Political partisanship alters the causality implicit in verb meaning Journal: Journal of Language and Social Psychology Manuscript ID Draft Manuscript Type: Keywords: Original Manuscript Social cognition, Cognition, Causality, Psycholinguistics, Political Affiliation Abstract: This research examined politics and causal attribution during the 0 U.S. Presidential Election by adapting the implicit causality (IC) task a psycholinguistics measure that reveals the causal information encoded in verbs. Results showed that Hillary Clinton and Donald Trump supporters judged their preferred candidate as causal for positive events and their non-preferred candidate as causal for negative events, demonstrating the social psychological utility of the IC task and expanding understanding of extralinguistic influences on causal attribution.
Page of 0 0 0 0 0 0 Abstract This research examined politics and causal attribution during the 0 U.S. Presidential Election by adapting the implicit causality (IC) task a psycholinguistics measure that reveals the causal information encoded in verbs. Results showed that Hillary Clinton and Donald Trump supporters judged their preferred candidate as causal for positive events and their non-preferred candidate as causal for negative events, demonstrating the social psychological utility of the IC task and expanding understanding of extralinguistic influences on causal attribution. Keywords: Social Cognition; Cognition; Causality; Psycholinguistics; Political Affiliation
Page of 0 0 0 0 0 0 Political partisanship alters the causality implicit in verb meaning In the sentence George assaulted Julie because would you predict the next word to be he (implicating George, the subject of the sentence, as the causal source of the event) or she (implicating Julie, the object of the sentence, as the causal source of the event)? Prior work in psycholinguistics using the implicit causality (IC) task (Garvey & Caramazza, ) has found that people s responses are mainly influenced by the given verb s semantic structure, which has been used to organize verbs into classes that tend to have implicit causality biases tendencies to compel selection of either the subject vs. the object as the cause of the event (Ferstl, Garnham, & Manouilidou, 0; Hartshorne & Snedeker, 0; Hartshorne, 0; Kipper-Schuler, 00). Yet, researchers have also argued that individual differences shape causal cognition (Rudolph & Forsterling, ). Recently, it was found that some moral values, the group-supporting binding values of loyalty, obedience, and preservation of purity (Graham et al., 0), are reliably associated with a tendency to select the sentence object as the causal source for harm events using the IC task (Niemi, Hartshorne, Gerstenberg, Stanley, & Young, under review). Based on this finding, when people who strongly endorse binding values encounter sentences such as: George assaulted Julie because, they might be more likely to continue with the referent to the object, in this case, she. Notably, although binding values are more strongly endorsed by political conservatives (Graham et al., 0; Graham, Nosek & Haidt, 00), the association between object-bias and binding values was not explained by political ideology. This suggests that the general tendency to see harm as victim-precipitated, as captured by the IC task, was better explained by the alliance-supporting nature of binding values than by politics (Niemi et al., under review). The present research builds upon these findings by investigating the role of individual differences in causal attributions using the IC task this time examining in detail the role of political alliances and hostilities in attributions for harmful and positive events. In prior work, political preferences were measured very simply, with an item gauging participants liberal or conservative affiliation. Judgments of harmful events were made using subjects
Page of 0 0 0 0 0 0 and objects with generic first names. In the present research, we hypothesized that people s support for Hillary Clinton or Donald Trump during the 0 U.S. Presidential Election would motivate them to attribute negative events to the opponent, and positive events to the preferred candidate in the IC task. If so, this would indicate that the IC task has additional utility as social psychological instrument because it reveals people s zeal and hostility toward specific targets in addition to previously observed general tendencies to attribute harmful events to victims when they more strongly endorsed group-oriented moral values (Niemi et al., 0; under review). Moreover, if IC responses are found to reflect people s political allegiances, this would indicate that, in addition to the causal information encoded in verb meaning (e.g., Hartshorne, 0), IC responses for multiple verbs with positive and negative moral valence should be expected to vary flexibly based on individual differences in ideology and situational constraints including the political climate. During the 0 U.S. Presidential Election, each side saw the other as unusually divisive: a 0 Pew Research Centre survey showed that over half of both Democrat and Republican respondents considered the opposing political party as more close-minded than the average American (Fingerhut, 0). The current research, carried out before and after the election, examined how participants explained events involving Donald Trump and Hillary Clinton, including confrontation and hostile discourse, such as mocking and interrupting. In a study conducted in the months before the 0 U.S Presidential Election and in a replication dataset collected in the days following election day, participants who supported either Trump or Clinton viewed sentences in the form [Trump/Clinton] [verb]ed [Clinton/Trump] because and then indicated whether the next word should be he (implicating Trump as the causal factor) or she (implicating Clinton as the causal factor). Verbs indicated either negative events (e.g., interrupted ) or positive events (e.g., thanked ). For both datasets, we hypothesized an interaction between event type and political affiliation for causal attributions, such that Trump supporters would be more likely to judge Clinton as the cause of negative events and Trump as the cause of positive events, and the opposite for Clinton supporters.
Page of 0 0 0 0 0 0 Method Participants For Study, Amazon Mechanical Turk workers over the age of and with United States IP addresses were recruited online during the first two weeks of October 0. After exclusions, the final sample was 0. This sample size approximates those in which individual differences were previously found to factor into implicit causality responses for people in different groups (Niemi et al., 0). Of the 0 participants, % planned to vote for Hillary Clinton and % planned to vote for Donald Trump, % were female and % were male, % were White or Caucasian, and 0% had a Bachelor s degree or higher. Average age was years (SD= years). Participant information for the replication dataset (final sample n=0), collected post-election can be found in the Supplemental Materials (SM) Section. Procedure This study protocol was approved by an institutional review board and was carried out in accordance with the APA Code of Ethics. After indicating their informed consent, participants responded to items about their political attitudes and then completed the implicit causality task (Garvey & Caramazza, ; Niemi, Hartshorne, Gerstenberg, & Young, 0), described in the following section. Participants also completed mathematical processing items; those data are reported elsewhere and were not analyzed in relation to the present measures (Niemi, Young, Cordes, & Woodring, 0). Additional measures from the primary study and from the replication dataset are described in SM Section. Lastly, participants provided demographic information. The replication dataset used the same procedure. Materials To gauge voting intentions, participants were asked: Who are you voting for in the upcoming election? with the options being Donald Trump, Hillary Clinton, other, or not voting. In the implicit causality task, participants were presented with prompts in the form of [Trump/Clinton] [verb]ed [Clinton/Trump] because ; for each prompt, participants were asked to predict the next word in the sentence, with the options being he or she. Responses were coded to indicate that the pronoun
Page of 0 0 0 0 0 0 referring to the object () vs. subject (0) of the sentence was selected. The verbs were divided into two groups, Set A and Set B, containing a total of verbs conveying positive events and verbs conveying negative (Table ). Participants viewed one set of verbs (either Set A or Set B containing positive and negative events each) either in the format Clinton [verb]ed Trump, or in the format Trump [verb]ed Clinton. The second set of verbs viewed (the yet unseen set; e.g., Set B if a participant first saw Set A) utilized the prompt format not presented with the first set of verbs (e.g., Clinton [verb]ed Trump, if a participant first saw Trump [verb]ed Clinton ). Prompt format order and verb set order were randomly assigned. Within each set, the individual verbs were presented in randomized order. Table. Verbs and Sets in the Implicit Causality Task Set A Verbs Set B Verbs Positive Verbs complimented praised inspired interested forgave thanked impressed comforted Negative Verbs interrupted attacked disgusted intimidated approached confronted criticized mocked frustrated annoyed ran against took on crushed outdid squashed beat The demographic information collected included level of education, age, gender (male, female, or other), ethnicity (open-response), religiosity on a scale of = not at all religious to = very religious, and political ideology on a scale of = very conservative to = very liberal. Results All statistical analyses were completed using R software version.. (R Core Team, 0). Because we did not make any predictions about the effects of demographic characteristics, the present
Page of 0 0 0 0 0 0 analyses do not include demographic variables as covariates. The results for Study and for the replication dataset remain the same when demographic variables are included in the model, see SM Sections and for details. Based on Niemi et al. (under review), to examine causal attributions (=object vs. 0=subject), the present analyses utilized the lme software package (Bates, Maechler, Bolker, & Walker, 0) to test a generalized linear mixed-effects regression model (link= logit ) which included event type (=negative vs. 0=positive) and political affiliation (=Trump supporter vs. 0=Clinton supporter), as fixed predictors, and participant ID and verb, as random effects with random intercepts only, at Step ; the interaction between event type and political affiliation was added at Step. Because the outcome variable was binary, we used Wald to compute significance and % CIs around beta-estimates. For ease of analyses, and because we expected responses to differ based on who was in the subject or object position, this model was run separately for Trump-as-object and for Clinton-asobject prompts. Because we expected attributions to differ based on whether the event was positive or negative, we broke down interactions by event type, using the procedures recommended by Aiken, West, and Reno (). Trump-as-Object We first analyzed participants responses to prompts in the form of Clinton [verb]ed Trump because. A response of indicates that the object (here, Trump) was the cause of the event, whereas a response of 0 indicates that the subject (here, Clinton) was the cause of the event. In Step of the regression model, there were no significant main effects of either event type, b =.0, SE =., Z =.0, p =., % CI = [-.,.], or political affiliation, b = -.0, SE =.0, Z = -., p =., % CI = [-.,.0]. At Step, as predicted, there was a significant interaction between event type and political affiliation, b = -., SE =., Z = -., p <.00, % CI = [-., -.]. Within positive events, Trump (vs. Clinton) supporters were more likely to identify Trump as the causal factor (see Top Left panel of Figure ), b =., SE =., Z =., p <.00, % CI = [.,.0]. By contrast, for negative events, Trump (vs. Clinton) supporters were less likely to identify Trump as the causal factor (see Top Left panel, Figure ), b = -0., SE =.0, Z = -.0, p <.00, % CI = [-., -
Page of 0 0 0 0 0 0.]. In general, attributions for negative versus positive events did not significantly differ among Trump supporters, b = -0., SE =., Z = -.0, p =.0, % CI = [-.,.]; or Clinton supporters, b =., SE =., Z =.0, p =., % CI = [-.,.]. This same pattern of results was found in the replication dataset, see Top Right panel, Figure and SM Section. Percent Selecting Object (Trump) Percent Selecting Object (Clinton) 0 0 0 0 Positive Clinton Supporter Positive Clinton Supporter Negative Trump Supporter Negative Trump Supporter Percent Selecting Object (Trump) Percent Selecting Object (Clinton) 0 0 0 0 Positive Clinton Supporter Positive Clinton Supporter Negative Trump Supporter Negative Trump Supporter Figure. Percent of participants who indicated the object of the sentence as the cause of the event. The left-side panels display data from Study ; the right-side panels display data from the Replication Dataset in the Supplementary Material.
Page of 0 0 0 0 0 0 Clinton-as-Object We next analyzed participants responses to prompts in the form of Trump [verb]ed Clinton because. In this case, a response of indicates that Clinton was the cause of the event, whereas response of 0 indicates that Trump was the cause of the event. At Step of the model, there was no effect of event type, b = -., SE =.0, Z = -., p =.0, % CI = [-.,.0]. Differing from the Trump-as-Object condition, there was a significant effect of political affiliation, b =., SE =.0, Z =., p <.00, % CI = [.,.], such that regardless of event type, Trump (vs. Clinton) supporters were significantly more likely to indicate Clinton as the causal factor. This effect was qualified by the predicted significant interaction between event type and political affiliation, b =., SE =., Z =., p <.00, % CI = [.0,.0], which we broke down by event type. Analogous to Clinton supporters attributions to Trump in the Trump-as-Object condition (see Top Panel, Figure ), within positive events, Trump (vs. Clinton) supporters were less likely to indicate Clinton as the causal factor (see Bottom Left Panel, Figure ), b = -0., SE =.0, Z = -., p <.00, % CI = [-., -.]. By contrast, for negative events, Trump (vs. Clinton) supporters were more likely to identify Clinton as the cause (see Bottom Left Panel, Figure ), b =., SE =.0, Z =., p <.00, % CI = [.0,.]. Analogous to how event type did not affect Clinton supporters causal attributions for Trump, negative events were not attributed to Clinton significantly differently from positive events by Trump supporters, b =.0, SE =., Z = 0.0, p =., % CI = [-.,.0]. By contrast, Clinton supporters were significantly less likely to identify Clinton as the cause if the event was negative compared to if the event was positive, b = -., SE =., Z = -., p =.00, % CI = [-.00, -.]. A similar pattern of results was found in the replication dataset, see Bottom Right panel, Figure and SM Section. Discussion The present research shows, in Study and in a replication dataset, that participants preferences for Hillary Clinton vs. Donald Trump during the 0 U.S. Presidential Election influenced their causal
Page of 0 0 0 0 0 0 judgments of events involving the two candidates. As hypothesized, for positive events (e.g., he or she thanked, interested, praised ), both Trump and Clinton supporters were more likely to choose their preferred candidate as the causal factor, regardless of whether that candidate occupied the sentence subject or object position. For negative events (e.g., mocked, attacked, criticized ), Trump and Clinton supporters were more likely to choose their non-preferred candidate as the causal factor, again regardless of that candidate s position in the sentence. Comparing causal attributions across event type, in general, Trump and Clinton supporters causal attributions for positive vs. negative events did not differ. The finding that participants were biased to see their preferred candidate as the cause of some of the positive events and the non-preferred candidate as the cause of some of the negative events is particularly striking from a psycholinguistic standpoint. Many of the verbs for which we observed this effect reliably induce people to select the object as the causal impetus (i.e., upwards of % of the time for positive verbs in the judgment verb class: thanked, praised, complimented, forgave, Hartshorne, 0; Ferstl et al., 0). The tendency for participants to most often judge the non-preferred candidate as the cause of negative events like disgusted, annoyed, frustrated, and attacked is similarly notable as there is little lexical backing for causal attribution to the sentential object for these typically subject-biased verbs (Ferstl et al., 0). Furthermore, when the non-preferred candidate was in the sentence object position, this tendency led participants to be, in effect, victim-blaming. Some previous work suggests that blaming the victim for their harmful situation might typically occur among political conservatives and be less prevalent among political liberals (Lambert & Raichle, 000; cf. Niemi & Young, 0). The present data suggest that, in a polarizing environment, shifting blame to protect one s preferred candidate might be likely in conservatives and liberals alike. The finding that Trump and Clinton supporters attributed positive and negative events to their preferred and non-preferred candidates roughly equivalently may be due to the commonly held view that politicians are dishonest (Gallup, 0). Even positive acts of the opponent candidate, therefore, can easily be viewed in a negative light: as fake. Participants may have understood the non-preferred candidate as having nefarious reasons for seemingly positive acts (e.g., Clinton thanked Trump
Page 0 of 0 0 0 0 0 0 because she was trying to pander to Trump supporters ), in addition to straightforwardly malicious intentions for negative acts. Indeed, news media characterized both Trump and Clinton this way (Greenberg, 0). Future research might test this hypothesis by coding participants open-ended responses to IC prompts like Trump [verbed] Clinton because for whether event valence matches reasoning valence (i.e., a positive reason for a positive act and vice-versa). A limitation of the present research is that we were not able to examine the effects of political party affiliation outside of a political contest, when polarization and hostile discourse might be less prevalent. To investigate whether our results are specific to investments in a political contest, future research might examine effects on causal attributions from party affiliation in general. The present work holds both theoretical and practical implications. Theoretically, the present research advances understanding of the factors influencing causal cognition by further demonstrating that verbs implicit causality biases are determined not only by lexical semantics (Hartshorne, 0), namely the morally relevant nature of the verb, but also by who occupies the subject and object position of a sentence, their perceived political affiliation, and how it aligns with one s own (Niemi et al., under review). Thus, the current research provides additional evidence of the power of personal beliefs about society in shaping perceptions of causality (Alicke et al., 0; Niemi et al., under review; Niemi et al., 0), displays the theoretical utility of the implicit causality task in the process, and supplements current literature on relationships between political ideology and cognition (Jost & Amodio, 0; Washburn & Skitka, 0). Practically, the present research contributes to an understanding of the psychological factors that potentially played a role in the outcome of the 0 U.S. Presidential Election (Azevedo, Jost, & Rothmund, 0; Bock, Byrd-Craven, & Burkley, 0; Choma & Hanoch, 0) by highlighting a way that individuals support for a particular presidential candidate may have altered their perceptions of and subsequent reactions toward events involving the two candidates, such as debates. Moreover, the findings show that the implicit causality task is a methodologically lean way to reveal how political alliances shape causal attributions for political events, indicating that lexical semantics factor into the causal cognition
Page of 0 0 0 0 0 0 that drives political partisanship even as we cover politics in the news, and consume that news (Faris et al., 0). In conclusion, the current research presents further evidence of the practical and theoretical value of analysis of implicit causality as an indicator of beliefs and biases, reveals another role for individual differences in causal attributions, and provides a unique take on cognitive processes involved in people s perceptions of the 0 U.S Presidential Election candidates.
Page of 0 0 0 0 0 0 References Aiken, L. S., West, S. G., & Reno, R. R. (0). Multiple regression: Testing and interpreting interactions. Newbury Park: SAGE Publications. Alicke, M., Mandel, D. R., Hilton, D. J., Gerstenberg, T., Lagnado, D. A. (0). Causal conceptions in social explanations and moral evaluations: A historical tour. Azevedo, F., Jost, J. T., & Rothmund, T. (0). Making America Great Again : System Justification in the U.S. Presidential Election of 0. Translational Issues in Psychological Science, (), 0. Bates, D., Mächler, M., Bolker, B., & Walker, S. (0). Fitting linear mixed-effects models using lme. Journal of Statistical Software, (). DOI: 0./jss.v0.i0 Bromwich, J. E. (0, December ). Protests of Trump's election continue into third day. The New York Times. Retrieved from https://www.nytimes.com/0///us/trump-electionprotests.html Bock, J., Byrd-Craven, J., & Burkley, M. (0). The role of sexism in voting in the 0 presidential election. Personality and Individual Differences,,. https://doi.org/0.0/j.paid.0.0.0 Choma, B. L., & Hanoch, Y. (0). Cognitive ability and authoritarianism: Understanding support for Trump and Clinton. Personality and Individual Differences, 0,. https://doi.org/0.0/j.paid.0.0.0 DeSilver, D. (0, December 0). Why Electoral College wins are bigger than popular vote ones. Pew Research Center. Retrieved from http://www.pewresearch.org/facttank/0//0/why-electoral-college-landslides-are-easier-to-win-than-popular-vote-ones/ Faris, R., Roberts, H., Etling, B., Bourassa, N., Zuckerman, E., & Benkler, Y. (0). Partisanship, Propaganda, and Disinformation: Online Media and the 0 U.S. Presidential Election (Research Publication No. 0-). Cambridge, MA. Retrieved from https://papers.ssrn.com/sol/papers.cfm?abstract_id=0
Page of 0 0 0 0 0 0 Garvey, C., & Caramazzo, A. (). Implicit causality in verbs. Linguistic Inquiry,(), -. https://www.jstor.org/stable/ Gentzkow, M. (0). Polarization in 0. Toulouse Network for Information Technology Whitepaper. Retrieved from https://web.stanford.edu/~gentzkow/research/polarizationin0.pdf Graham, J., Haidt, J., & Nosek, B. A. (00). Liberals and conservatives rely on different sets of moral foundations. Journal of Personality and Social Psychology,, 0-0. DOI: 0.0/a00 Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (0). Mapping the moral domain. Journal of Personality and Social Psychology, 0, -. DOI: 0.0/a00 Greenberg, D. (0, August/July). Are Clinton and Trump the biggest liars ever to run for Ferstl, E. C., Garnham, A., & Manouilidou, C. (0). Implicit causality bias in English: A corpus of 00 verbs. Behavior Research Methods, (), -. DOI: 0./s-00-00- Fingerhut, H. (0, December ). Partisanship and political animosity in 0. Pew Research Centre. Retrieved from http://www.people-press.org/0/0//partisanship-and-politicalanimosity-in-0/ Gallup. (0, December -). Honesty/Ethics in professions. Retrieved from https://news.gallup.com/poll//honesty-ethics-professions.aspx President?: A short history of White House fabulists. Politico. Retrieved from https://www.politico.com/magazine/story/0/0/0-donald-trump-hillary-clinton-ushistory-presidents-liars-dishonest-fabulists-0 Hartshorne, J. K. (0). What is implicit causality? Language, Cognition and Neuroscience,, 0-.
Page of 0 0 0 0 0 0 Hartshorne, J. K. & Snedeker, J. (0). Verb argument structure predicts implicit causality: The advantages of finer-grained semantics. Language and Cognitive Processes, (0), - 0. DOI: 0.00/00.0.0 Jost, J. T., & Amodio, D. M. (0). Political ideology as motivated social cognition: Behavioral and neuroscientific evidence. Motivation and Emotion, (), -. Kipper-Schuler, K. (00). VerbNet: A broad-coverage, comprehensive verb lexicon. Ph.D. thesis, University of Pennsylvania. Lambert, A. J., & Raichle, K. (000). The role of political ideology in mediating judgments of blame in rape victims and their assailants: A test of the Just World, Personal Responsibility, and Legitimization hypotheses. Personality and Social Psychology Bulletin, (),. https://doi.org/0./00000 Niemi, L., Hartshorne, J. K., Gerstenberg, T., Stanley, M., & Young, L. (under review). Moral values in causal attribution: Evidence from the implicit verb causality task and explicit judgments. Niemi, L., Hartshorne, J. K., Gerstenberg, T., & Young, L. (0). Implicit measurement of motivated causal attribution. Cognitive Science Society Proceedings. Niemi, L., Woodring, M., Young, L. & Cordes, S. (0). Partisan mathematical processing of political polling statistics: It s the expectations that count. Cognition,, -0. Niemi, L. & Young. L. (0). When and why we see victims as responsible: The impact of ideology on attitudes toward victims. Personality and Social Psychology Bulletin, (), -. Rudolph, U., & Forsterling, F. (). The psychological causality implicit in verbs: A review. Psychological Bulletin,, -. Washburn, A. N., & Skitka, L. J. (0). Science Denial Across the Political Divide: Liberals and Conservatives Are Similarly Motivated to Deny Attitude-Inconsistent Science. Social Psychological and Personality Science, (), 0. https://doi.org/0./000
Page of 0 0 0 0 0 0 Footnotes. Excluded participants reported they did not plan to vote for either Hillary Clinton or Donald Trump in the 0 United States Presidential Election (N=), did not complete the primary measures of interest (N=), or indicated disagreement or only somewhat agreement with the statement The United States is geographically north of Central America (N=0) i.e., failed the attention check.. We used the same exclusions for the replication dataset and we only analyzed the data of participants who voted for Clinton or Trump.
Page of 0 0 0 0 0 0 Table of Contents Section : Study Models Including Key Demographic Variables... Section : Replication Dataset... Section : Replication Dataset Models Including Key Demographic Variables... Section : Additional Measures...
Page of 0 0 0 0 0 0 Section : Study Models Including Key Demographic Variables Table S Study Steps - of the Model Containing Political Affiliation (=Trump Supporter, 0=Clinton Supporter), Event Type (=Negative, 0=Positive), Education, Gender (=Female, 0=Male), Liberalism, Religiosity, and Political Affiliation x Event Type Predicting Causal Attributions (=Object, 0=Subject) for Trump-As-Object Sentences Model Effect B SE Z P % CI Constant -0. 0. -.00 0.0 -., 0. Political Affiliation -0. 0.0 -. 0. -0., 0.0 Event Type 0.0 0. 0.0 0. -0., 0. Education 0.0 0.0 0. 0. -0.0, 0.0 Gender -0. 0.0 -. 0.00-0., -0.0 Liberalism 0.0 0.0 0. 0. -0.0, 0.0 Religiosity 0.0 0.0. 0.00 0.0, 0.0 Constant -0. 0. -. 0.0 -., 0. Political Affiliation 0. 0.. <0.00 0.,.00 Event Type 0.0 0..0 0.0-0.,. Education 0.0 0.0 0. 0. -0.0, 0.0 Gender -0.0 0.0 -. 0.00-0., -0.0 Liberalism 0.0 0.0 0. 0. -0.0, 0.0 Religiosity 0.0 0.0. 0.00 0.0, 0.0 Political Affiliation x Event Type -. 0. -. <0.00 -., -.0
Page of 0 0 0 0 0 0 Table S Study Simple Effects of Political Affiliation (=Trump Supporter, 0=Clinton Supporter) and Event Type (=Negative, 0=Positive) from the Model Containing Political Affiliation, Event Type, Gender, Religiosity, and Political Affiliation x Event Type Predicting Causal Attributions for Trump-As-Object Sentences Effect B SE Z P % CI Trump Supporter: Negative vs Positive Event -0. 0. -.0 0.0 -., 0. Clinton Supporter: Negative vs Positive Event 0. 0..0 0. -0.,. Negative Event: Trump vs Clinton Supporter -0. 0.0 -. <0.00-0., -0. Positive Event: Trump vs Clinton Supporter 0. 0.. <0.00 0.0, 0. Note. Liberalism and education were excluded when breaking down this interaction because neither variable was statistically significant in the main model
Page of 0 0 0 0 0 0 Table S Study Steps - of the Model Containing Political Affiliation (=Trump Supporter, 0=Clinton Supporter), Event Type (=Negative, 0=Positive), Education, Gender (=Female, 0=Male), Liberalism, Religiosity, and Political Affiliation x Event Type Predicting Causal Attributions (=Object, 0=Subject) for Clinton-As-Object Sentences Model Effect B SE Z P % CI Constant 0. 0.. 0. -0.,. Political Affiliation 0. 0.0.0 0.0 0.0, 0. Event Type -0. 0. -. 0.0 -., 0.0 Education -0.0 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -. <.0.00-0., -0. Liberalism -0.0 0.0 -. 0. -0.0, 0.0 Religiosity 0.0 0.0 0. 0. -0.0, 0.0 Constant 0. 0.. 0.0-0.0,.0 Political Affiliation -0. 0. -. <0.00-0., -0. Event Type -.0 0. -. 0.00 -.0, -0. Education -0.0 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -.0 <0.00-0., -0. Liberalism -0.0 0.0 -. 0. -0.0, 0.0 Religiosity 0.0 0.0 0. 0. -0.0, 0.0 Political Affiliation x Event Type.0 0.. <0.00.0,.
Page 0 of 0 0 0 0 0 0 Table S Study Simple Effects of Political Affiliation (=Trump Supporter, 0=Clinton Supporter) and Event Type (=Negative, 0=Positive) from the Model Containing Political Affiliation, Event Type, Gender and Political Affiliation x Event Type Predicting Causal Attributions for Clinton-As-Object Sentences Effect B SE Z P % CI Trump Supporter: Negative vs Positive Event 0.0 0. 0.0 0. -0., 0. Clinton Supporter: Negative vs Positive Event -. 0. -. 0.00 -.0, -0. Negative Event: Trump vs Clinton Supporter 0. 0.0. <0.00 0., 0. Positive Event: Trump vs Clinton Supporter -0. 0.0 -. <0.00-0., -0. Note. Liberalism, religiosity, and education were excluded when breaking down this interaction because none of these variables were statistically significant in the main model
Page of 0 0 0 0 0 0 Method Section : Replication Dataset The goal of this study was to replicate the findings from Study. Participants. Mechanical Turk workers over the age of and with United States IP addresses were recruited in the days following the 0 US Presidential Election on November th (November th through November th ). After exclusions, the final sample was 0. Of the 0 participants, 0% indicated voting for Hillary Clinton and 0% indicated voting for Donald Trump, % were female, % were White or Caucasian, and % had a Bachelor s degree or higher. Average age was years (SD= years). Procedure and Materials. The procedure was identical to that of Study. The same measures and materials from Study were also used, with one unrelated measure added at the end, see Supplemental Materials (SM) Section. Results As in Study, all statistical analyses were completed using R software version.. (R Core Team, 0). When demographic variables are included in the model, the results do not change, see SM Section for details. Analyses were conducted as in Study. Trump-as-Object. At Step of the model, there were no main effects of either event type, b =.0, SE =., Z = 0.0, p =., % CI = [-.,.], or political affiliation, b = -.0, SE =.0, Z = - 0., p =., % CI = [-.,.]. Replicating Study, the interaction between the two variables at Step was significant, b = -.00, SE =., Z = -., p <.00, % CI = [-., -.]. In keeping with Study, we first broke down the interaction by event type. Once again, for positive events, Trump (vs. Clinton) supporters were more likely to indicate Trump as the causal factor, b =., SE =., Z =.0, p <.00, % CI = [.,.].For negative events, Trump (vs. Clinton) supporters were less likely to Excluded participants reported they had not voted for either Hillary Clinton or Donald Trump (N=0), did not complete the primary measures of interest (N=), or indicated disagreement or only somewhat agreement with the statement The United States is geographically north of Central America (N=),
Page of 0 0 0 0 0 0 indicate Trump as the causal factor, b = -0., SE =.0, Z = -., p <.00, % CI = [-., -.], see Top Right Panel of Figure in the main text. Comparing across event types, negative vs. positive events did not elicit significantly different attributions among either Clinton supporters, b =., SE =., Z = 0., p =., % CI = [-.,.0], or Trump supporters, b = -0., SE =., Z = -., p =., % CI = [-.,.0]. These findings replicate Study. Clinton-as-Object. At Step of the model, there was a main effect of political affiliation such that, regardless of event type, Trump (vs. Clinton) supporters were more likely to choose Clinton as the cause of the event, b =., SE =.0, Z =., p <.00, % CI = [.,.0]. There was a marginally significant main effect of event type such that, regardless of political affiliation, Clinton was somewhat less likely to be chosen as the cause when the event was negative (vs. positive), b = -0., SE =., Z = -., p =.0, % CI = [-.,.0]. These main effects were qualified by an interaction, also seen in Study, between event type and political affiliation at Step, b =., SE =., Z = 0.0, p <.00, % CI = [.,.], which we again broke down by event type. Replicating Study, for positive events, Trump (vs. Clinton) supporters were less likely to indicate Clinton as the causal factor, b = -0., SE =., Z = -., p =.00, % CI = [-., -.]; for negative events, Trump (vs. Clinton) supporters were more likely to indicate Clinton as the causal factor, b =., SE =.0, Z =., p <.00, % CI = [.0,.0], see Bottom Right Panel in Figure in the main text. As in Study, negative vs. positive events did not elicit significantly different attributions among Trump supporters, b = -0.0, SE =.0, Z = -0.0, p =., % CI = [-.,.]. However, among Clinton supporters, Clinton was less likely to be chosen as the causal factor for negative (vs. positive) events, b = -., SE =.0, Z = -., p =.00, % CI = [-., -.].
Page of 0 0 0 0 0 0 Section : Replication Dataset Models Including Key Demographic Variables Table S Replication Dataset Steps - of the Model Containing Political Affiliation (=Trump Supporter, 0=Clinton Supporter), Event Type (=Negative, 0=Positive), Education, Gender (=Female, 0=Male), Liberalism, Religiosity, and Political Affiliation x Event Type Predicting Causal Attributions (=Object, 0=Subject) for Trump-As-Object Sentences Model Effect B SE Z P % CI Constant -0. 0. -. 0. -., 0. Political Affiliation 0.0 0. 0. 0. -0., 0. Event Type 0.0 0. 0.0 0. -0., 0. Education -0. 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -.0 0.0-0., -0.0 Liberalism -0.0 0.0. 0.0-0.0, 0.0 Religiosity 0.0 0.0. 0. -0.0, 0.0 Constant -0. 0. -.0 0.0 -., -0.0 Political Affiliation 0. 0..0 <0.00 0., 0. Event type 0. 0. 0. 0. -0.,. Education -0.0 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -. 0.0-0., -0.0 Liberalism -0.0 0.0. 0. -0.0, 0.0 Religiosity 0.0 0.0. 0.0-0.0, 0.0 Political Affiliation x Event Type -.00 0. -. <0.00 -., -0.
Page of 0 0 0 0 0 0 Table S Replication Dataset Simple Effects of Political Affiliation (=Trump Supporter, 0=Clinton Supporter) and Event Type (=Negative, 0=Positive) from the Model Containing Political Affiliation, Event Type, Gender, and Political Affiliation x Event Type Predicting Causal Attributions for Trump-As-Object Sentences Effect B SE Z P % CI Trump Supporter: Negative vs Positive Event -0. 0. -.0 0. -., 0.0 Clinton Supporter: Negative vs Positive Event 0. 0. 0. 0. -0.,. Negative Event: Trump vs Clinton Supporter -0. 0.0 -. <0.00-0., -0.0 Positive Event: Trump vs Clinton Supporter 0. 0.. <0.00 0., 0. Note. Liberalism, religiosity, and education were excluded when breaking down this interaction because none of these variables were statistically significant in the main model
Page of 0 0 0 0 0 0 Table S Replication Dataset Steps - of the Model Containing Political Affiliation (=Trump Supporter, 0=Clinton Supporter), Event Type (=Negative, 0=Positive), Education, Gender (=Female, 0=Male), Liberalism, Religiosity, and Political Affiliation x Event Type Predicting Causal Attributions (=Object, 0=Subject) for Clinton-As-Object Sentences Model Effect B SE Z P % CI Constant 0. 0.0 0. 0. -0.,.0 Political Affiliation 0.0 0.. <0.00 0., 0. Event Type -0. 0. -.0 0.0 -., 0.0 Education -0.0 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -. <.0.00-0., -0.0 Liberalism -0.0 0.0-0. 0. -0.0, 0.0 Religiosity 0.0 0.0-0. 0. -0.0, 0.0 Constant 0. 0.. 0. -0.,. Political Affiliation -0.0 0. -. <0.00-0., -0. Event Type -. 0.0 -.0 0.00 -.0, -0. Education -0.0 0.0-0. 0. -0.0, 0.0 Gender -0. 0.0 -. <0.00-0.0, -0. Liberalism -0.0 0.0-0. 0.0-0.0, 0.0 Religiosity -0.0 0.0-0. 0.0-0.0, 0.0 Political Affiliation x Event Type. 0. 0.0 <0.00 0.,.
Page of 0 0 0 0 0 0 Table S Replication Dataset Simple Effects of Political Affiliation (=Trump Supporter, 0=Clinton Supporter) and Event Type (=Negative, 0=Positive) from the Model Containing Political Affiliation, Event Type, Gender, and Political Affiliation x Event Type Predicting Causal Attributions for Clinton-As-Object Sentences Effect B SE Z P % CI Trump Supporter: Negative vs Positive Event -0.0 0.0-0.0 0. -0., 0. Clinton Supporter: Negative vs Positive Event -. 0.0 -. 0.00 -.0, -0. Negative Event: Trump vs Clinton Supporter 0. 0.0. <0.00 0.,.0 Positive Event: Trump vs Clinton Supporter -0. 0. -. <0.00-0., -0. Note. Liberalism, religiosity, and education were excluded when breaking down this interaction because none of these variables were statistically significant in the main model
Page of 0 0 0 0 0 0 Section : Additional Measures In both Study and in the replication dataset, participants also did a sentence completion task after completing the implicit causality task and before completing the demographics. Participants were presented with a subset of the study prompts again (specifically, the verbs 'outdid' and 'beat'). Half of participants were randomly assigned to read Clinton outdid Trump and then Trump beat Clinton; the other half read Trump outdid Clinton and then Clinton beat Trump. For each prompt, they were given a text box in which to complete the sentence. In replication dataset only, after the sentence completion task and before the demographics, participants completed a version of the implicit causality task meant to measure gender bias (Niemi, Hartshorne, Gerstenberg, & Young, 0). Participants were presented with prompts in the form of [Male name/female name] [verb]ed [Female name/male name] because ; for each prompt, participants were asked to predict the next word in the sentence, with the options being he or she. Names were randomly inputted from a list of generic male names (e.g., Max, George, Ben) and a list of generic female names (e.g., Melissa, Julie, Carol). The verbs were divided into two groups, Set A and Set B, as described in the main manuscript (see Table ). Participants viewed one set of verbs (either Set A or Set B) either in the format Male name [verb]ed Female Name, or in the format Female name [verb]ed Male name because. The second set of verbs viewed (the unseen set) utilized the prompt format not presented with the first set of verbs. Prompt format order and verb set order were randomly assigned. Within each set, the individual verbs were presented in randomized order.