlations between NBNS measured across the same individuals at different points in time. H3. Negativity biases in news selection will, in turn, be correlated with more negative political and economic perceptions (i.e., government satisfaction and economic sentiment). 2. A survey-based measure of negativity biases in news selection The most frequently-used measures of negativity biases are based on either self-reports or lab studies. Each of these presents some challenges. The principal drawback of self-reports is that participants have difficulty reporting on their own internal states and often struggle to accurately report the emotions or arousal felt in response to different stimuli (e.g., Robinson and Clore, 2002). Participants may also only remember a narrow amount of what they experience. Moreover, some attitudes and feelings can stay below the level of consciousness, making it hard for individuals to explicitly report them (see, e.g., Hofmann et al., 2005). There may also be some shame in reporting one's true feelings or level of arousal in response to certain images or content (Granberg and Holmberg, 1991), and certain groups may be socialized to under- or over-report their attitudes and feelings. This dynamic may well be at play where attentiveness to negative news is concerned. There is, after all, evidence that survey respondents say that news should be more positive, even as they choose to read negative rather than positive news content (Trussler and Soroka, 2014). Lab-based measures of negativity biases include both psychophysiological variables (especially skin conductance) and lab-experimental games (such as work using Beanfest; i.e., Pietri, Fazio, and Shook, 2013; Rocklage and Fazio, 2014). Both typically need to be collected in a lab setting, which can be costly and time-consuming. Most importantly, lab experiments greatly restrict sample reach. These constraints will almost certainly pose problems for work interested in negativity as moderator of media effects on the public at large. The advantages of a brief and implicit measure of negativity biases, focused in particular on news consumption, and deployable in a simple online survey, are relatively clear. Does the direct or moderating impact of negativity biases fluctuate during periods of political intensity, such as campaigns? Are they impacted by the general political climate, country or culture? Answering these questions requires a measure that allows for broader samples (potentially cross-national), that reduces cost burdens (particularly when repeatedly measured over time), and that allows for quicker distribution during politically-charged periods or events. The survey-based method used here has these advantages. We explore negativity biases in news selection (NBNS) using three different surveys. The first is a campaign-period panel survey, fielded in the U.S. in 2016; the second is the online wave of the 2015 Canadian Election Study; and the third is the sixteenth and twenty-ninth waves of 3
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
Table 1 Headline selections, by country.
Table 2 Modeling NBNS. Sweden
Canada
USA
10.09% 16.26% 35.52% 38.12%
NA NA NA NA
14.23% 28.97% 18.64% 38.16%
25.20% 33.38% 21.24%
14.67% 57.49% 10.91%
13.98% 47.48% 14.86%
20.19%
16.93%
23.68%
11.76% 18.85%
25.75% 25.31%
31.36% 22.67%
16.22% 53.17%
20.54% 28.40%
15.24% 30.73%
30.22%
22.43%
22.42%
19.75%
23.99%
21.41%
9.41%
22.86%
18.51%
40.62%
30.72%
37.66%
12.99% 33.87% 16.35% 36.80%
NA NA NA NA
19.65% 35.89% 21.91% 22.54%
DV: NBNS Sweden
a
Politics Assistance in Sight for Congressional Leadership Congress Fumbles Again Parties Succeed in Rebuilding Bases of Support Support for Government at All-Time Low Economics Employment Up from Last Month Experts Deeply Worried about Rising Cost of Living Has Employment Already Peaked? Future Prospects Worsen Inflation Figures Released: Outlook is Positive Foreign Explosions Shock Diplomats Across the Middle East Foreign Leaders Convene to Improve Trade Relations Global Trade Summit Widely Criticized Positive Shift in Middle East Talks Environment Monthly Trend Suggests Improvement in Global Warming Report Suggests Rising Concerns about Rising Temperatures Scientists Offer Warnings about Depleted Ozone Layer Successful Reforestation Offers Signs of Hope Health Doctors' Healthy Eating Tips Easy Ways to Improve Heart Health Meals That Can Harm Your Health Why Are Heart Attacks On the Rise?
Age Sex (1 = Female) Education (1 = College) Income (terciles) Incumbent Support Political Interest Left-Right Ideology Pol. Interest*Ideology Big 5: Extraversion
−0.203*** (0.013) −0.035*** (0.005) −0.012* (0.006) −0.035*** (0.007) −0.054*** (0.006) 0.031** (0.011) 0.007*** (0.001) −0.004* (0.002)
Big 5: Agreeableness Big 5: Conscientiousness Big 5: Emotional Stability Big 5: Openness Constant Observations R2
a Note that the politics headlines were adjusted to refer to domestic politics, not “Congressional Leadership,” for instance, in Sweden.
0.546*** (0.012) 10,631 0.041
Canada
US
−0.045 (0.032) 0.012 (0.013) −0.007 (0.015) 0.005 (0.016) −0.023 (0.016) 0.018 (0.033) −0.010 (0.005) −0.003 (0.006) −0.011* (0.005) −0.018** (0.007) 0.014* (0.007) −0.006 (0.006) −0.007 (0.006) 0.723*** (0.053) 2,480 0.019
−0.052 (0.087) 0.026 (0.032) 0.037 (0.042) −0.013 (0.043) −0.044 (0.054) −0.058 (0.071) −0.013 (0.013) 0.005 (0.016)
0.595*** (0.091) 316 0.015
Cells contain OLS coefficients with standard errors in parentheses. * p < .05; ** p < .01; *** p < .001.
which we asked 100 Amazon MTurkers to report the “tone” of each headline on a scale from 1 to 5 (where 1 is most negative, 5 is most positive, and 3 is neutral). Results are shown in Appendix Fig. 2; they confirm that coders are readily able to identify positive and negative headlines. With these diagnostics in hand, we turn to an analysis of NBNS.
distributions lean to the right, i.e., in the direction of more rather than less negativity. These results fit with past work on negativity biases. The Swedish data, in contrast, lean in the direction of positivity. We have no particular expectations in the Swedish case given the absence of past 1work in that country. What matters most, we suspect, is that we find a good amount of variation within each of the three countries. In both Sweden and the US, roughly 15% of the sample is at the bottom (0) and top (1) of the index. In Canada, with the four-point scale, roughly 35% of the sample is at 0 and 1. We take these results as evidence that there is a good degree of within-sample variation in NBNS.
3. Individual-level variation in NBNS We begin by looking at the distribution of the NBNS measure across the three surveys, shown in Fig. 1. Both the Canadian and U.S.
Fig. 1. The distribution of NBNS, by sample. 4
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
Exploring individual-level correlates of the NBNS measure is of some importance, particularly as a test of H1a-H1c. Table 2 accordingly presents OLS models of NBNS in each of the Swedish, Canadian and U.S. surveys, where NBNS is modeled as a function of basic demographics alongside incumbent support, political interest, left-right ideology, and an interaction between interest and ideology. The interaction allows for the possibility that the connection between left-right ideology and NBNS is moderated by political interest; we expect that both ideology and interest will be positively associated with NBNS, but that the impact of each may be contingent on the other. The specification of political interest and left-right ideology variables vary across samples due to differences in survey questions. Interest is tabulated here using a binary variable that cuts each sample roughly in half based on political interest (where 1=high interest), and ideological self-placement is an 11-point left-right scale in Sweden and Canada, and a 7-point partisan identification scale in the U.S. (where higher values indicate conservative ideology). Full details on question wording are available in the Appendix. We also include a binary incumbent support variable, equal to one for respondents voting for the incumbent President (in the U.S.) or Government (in Canada and Sweden). This variable captures the possibility that respondents unhappy with the current government will be more inclined towards negative information (independent of political ideology). The Canadian survey is only one that asks the Ten-Item Personality Index (TIPI; Gosling et al., 2003) of all respondents, and so we include each of the Big 5 personality traits in this country's model. Doing so offers a test of H1c (Fig. 2). The first column of Table 2 shows results for Sweden. These are in line with our expectations for H1a and H1b. Women show significantly lower levels of NBNS than men, and both political interest and conservative political ideology are associated with higher NBNS. There also is a significant interaction between political interest and left-right ideology. Fig. 2 illustrates this interaction (i.e., the estimated impact of ideology on NBNS conditional on low versus high political interest). Amongst those who identify as being at the far right of the ideological spectrum, political interest makes little difference – NBNS is relatively high in general. Moving to the left of the ideological scale suggests a difference, however. Those who are more liberal-leaning with high political interest do not decrease in NBNS at the same rate as those with low political interest. Put differently, amongst those on the left of the
ideological scale, political interest increases negativity bias in news selection and amongst those on the right, NBNS is comparatively high regardless of interest. This is line with work suggesting a connection between right-wing ideology and other negativity biases (e.g., Hibbing et al., 2014). Results in both Canada and the U.S. are less clear than the Swedish data. There are no statistically significant coefficients in the U.S. model, which might be explained by the fact that the sample drops to just over 300 respondents with the inclusion of the interest and ideology variables.2 We do not probe those results any further here but note that in subsequent sections even the small US sample suggests connections between NBNS and a range of political economic attitudes. The Canadian sample also finds no support for H1a or H1b, but results for H1c are roughly as expected: extraversion and agreeableness are negatively associated with NBNS, while conscientiousness is positively associated with NBNS. (The impact of these traits does not change in models that include just one trait at a time; nor are there significant coefficients for demographics, interest and ideology when the Big 5 variables are excluded from the model.)3 Our exploration of the demographic and attitudinal predictors of NBNS necessarily requires some triangulation across surveys with varying advantages. Even so, we take results in Table 2 as a preliminary indication that NBNS may reveal meaningful differences in ways in which individual select news content. 4. NBNS over time To further explore the durability of the NBNS measure (H2), we examine a re-fielding of the measure three years later in the Swedish sample. This time, we fielded NBNS questions in the months leading up to the September 2018 Swedish General Election. (Note that this is in contrast to the 2015 wave, which came roughly a year after the 2014 election.) In the 2018 survey, we were able to re-interview 1,437 respondents from the 2015 wave and ask either (a) the NBNS battery in exactly the same form as in 2015, or (b) the NBNS battery, but using an entirely new set of topics and headlines. Respondents were randomly assigned to either condition (a) or (b). The new battery is listed in Table 3, which also shows the percentage of respondents selecting each headline (as in Table 1), based on the 735 respondents assigned to the (b) condition. This new battery removes the two explicitly political headline batteries. Our aim in this instance is to capture a variant of NBNS that is (perhaps) more independent of partisanship. Fig. 3 shows the distribution of the NBNS measure using respondents from the 2018 wave. The first panel shows the distribution from the 2015 survey but using the subsequently re-tested 2018 respondents only; the second panel shows the distribution from 2018 using the sample that received the original headlines; the third shows the distribution from 2018 using the sample that received the new headlines. There is little difference in the distributions in the first two panels – using the same headlines produces roughly the same distribution in NBNS in 2018 as it did in 2015. The new headlines, in contrast, produce a distribution that leans further towards negativity. That on its own is not a problem, of course – the central issue is the extent to which the 2018 indices are correlated with the 2015 index. But the difference does point to the fact that the NBNS measure is not entirely independent of 2 It is the interest variable that matters most in this regard, as it is based on questions that were not asked of the entire sample. That said, dropping interest increases the sample size but does not reveal any other statistically significant effects. 3 The Swedish survey actually includes the TIPI scale, but just for a small (N=~350) subset of respondents. Even so, there are statistically significant negative bivariate correlations between extraversion and NBNS, as well as agreeableness and NBNS, in those data. We take this as a confirmation that results in the second column of Table 2 are not unique to Canada.
Fig. 2. Political ideology and NBNS, moderated by political interest, Swedish data. 5
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
mostly a matter of a one-headline difference. We believe that it is reasonable to anticipate moderate rather than strong correlations given that our measure is not a nuanced intervallevel measure built from dozens of survey items, but rather a simple 4item count variable. And we expect that there is over-time variation, resulting from a changing news environment. Furthermore, and like for any survey measure, we expect some noise in the measurement of NBNS. NBNS is at least partly responsive to the topics of the headlines, and likely the information context in which the data are collected. In what ways is the ‘information context’ in 2018 different from the one in 2015? As noted above, the 2018 data are from the midst of an election campaign, while the 2015 data are not. The parties in government are the same, however; and macroeconomics did not change dramatically from 2015 to 2018. Of course, for any one individual, the information context is a function of much more than national-level political and economic variables – it is, perhaps primarily, the product of more regional and personal factors as well. Accounting for the contextual factors that shift NBNS likely requires further work. In the meantime, our results highlight what appears to be a notable degree of durability in NBNS over time. We take this as preliminary evidence in support of H2.
Table 3 Attentiveness to headlines, Swedish 2018 data. Sweden Education Teachers Concerned as Class Sizes Exceed Capacity Early Warning Signs of Problems with Math and Science Curriculum Student Projects Bring Solutions to Local Community Issues School Board Approves Expanding Resources for Teachers Technology Smart Watches Raise Serious Concerns about Privacy Microchip Manufacturers Failing Workplace-Safety Standards New Autonomous Vehicle Company Brings Mobility to the Disabled Artificial Intelligence App Promises to Boost Work-Life Balance Economy Falling GDP Expected to Slow Economic Growth Recent Outsourcing Incentives Hurting Middle-Class Firms See Signs of Economic Recovery Record Increases in Consumer Confidence Science Funding for basic research at 10-year low Damaged satellite major setback for space research Scientists Hail Breakthrough in Flu Research NASA Discovers New Galaxy with Advanced Telescope Culture Death of the Orchestra? Theaters Concerned with Declining Ticket Sales Thousands of Historic Paintings Lost in Recent Hurricane Instagram Opens Opportunities for Budding Artists Study Shows Teen Reading Interest at All Time High
25.65% 26.60% 29.60% 18.14% 33.02% 8.83% 41.30% 16.85% 35.65% 26.53% 17.69% 20.14% 21.69% 2.05% 48.02% 28.24% 9.80%
5. NBNS and political attitudes
39.46% 8.98% 41.77%
We now examine the broader significance of NBNS, in particular, the relationship that NBNS may have with a range of political and
Fig. 3. The distribution of NBNS, Swedish data, 2015 and 2018.
the topics used in the five questions. Even so, Fig. 4 shows (jittered) scatterplots of the 2015 index against both variants from 2018. The Pearson's r correlation between measures in the first panel is .38; in the second panel it is .22. Both are statistically significant at p < .001. It is somewhat difficult to know whether Fig. 4 points primarily to variability or stability in NBNS. We lean towards the latter interpretation, however. The stability is seen particularly when we collapse each measure into three categories for low-, medium-, and high-negativity groups (corresponding to 0-1 negative headlines, 2-3 negative headlines, and 4-5 negative headlines): A cross-tabulation of the 2015 and 2018 results finds over 50% of the cases on the diagonal. To be clear, this means about 50% of respondents are in exactly the same category three years later. Just as importantly, only 16 respondents switch from the low-negativity group to the high-negativity group, and another 16 switch from the high-negativity group to the low-negativity group. To the extent there is variation over time for most respondents, then, is
economic preferences (H3). In particular, we use the NBNS as a predictor for (a) retrospective economic evaluations, (b) government satisfaction, (c) the belief that governments tend to keep their promises, and (d) satisfaction with democracy. We regard these as tests of the concurrent validity of the NBNS measure and expect that respondents who tend to select negative news will have more negative views on each of these dimensions. We also see these results as tests of the potential importance of NBNS, relevant not just to news selection but to a range of political attitudes. We rely on relatively simple OLS regression models in Tables 4 and 5, controlling for age, sex, education, income, and whether the respondent voted for the incumbent party, as above. Full details for the other variables are available in the Appendix. The critical results – coefficients for the negativity variable – are in the top row of each table. In every case, the estimated coefficient is negative; in all but one, the coefficient is statistically significant. It clearly is the case that NBNS is associated with more negative views of 6
Personality and Individual Differences xxx (xxxx) xxxx
1.0
1.0
0.8
0.8
Revised NBNS, 2018
NBNS, 2018
S. Bachleda, et al.
0.6 0.4 0.2 0.0
0.6 0.4 0.2 0.0
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
NBNS, 2015
0.4
0.6
0.8
1.0
NBNS, 2015
Fig. 4. Scatterplots of 2015 and 2018 NBNS measures.
the economy and politics. Whether the estimated effect here is truly causal is unclear, of course. We can imagine two different possibilities. First, it may be that there are people who are attracted to negative information and thus tend to develop more negative views of the political world. Second, it may be that those who hold negative views of the economy or government also tend to gravitate towards reading negative information. (In the first case, NBNS is a stable trait; in the second, it may or may not be context-dependent.) Models in Tables 4 and 5 are intended primarily to explore the correlation between NBNS and attitudes, controlling for some basic demographics and political preference; not to make strong claims about whether NBNS is causally prior to, for instance, satisfaction with government. That said, as noted above, we suspect that NBNS reflects a partly durable predisposition to read negative information. Since we control for support for the incumbent, the estimated impact of the NBNS measure reflects its impact above and beyond incumbent support. This, at least, makes clear that the NBNS measure is not just about partisan preference. Note also that questions in the NBNS measure are not exclusively about political affairs, and the inclination to select negative headlines across topics such as health and environment should not, we suspect, be driven by an antipathy towards the current government. This too points to the possibility that the NBNS measure is independent of, and perhaps causally prior to, a range of (partisan) attitudes about politics and government. Table 6 repeats analyses from Tables 4 and 5, this time using the
Table 5 NBNS and economic sentiment.
NBNS Age Sex (1=Female) Education (1=College) Income (terciles) Incumbent Support Constant Observations R2
DV: Economic Sentiment Sweden Canada
USA
−0.105*** (0.013) −0.090*** (0.018) −0.003 (0.007) 0.045*** (0.007) −0.044*** (0.008) 0.170*** (0.007) 0.434*** (0.014) 10,317 0.066
−0.186** (0.057) −0.006 (0.083) −0.085** (0.031) 0.024 (0.040) 0.093* (0.040) 0.265*** (0.033) 0.474*** (0.068) 629 0.131
−0.074*** (0.017) −0.052* (0.025) −0.027* (0.010) −0.041*** (0.012) −0.009 (0.013) 0.240*** (0.013) 0.268*** (0.020) 3,167 0.117
Cells contain OLS coefficients with standard errors in parentheses. * p < .05; ** p < .01; *** p < .001.
Swedish 2018 respondents, and comparing results across three NBNS measures: the 2015 version, and the 2018 versions with the original and revised headlines. All NBNS measures are correlated with economic and
Table 4 NBNS and political attitudes.
NBNS Age Sex Education Income Incumbent Support Constant Observations R2
DV: Government Satisfaction Sweden Canada
USA
Keep Promises Sweden
Canada
Democratic Satisfaction Sweden Canada
−0.105*** (0.008) −0.114*** (0.012) 0.059*** (0.005) 0.033*** (0.005) −0.065*** (0.005) 0.270*** (0.005) 0.333*** (0.009) 11,565 0.268
−0.143*** (0.043) −0.212** (0.064) −0.021 (0.024) 0.044 (0.031) −0.019 (0.031) 0.381*** (0.025) 0.405*** (0.052) 671 0.281
−0.118*** (0.031) −0.091* (0.044) −0.019 (0.017) 0.075*** (0.020) 0.053** (0.020) 0.085*** (0.018) 0.412*** (0.036) 1,134 0.059
−0.023 (0.015) −0.005 (0.023) 0.021* (0.009) 0.017 (0.011) 0.013 (0.012) −0.022 (0.012) 0.343*** (0.018) 3,157 0.005
−0.116*** (0.010) −0.069*** (0.014) 0.044*** (0.005) 0.050*** (0.006) 0.064*** (0.007) 0.112*** (0.006) 0.522*** (0.011) 10,519 0.076
−0.058*** (0.015) −0.032 (0.022) −0.004 (0.009) −0.033** (0.011) 0.056*** (0.011) 0.508*** (0.011) 0.304*** (0.018) 3,206 0.411
Cells contain OLS coefficients with standard errors in parentheses. * p < .05; ** p < .01; *** p < .001. 7
−0.069*** (0.015) 0.020 (0.022) −0.017 (0.009) 0.002 (0.011) 0.063*** (0.011) 0.159*** (0.011) 0.536*** (0.018) 3,147 0.085
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
Table 6 NBNS and economic sentiment/political attitudes, Swedish data.
NBNS 2015 NBNS 2018
DV: Economic Sentiment (2018)
Government Satisfaction (2018)
−0.063 (0.039)
−0.084*** (0.024)
NBNS 2018, revised Age Sex Education Income Voted for Incumbent Constant Observations R2
0.013 (0.008) 0.026 (0.021) 0.067** (0.023) −0.050 (0.026) 0.294*** (0.024) 0.439*** (0.055) 1,295 0.125
−0.183*** (0.055) 0.008 (0.011) 0.032 (0.030) 0.069* (0.032) −0.064 (0.035) 0.328*** (0.033) 0.481*** (0.076) 640 0.172
−0.152* (0.059) 0.017 (0.010) 0.044 (0.029) 0.070* (0.032) −0.051 (0.036) 0.255*** (0.033) 0.458*** (0.075) 675 0.112
−0.005 (0.005) −0.032* (0.013) 0.028* (0.014) −0.055*** (0.016) 0.423*** (0.015) 0.423*** (0.034) 1,348 0.409
−0.212*** (0.035) −0.014* (0.007) −0.009 (0.019) 0.036 (0.021) −0.070** (0.023) 0.424*** (0.021) 0.476*** (0.048) 664 0.433
−0.122*** (0.035) 0.004 (0.006) −0.038* (0.018) 0.036 (0.019) −0.058** (0.022) 0.417*** (0.020) 0.417*** (0.044) 706 0.423
Cells contain OLS coefficients with standard errors in parentheses. * p < .05; ** p < .01; *** p < .001.
political attitudes, although the link is weakest for the 2015 measure. This is expected given that the dependent variables in these models are from the 2018 wave, three full years later after the 2015 NBNS measure. The 2018 version with the original headlines shows the strongest correlation with attitudes. Note that the revised headlines include no explicitly political headlines, however. This was purposeful, in order to better separate the measurement of NBNS from current political attitudes. That it reduces the connection between NBNS and political attitudes makes sense, then. But the fact that this revised measure of NBNS still exhibits a connection with political and economic attitudes also indicates the possibility that NBNS reflects an underlying tendency to prioritize negative news over positive news. Indeed, the Swedish panel data allow for one further test of whether NBNS reflects preferences that are durable over time: we have access to 13 waves prior to the measurement of NBNS and are thus able to track whether NBNS – captured in 2015 – is predictive of differences in political attitudes as far back as 2011. Fig. 5 shows the trend in satisfaction with government in Sweden over 14 surveys, separating respondents at the median of the NBNS index as fielded in the 2015 survey. The figure focuses on satisfaction with government (rather than the other dependent variables in Tables 4 and 5) because it is the question that has been asked most frequently in the Swedish panel over
the preceding four years. Fig. 5 suggests a clear and durable difference in government satisfaction, based on NBNS, over the entire four-year period. Put differently, NBNS captured in May 2015 appears to be related not just to attitudes about government in May 2015, but to attitudes about government four years prior. It is of some significance that the partisanship of the Swedish government has changed over this time period as well. A center-right alliance (led by the Moderate Party) was in power until the September 2014 election, at which point a green-left coalition (led by the Social Democrats) took power. It thus cannot be the case that the gap between low- and high-NBNS respondents is attributable to partisans who are systematically excluded from power. There is, in sum, evidence that a battery of headline-selection questions distinguishes between respondents who have more negative (or positive) views of the political and economic world around them. We take this as a preliminary sign that NBNS reflects durable differences in news-seeking behavior; and thus as evidence in support of H3. Importantly, the NBNS measure appears to distinguish attitudes not just at the time it is asked, but four years prior; and results are correlated, albeit moderately, when asked of the same respondents nearly three years later.
Fig. 5. Satisfaction with government, by NBNS 8
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
6. Discussion
preferences as an important avenue for future research. There is also potential for work that explores the degree to which more nuanced variation in tone – i.e., not just positive or negative, but gradations across that range – matters for story selection and measures of negativity biases. Our headlines do not vary in tone much within the negative and positive categories (see Appendix Fig. 2); this was done by design. But past work has suggested nonlinearities in negativity biases (e.g., Ito and Cacioppo, 2005), and these could be more fully explored using headlines that vary systematically in degrees of positivity or negativity. Finally, an exploration of the relationship between NBNS and other measures of negativity biases will be critical for future work. Given that other more standard measures of negativity biases are primarily labbased, we have not examined them in the survey data used here. However, understanding the extent to which NBNS is a domain-specific negativity bias, versus the consequence of a more domain-general bias, requires further research. Our results provide only a first step in this direction. In doing so, however, we regard the preceding analyses as a first signal that individual-level differences in news preferences may be one way in which personality differences are relevant to political attitudes and behavior.
There is reason to expect that individual-level variation in negativity biases has an important and durable impact on individuals’ news media use, as well as on a range of economic and political attitudes. This paper has taken a first step toward measuring a negativity bias in news selection. We find that while on balance there is a bias towards negativity, there are individual-level differences. These differences appear to be partly pre-dispositional; that is, they appear to be durable, demonstrated both by correlations with demographic, partisan and personality measures, and by within-respondent correlations across time. We also find that these individual-level differences are correlated with a variety of economic and political attitudes. We take these results as evidence of the potential importance of negativity biases in news selection (NBNS) in understanding attitudes about governments, the economy, and other politically and economically-relevant attitudes. We also suspect that NBNS moderates the impact of news content – those who are high in NBNS may select into a rather different information stream than those who are low in NBNS, which could subsequently shape their political perspectives. Although this application of the measure is not tested here, we thus see disentangling the relationship between political news selection and political Appendix
This appendix includes three sections detailing (1) survey samples, (2) question wording, and (3) additional diagnostics. Samples U.S. Sample Data for the U.S. study were collected as part of an online panel survey from a sample provided by Qualtrics, which recruited subjects using ClearVoice research. ClearVoice maintains a standing panel of survey respondents who were recruited to the platform through a combination of targeted emails, advertisements, and website intercepts. These individuals then opt-in to taking surveys and are recruited to participate in individual studies either by email or by clicking on a dashboard link. ClearVoice sent emails to 61,865 panelists with the goal of recruiting a broad national sample of at least 3,667 Americans to participate in the study. Swedish Sample Data for the first Swedish sample come from the Citizen Panel (original Swedish name: Medborgarpanelen – MP), which is a panel survey fielded online by the Laboratory of Opinion Research (LORE). Specifically, the data come from Citizen Panel 16 (MP16), which was fielded between June 9 and June 30, 2015. The panel used a mixed sampling design whereby 84 percent of the gross sample were opt-in and the remaining 16 percent were probability based. The panel wave included five separate modules and our data come from module 3 (Negativity Biases). This module yielded 12,867 complete responses for an AAPOR participation rate (RR5) of 92%. Data for the second Swedish sample also come from the Citizen Panel. Specifically, the data come from Citizen Panel 29 (MP29), which was fielded between March 22 and April 16, 2018. The panel used a mixed sampling design whereby 76 percent of the gross sample were opt-in and the remaining 24 percent were probability based. The panel wave included five separate modules and our data come from module 2 (Negativity Biases in News Selection). Additional information about the Citizen Panels can be found at http://lore.gu.se/surveys/citizen. Canadian Sample The Canadian data come from the 2015 Canadian Election Study. Full documentation for the study can be found at: http://ces-eec.arts.ubc.ca/ english-section/surveys/. The study was funded by the Social Sciences and Humanities Research Council of Canada.
Appendix Fig. 1. Distribution of selected headlines, by headline order. 9
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
Appendix Fig. 2. MTurk headline ratings
Question Wording Retrospective Economic Evaluations Sweden: Question: “Retrospective economic perceptions: Swedish economy over the past 12 months”; Original Swedish wording: “Hur har enligt din mening nedanstående ekonomiska förhållanden förändrats under de senaste 12 månaderna? - Den svenska ekonomin”; Response options: Gotten much better, Gotten somewhat better, Stayed about the same, Gotten somewhat worse, Gotten much worse, DK; Recoded to three categories: worse, same, better Canada: Question: “Now the economy. Over the past year, has Canada's economy …”; Response options: Gotten Better, Gotten Worse, Stayed about the Same, (DK/Refused omitted) USA: Question: “Now thinking about the economy in the country as a whole, over the past year how much better or worse would you say the nation's economy has gotten?”; Response options: Much worse, A little worse, About the same, A little better, Much better; Recoded to three categories: worse, same, better Government Satisfaction Sweden: Question: “Evaluation of Social Democrats” (wave 16); Original Swedish wording: “Hur tycker du att Socialdemokraterna sköter sin uppgift i regeringen?”; Response options: 7-point unlabeled scale from very bad to very good Canada: Question: "How satisfied are you with the performance of the federal government under [Stephen Harper]"; Response options: Not satisfied at all, Not very satisfied, Fairly satisfied, Very satisfied (DK/Refused excluded) USA: Question: “Overall, do you approve, disapprove, or neither approve nor disapprove of the way Barack Obama is handling his job as President?”; Response options: Strongly approve, Somewhat approve, Neither approve nor disapprove, Somewhat disapprove, Strongly disapprove Keep Promises Sweden: Question: “Swedish politicians usually keep election promises”; Original Swedish wording: “I vilken utsträckning instämmer du i följande påståenden om / svenska politiker? / -Svenska politiker håller oftast sina vallöften”; Response options: 5-point Likert scale from Strongly agree to Strongly disagree Canada: Question: "Do political parties keep their election promises?"; Response options: Most of the time, Some of the time, Hardly ever or never (DK/Refused omitted) Satisfaction with Democracy Sweden: Question: “Satisfaction with Democracy: Sweden”; Original Swedish wording: “På det hela taget, hur nöjd är du med det sätt på vilket demokratin fungerar i?: - Sverige”; Response options: Very satisfied, Fairly satisfied, Not very satisfied, Not satisfied at all Canada: Question: “On the whole, how do you feel about the way democracy works in Canada?”; Response options: Very satisfied, Fairly satisfied, Not very satisfied, Not satisfied at all (DK/Refused omitted) Incumbent support Sweden: If “Feel close to a certain party” = Social Democrats or Green Party then 1, 0 otherwise (wave 13) Canada: Conservative selected as vote choice = 1 (else = 0) using pre-campaign wave question USA: Democrat = 1 (else=0) using standard wave one Party ID question Political Interest Sweden: Based on a four-point scale, recoded a binary variable in which ‘very interested’ is equal to 1. Canada: Based on a 10-point scale, collapsed into a binary variable in which values exceeding 8 are equal to 1. USA: There is no standard measure of political interest available in the US survey, but we combine variables capturing the time (in minutes and hours) that respondents spend getting news from newspapers or online on an average day, and use the resulting measure to split the sample roughly in half.
10
Personality and Individual Differences xxx (xxxx) xxxx
S. Bachleda, et al.
Political Ideology Sweden: Based on a 10-point scale, where 0 is left-leaning and 10 is right-leaning. Canada: Based on a 10-point scale, where 0 is left-leaning and 10 is right-leaning. USA: Based on the standard 7-point party identification measure, where 1 is strongly Democratic and 7 is strongly Republican. Demographic Variables Age: age in years (Canada, Sweden, USA); Sex: Dummy variable where female =1 (Canada, Sweden, USA); Education: Dummy where college=1 (Canada, Sweden, USA); Income: Income in terciles (Canada, Sweden, USA) Additional Diagnostics Appendix Figure 1 shows the percent of respondents selecting headlines, based on the ordering of (randomized) headlines. There is no sign that headline placement mattered to headline selection. Appendix Fig. 2 shows the mean tone, and the associated 95% confidence interval, for each headline. Red squares indicate negative headlines, and blue squares indicate positive headlines. The vertical axis shows the proportion of respondents picking each headline, based on the US sample.
Kemp, A. H., Silberstein, R. B., Armstrong, S. M., & Nathan, P. J. (2004). Gender differences in the cortical electrophysiological processing of visual emotional stimuli. NeuroImage, 21(2), 632–646. Klein, J. G. (1991). Negativity effects in impression formation: A test in the political arena. Personality and Social Psychology Bulletin, 17(4), 412–418. https://doi.org/10. 1177/0146167291174009. Knobloch-Westerwick, S. (2012). Selective exposure and reinforcement of attitudes and partisanship before a presidential election. Journal of Communication, 62(4), 628–642. Lamberson, P. J., & Soroka, Stuart (2018). A model of attentiveness to outlying news. Journal of Communication, 68(5), 942–964. Lau, R. R. (1982). Negativity in political perception. Political Behavior, 4(4), 353–377. Lilienfeld, S. O., & Latzman, R. D. (2014). Threat bias, not negativity bias, underpins differences in political ideology. [Peer commentary on “Differences in negativity bias underlie variations in political ideology,” by J. R. Hibbing, K. B. Smith and J. R. Alford]. Behavioral and Brain Sciences, 37(3), 318–319. https://doi.org/10.1017/ S0140525×1300263X. Mondak, J. J. (2010). Personality and the foundations of political behavior. Cambridge, England: Cambridge University Press. Mondak, J. J., Hibbing, M. V., Canache, D., Seligson, M. A., & Anderson, M. R. (2010). Personality and civic engagement: An integrative framework for the study of trait effects on political behavior. American Political Science Review, 104(1), 85–110. https://doi.org/10.1017/S0003055409990359. Norris, C. J., Larsen, J. T., Crawford, L. E., & Cacioppo, J. T. (2011). Better (or worse) for some than others: Individual differences in the positivity offset and negativity bias. Journal of Research in Personality, 45(1), 100–111. https://doi.org/10.1016/j.jrp. 2010.12.001. Patterson, T. E. (1994). Out of order. New York: Vintage Books. Pietri, E. S., Fazio, R. H., & Shook, N. J. (2013). Weighting positive versus negative: The fundamental nature of valence asymmetry. Journal of Personality, 81(2), 196–208. https://doi.org/10.1111/j.1467-6494.2012.00800.x. Robinson, M. D., & Clore, G. L. (2002). Belief and feeling: Evidence for an accessibility model of emotional self-report. Psychological Bulletin, 128(6), 934–960. Rocklage, M. D., & Fazio, R. H. (2014). Individual differences in valence weighting: When, how, and why they matter. Journal of Experimental Social Psychology, 50, 144–157. https://doi.org/10.1016/j.jesp.2013.09.013. Rohrmann, S., Hopp, H., & Quirin, M. (2008). Gender differences in psychophysiological responses to disgust. Journal of Psychophysiology, 22(2), 65–75. https://doi.org/10. 1027/0269-8803.22.2.65. Rozin, P., & Royzman, E. B. (2001). Negativity bias, negativity dominance, and contagion. Personality and Social Psychology Review, 5(4), 296–320. Soroka, S. N. (2006). Good news and bad news: Asymmetric responses to economic information. Journal of Politics, 68(2), 372–385. Soroka, S. N., Gidengil, E., Fournier, P., & Nir, L. (2016). Do women and men respond differently to negative news. Politics & Gender, 12(2), 344–368. https://doi.org/10. 1017/S1743923×16000131. Soroka, S. N., & McAdams, S. (2015). News, politics, and negativity. Political Communication, 32(1), 1–22. Stevens, J. S., & Hamann, S. (2012). Sex differences in brain activation to emotional stimuli: A meta-analysis of neuroimaging studies. Neuropsychologia, 50(7), 1578–1593. Stroud, N. J. (2007). Media effects, selective exposure, and Fahrenheit 9/11. Political Communication, 24(4), 415–432. Stroud, N. J. (2010). Polarization and partisan selective exposure. Journal of Communication, 60(3), 556–576. Trussler, M., & Soroka, S. (2014). Consumer demand for cynical and negative news frames. The International Journal of Press/Politics 1940161214524832. Webster, S. W. (2018). It's personal: The big five personality traits and negative partisan affect in polarized U.S. politics. American Behavioral Scientist, 62(1), 127–145. Wrase, J., Klein, S., Gruesser, S. M., Derik, H., Herta, F., Karl, M., Braus, D. F., & Heinz, A. (2003). Gender differences in the processing of standardized emotional visual stimuli in humans: A functional magnetic resonance imaging study. Neuroscience Letters, 348(1), 41–45.
References Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D (2001). Bad is stronger than good. Review of General Psychology, 5(4), 323–370. Bradley, S. D., Angelini, J. R., & Lee, S. (2007). Psychophysiological and memory effects of negative political ADS: Aversive, arousing, and well remembered. Journal of Advertising, 36(4), 115–127. https://doi.org/10.2753/JOA0091-3367360409. Buck, R. (2014). Emotional attachment security as the origin of liberal-conservative differences in vigilance to negative features of the environment. [Peer commentary on “Differences in negativity bias underlie variations in political ideology,” by J. R. Hibbing, K. B. Smith and J. R. Alford]. Behavioral and Brain Sciences, 37(3), 308–309. https://doi.org/10.1017/S0140525×13002525. Caprara, G. V., Schwartz, S., Capanna, C., Vecchione, M., & Barbaranelli, C. (2006). Personality and politics: Values, traits, and political choice. Political Psychology, 27(1), 1–28. https://doi.org/10.1111/j.1467-9221.2006.00447.x. Darley, W. K., & Smith, R. E. (1995). Gender differences in information processing strategies: An empirical test of the selectivity model in advertising response. Journal of Advertising, 24(1), 41–56. Dodd, M. D., Balzer, A., Jacobs, C. M., Gruszczynski, M. W., Smith, K. B., & Hibbing, J. R. (2012). The political left rolls with the good and the political right confronts the bad: connecting physiology and cognition to preferences. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 367(1589), 640–649. https://doi.org/ 10.1098/rstb.2011.0268. Federico, C. M., Johnston, C. D., & Lavine, H. G. (2014). Context, engagement, and the (multiple) functions of negativity bias. [Peer commentary on “Differences in negativity bias underlie variations in political ideology,” by J. R. Hibbing, K. B. Smith and J. R. Alford]. Behavioral and Brain Sciences, 37(3), 311–312. https://doi.org/10.1017/ S0140525×13002550. Feldman, S. (2013). Political Ideology. In L. Huddy, D. O. Sears, & J. S. Levy (Eds.). The Oxford handbook of political psychology(2nd ed). Oxford: Oxford University Press. Retrieved from http://www.oxfordhandbooks.com/view/10.1093/oxfordhb/ 9780199760107.001.0001/oxfordhb-9780199760107-e-019. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38. Garrett, R. K. (2009). Echo chambers online? Politically motivated selective exposure among internet users. Journal of Computer Mediated Communication, 14(2), 265–285. Garrett, R. K., & Stroud, N. J. (2014). Partisan paths to exposure diversity: differences in pro- and counterattitudinal news consumption. Journal of Communication, 64(4), 680–701. Gosling, S. D., Rentfrow, P. J., & Swann, W. B., Jr (2003). A very brief measure of the big five personality domains. Journal of Research in Personality, 37, 504–528. Grabe, M. E., & Kamhawi, R. (2006). Hard wired for negative news? Gender differences in processing broadcast news. Communication Research, 33(5), 346–369. Granberg, D., & Holmberg, S. (1991). Self-reported turnout and voter validation. American Journal of Political Science, 35(2), 448–459. Helson, H. (1964). Adaptation-level theory: An experimental and systematic approach to behavior. New York: Harper and Row. Hibbing, J. R., Smith, K. B., & Alford, J. R. (2014). Differences in negativity bias underlie variations in political ideology. Behavioral and Brain Sciences, 37(3), 297–307. https://doi.org/10.1017/S0140525×13001192. Hofmann, W., Gawronski, B., Gschwendner, T., Le, H., & Schmitt, M. (2005). A metaanalysis on the correlation between the Implicit Association Test and explicit selfreport measure. Personality and Social Psychology Bulletin, 31, 1369–1385. Holbrook, A. L., Krosnick, J. A., Visser, P. S., Gardner, W. L., & Cacioppo, J. T. (2001). Attitudes toward presidential candidates and political partisanship: Initial optimism, inertial first impressions, and a focus on flaws. American Journal of Political Science, 45(4), 930–950. https://doi.org/10.2307/2669333. Ito, T., & Cacioppo, J. (2005). Variations on a human universal: Individual differences in positivity offset and negativity bias. Cognition and Emotion, 19(1), 1–26. https://doi. org/10.1080/02699930441000120. Jost, J. T., Glaser, J., Kruglanski, A. W., & Sulloway, F. J. (2003). Political conservatism as motivated social cognition. Psychological Bulletin, 129(3), 339–375. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 363–391.
11