Replication of changing hearts and minds? Why media messages designed to foster empathy often fail (Gubler et al., 2022)

Prochazka, Jakub; Pandey, Shubham; Castek, Ondrej; Firouzjaeiangalougah, Mojtaba Working Paper Replication of changing hearts and minds? Why media messages designed to foster empathy often fail (Gubler et al., 2022) MUNI ECON Working Paper, No. 2024-02 Provided in Cooperation with: Masaryk University, Faculty of Economics and Administration Suggested Citation: Prochazka, Jakub; Pandey, Shubham; Castek, Ondrej; Firouzjaeiangalougah, Mojtaba (2024) : Replication of changing hearts and minds? Why media messages designed to foster empathy often fail (Gubler et al., 2022), MUNI ECON Working Paper, No. 2024-02, Masaryk University, Faculty of Economics and Administration, Brno, https://doi.org/10.5817/WP_MUNI_ECON_2024-02 This Version is available at: https://hdl.handle.net/10419/286872 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ n. 2024-02 ISSN 2571-130X DOI: 10.5817/WP_MUNI_ECON_2024-02 Replication of Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail (Gubler et al., 2022) Jakub Prochazka / Masaryk University, Faculty of Economics and Administration, Department of Business Management, Brno, Czech Republic Shubham Pandey / Indian Institute of Technology Bombay, Mumbai Ondrej Castek / Masaryk University, Faculty of Economics and Administration, Department of Business Management, Brno, Czech Republic Mojtaba Firouzjaeiangalougah / Masaryk University, Faculty of Economics and Administration, Brno, Czech Republic Replication of Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail (Gubler et al., 2022) Abstract This paper focuses on computational reproducibility and robustness replicability of Gubler et al.’s(2022) studies which examine the effect of media messages on empathic concern, dissonance, and out-group policy attitudes. The original paper tests four hypotheses using two online experiments with large samples from one US state (N1=5,800; N2=2,200). Regarding the first experiment, we successfully reproduced the effect that initial antipathy weakens the effect of humanizing treatment on empathic concern (H1). However, we show that the moderating effect is negligible and has little practical significance. Moreover, the individual effect estimates in our analyses slightly differed from the original paper due to different procedure of data cleaning and minor coding errors in the original paper. The most relevant difference was the opposite effect of gender than reported in the original paper. We also show that empathic concern might mediate the effect of humanizing treatment on attitudes toward immigrants (H3). The original study rejected the mediation hypothesis due to not finding a total effect of humanizing treatment on attitudes. In contrast, we found that humanization treatment has a positive indirect effect on attitudes through empathic concern. At the same time, it also has a direct negative effect on attitudes. For the second experiment (H1, H2a, H2b, H3), we attempted to reproduce the results using a different software. We partially succeeded once receiving support from the authors of the original study. We note throughout the report issues we have encountered. Masaryk University Faculty of Economics and Administration Authors: Jakub Prochazka / Masaryk University, Faculty of Economics and Administration, Department of Business Management, Brno, Czech Republic Shubham Pandey / Indian Institute of Technology Bombay, Mumbai Ondrej Castek / Masaryk University, Faculty of Economics and Administration, Department of Business Management, Brno, Czech Republic Mojtaba Firouzjaeiangalougah / Masaryk University, Faculty of Economics and Administration, Brno, Czech Republic Contact: [email protected] Creation date: 2024-03 Revision date: Keywords: Reproduction, Replication, Research Transparency, Open Science, Economics, Political Science, Persuasion, Political Communication, Empathic Concern JEL classification: B41, C10, C81, P49 Citation: Prochazka, J., Pandey, S., Castek, O., Firouzjaeiangalougah, M. (2024). Replication of Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail (Gubler et al., 2022). MUNI ECON Working Paper n. 2024-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2024-02 (https://creativecommons.org/licenses/by-nc-nd/4.0/) Licensing of the final text published in the journal is in no way conditional on this working paper licence. MUNI ECON Working Paper n. 2024-02 ISSN 2571-130X 1 Replication of Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail (Gubler et al., 2022) Jakub Prochazka1 Shubham Pandey2 Ondrej Castek1,3 Mojtaba Firouzjaeiangalougah1 Abstract This paper focuses on computational reproducibility and robustness replicability of Gubler et al.’s (2022) studies which examine the effect of media messages on empathic concern, dissonance, and out-group policy attitudes. The original paper tests four hypotheses using two online experiments with large samples from one US state (N1=5,800; N2=2,200). Regarding the first experiment, we successfully reproduced the effect that initial antipathy weakens the effect of humanizing treatment on empathic concern (H1). However, we show that the moderating effect is negligible and has little practical significance. Moreover, the individual effect estimates in our analyses slightly differed from the original paper due to different procedure of data cleaning and minor coding errors in the original paper. The most relevant difference was the opposite effect of gender than reported in the original paper. We also show that empathic concern might mediate the effect of humanizing treatment on attitudes toward immigrants (H3). The original study rejected the mediation hypothesis due to not finding a total effect of humanizing treatment on attitudes. In contrast, we found that humanization treatment has a positive indirect effect on attitudes through empathic concern. At the same time, it also has a direct negative effect on attitudes. For the second experiment (H1, H2a, H2b, H3), we attempted to reproduce the results using a different software. We partially succeeded once receiving support from the authors of the original study. We note throughout the report issues we have encountered. 1 Masaryk University, Faculty of Economics and Administration, Department of Business Management, Lipova 41a, 602 00 Brno, Czech Republic, [email protected]. 2 Indian Institute of Technology Bombay, Mumbai, [email protected]. 3 Corresponding author. Masaryk University, Faculty of Economics and Administration, Department of Business Management, Lipova 41a, 602 00 Brno, Czech Republic, [email protected]. 2 1. Introduction Gubler et al. (2022) examine the effect of media messages on empathic concern, dissonance, and out-group policy attitudes. They run two separate large-scale experiments with subjects being Anglos from a conservative western US state with a high prevalence of republicans (N = 5,800; N = 2,200). The samples were drawn from citizens of a particular US state described by authors as Western, very conservative. The name of the state was not disclosed in the paper. Data was collected in January 2012 (study 1) and September 2015 (study 2). Responses were obtained from subjects described as Anglos or white/Caucasian. Questionnaires were disseminated via e-mail to several subpopulations with response rates from 9% to 19%. We completed computational reproduction using the original dataset obtained from Harvard Dataverse4 and other materials (supplemental material, log file with R syntax) that are available on the webpage5 of one of the authors. Below is a summary of the hypotheses formulated by Gubler et al. (2022, p. 2160) and the support these hypotheses received in the original study. H1. Individuals with high pretreatment out-group antipathy – often the targets of humanizing media messages – will exhibit low levels of empathic concern as a result of humanizing information about the out-group, while individuals with low pretreatment antipathy toward the out-group will exhibit high levels of empathic concern. H1 assumes a moderation effect of pretreatment out-group antipathy on the relationship between humanizing message and empathic concern. The analytical method was multiple regression with OLS estimator. In Study 1, Gubler et al. (2022) found strong support for this hypothesis: “a stark difference between those two groups [low vs. high pretreatment antipathy] emerges”, “low antipathy respondents reported dramatically higher levels of empathy in the humanization and combined conditions compared to high antipathy respondents (pp. 2163-4)”. Support for these claims is given in Figure 2 (Gubler et al. 2022, p. 2164) and in Supplemental material of the original paper, Table G.9. The moderating effects of pretreatment antipathy on the effects of both treatments Humanization and Combined on empathic concern was significantly 4 https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/FUCDTT 5 https://davidaromney.com/publication/ 3 negative (both cases: β = -0.40, p < .001). The moderating effect holds exactly the same after including all control variables. In Study 2, Gubler et al. (2022) found support for H1: “Low antipathy participants reported significantly more empathic concern than high antipathy participants no matter the experimental condition, and both groups decreased in empathy when they were assigned to the “illegal” condition. But the effect of the illegal condition was over twice as large for high antipathy participants, and the predicted point estimate for those with high levels of antipathy assigned to that condition fell below the scale midpoint (p. 2166).” Support for these claims is given in Figure 5A, 5B (Gubler et al. 2022, p. 2166) and in Supplemental material of the original paper, Table G.11. The moderating effects of pretreatment antipathy on the effects of treatments Illegal condition on empathic concern was significantly negative (β = -0.05, p < .01 in sparse model (2), β = -0.05, p < .05 in full model including controls (3)). The moderating effect was notably weaker than in Study 1 (β = -0.40 vs β = -0.05). H2a. Individuals with high levels of out-group antipathy before treatment will on average exhibit higher levels of dissonance posttreatment. H2b. Individuals with low pretreatment antipathy will exhibit little or no change in dissonance levels. Gubler et al. (2022) approached both H2 as a moderation effect of pretreatment antipathy on the relationship between the treatment and dissonance, i.e., on the level of dissonance caused by the treatment. The analytical method was multiple regression with OLS estimator. H2 were tested in Study 2 only and Gubler et al. (2022) found support for both H2a and H2b. First, “participants with high levels of pretreatment outgroup antipathy were more likely than those with low antipathy to report dissonant affect regardless of treatment condition”, second “the difference between high and low antipathy participants in self-reported dissonance was more than three times larger in the illegal condition”, and finally “the difference in differences between high and low out-group antipathy is also significant (p = .02), representing key evidence of our hypothesized mechanism at work (p. 2167)”. Support is given for two claims in Supplemental material of the original paper, Table G.12. First, Illegal condition increases dissonance: β = 0.04, p < .001 in basic model (1), β = 0.02, p < .1 in sparse model with interaction term (2), and β = 0.02, p < .05 in full model including controls (3). 4 Second, the pretreatment antipathy is a significant moderator for Illegal condition β = 0.05, p < .05 in sparse model (2), and β = 0.05, p < .05 in full model including controls (3). All mentioned effects are weak. H3. While posttreatment empathy levels will be correlated with posttreatment political attitudes, the unpleasant affect from dissonance will result in small or zero average effects of the media message treatments on attitudes. Gubler et al. (2022) explain H3 followingly “in the presence of both empathy (pleasant affect) and dissonance (unpleasant affect), average experimental effects of humanization treatments on policy attitudes should be small or nonexistent (p. 2167).” H3 can be understood also as effect of humanization on attitude being mediated by empathic concern. In Study 1, Gubler et al. (2022) found support for this hypothesis: “The key finding overall is that neither of the conditions with humanizing messages had any discernible effect on policy attitudes (pp. 2168).” Support for this claim is given in Table 1 (Gubler et al. 2022, p. 2167) and in Supplemental material of the original paper, Tables G.13 and G.15. Treatment effects on support of harmful policies (i.e., negative attitude towards immigrants) are either statistically non-significant, or very weak: Humanization β = -0.01, n.s., Information β = 0.01, n.s., Combined β = -0.01, n.s., in basic model (1); Humanization β = -0.00, n.s., Information β = 0.04, p < .05, Combined β = -0.00, n.s., in sparse model with interaction terms (2); Humanization β = -0.01, n.s., Information β = 0.04, p < .05, Combined β = -0.00, n.s., in full model including controls (3). The results change slightly based on antipathy being measured as continuous (Table G.13) or dichotomous (Table G.15). In Study 2, Gubler et al. (2022) come to the same conclusion: “While there is some evidence that, relative to the legal condition, humanizing messages in the illegal condition decreased support for policy harm, the effect is quite small. Overall, as in study 1, support for policy harm is primarily a function of pretreatment antipathy toward immigrants, the effect of which is dramatically larger than the experimental conditions (p. 2168).” Support for this claim is given in Table 2 (Gubler et al., 2022, p. 2168) and in Supplemental material of the original paper, Tables G.14 and G.16. 5 Treatment effect on support of harmful policies is statistically significant (unlike Study 1), but very weak: β = -0.03, p < .01 in basic model (1); β = -0.04, p < .05 both in sparse model with interaction terms (2) and in full model including controls (3). The results change slightly based on antipathy being measured as continuous (Table G.14) or dichotomous (Table G.16). In the present paper, we first tested the computational reproducibility using a new version of the same software (R ver. 1.1.463) and following the published syntax. We then checked the computational reproducibility of the tests of the four hypotheses that were stated in the original paper using different software (SPSS ver. 28, MATLAB ver. 2022, Mathwork Inc). Finally, we perform two robustness checks on the results of the first study. To reproduce the hypotheses test, we first cleaned the data and computed the compound variables as described in the paper and its supplemental material. If the description was not clear, we looked for details in the provided R script in the log file. Regarding the first experiment, we reproduced the tests of H1 and H3 and performed robustness checks focused on the practical significance of the moderation effect of outgroup antipathy (H1) and on the existence of the mediation effect of empathic concern in the relationship between humanization messages and attitudes towards immigrants (H3). In the case of the second experiment, we were not able to reliably identify the variables needed for the analyses based on the dataset, log file, and supplemental materials. The original authors provided support which allowed us to computationally reproduce their results with minor differences. 2. Reproducibility First, we ran the original syntax in R ver. 1.1.463 using the original dataset to compare our results with the figures presented in the original manuscript and tables presented in the supplemental file. Each step of the analysis was carefully followed, and all necessary code snippets were executed as instructed. Using the same syntax, we were able to reproduce all the main estimates and all published figures (see Appendix 1 for detailed results) for both experiments except the Figure A.1 (flowchart in Supplemental material of the original paper). We found that the treatment labels are misassigned within Figure A1. The order should be Humanization, Information, Combined, 6 Control. Nevertheless, the treatments are labeled correctly in the log file and in the other parts of the manuscript. It seems that this presentation error did not influence the interpretation of hypothesis testing. 2.1 Study 1: Reproducibility using different software 2.1.1 Data cleaning Two authors tried to reproduce the data cleaning procedure of Study 1 using two different softwares (SPSS ver. 28 and MATLAB). None of us was able to reach the same sample size as described in the original study. We are aware that some differences in the data cleaning procedure might be caused by the decision of which of the duplicate cases should be kept for further analyses. In the first attempt with SPSS, we tried various ways to exclude the duplicates, but none of these efforts led to the same result as presented in the original study. Therefore, we have chosen the procedure best matching the description in the original manuscript. If a duplicate appeared in the data matrix, we preferred to keep the record that better met the other criteria for retention in the sample (i.e., did not indicate non-white ethnicity, did not indicate problems with video, finished the survey; see Figure A.1 in the Supplemental material of the original paper for more details). Following this approach, similar to the original study, we removed 149 duplicate cases (according to the identifier) and 168 non-whites (i.e., respondents who identified themselves as otherwise than White/Caucasian). However, in the next step, we identified only 1596 (not 1610 as in the original study) respondents with video issues and 384 (not 386) respondents who did not finish the questionnaire. We continued with the analysis in SPSS with a sample of 3514 respondents (not 3494 as in the original study) who met all the conditions described in the manuscript. These responses were divided among four treatments in the following way: Humanization: 847, Information: 949, Combined: 858, Control: 860. In the second attempt by the other co-author with MATLAB, the analysis started with a sample of 5811. We first dropped 1539 unfinished responses followed by dropping 568 responses with video issues. Then we dropped 195 non-whites followed by dropping 28 duplicate entries. Notably we removed duplicates at the last stage of data cleaning to minimize potential loss of any relevant response. The duplicate removal was automatically done by the MATLAB algorithm. Afterwards, 13 Figure 2: Marginal effects of the treatments on empathic concern, by levels of out-group antipathy. Figure 3: Empathy gap: difference between low and high antipathy individuals in reported empathic concern, by treatment condition. 14 2.2 Study 2: Reproducibility using different software 2.2.1 Data cleaning According to Figure A.2 (Supplemental material, p. 4 of the original paper), there were 2632 (3623 invited – 991 not responded) participants who responded to the second wave of the survey. However, the provided dataset with manuscript contains 2159 participants only. We first note that the original authors have not provided raw data for study 2. We thus could not recode from scratch. For instance, “non-finished” responses were already removed. Additionally, these items did not have original labels; instead all items were already assigned labels as icb1, icb2,..., icb10; item nr. 8 (icb8) was reverse coded, which is not mentioned in the original manuscript. Upon request, original authors informed us that the variables in the dataset have different order than the items in the survey, but we were not able to check it as we missed the codebook. We started the data cleaning procedure with a sample of 2159. 130 non-whites were removed. Then, we removed 47 responses which had no treatment assigned. Therefore, we performed all analysis with a final sample of 1982 responses which is the same as what the original authors reported. Out of the final sample, we identified 999 illegal and 983 legal responses (same as original manuscript). Also, we could not figure out whether Male was coded 1 or female was coded 1. 2.2.2 Operationalization of variables Study 2 tested all four hypotheses. Dissonance was manipulated separately from humanization (see Gubler et al., 2022, p. 2162). More importantly, to unbundle dissonance from humanization/empathy, the measurement was done in two waves. Outgroup antipathy, infrahumanization, and all demographics were measured in the first wave, which allowed to divide respondents in the group of “low antipathy” and “high antipathy” at the scale midpoint. Outgroup antipathy was yet again computed using the adapted “ethos of conflict” measure developed by Bar-Tal et al. (2009, 2012), Roccas et al. (Roccas, Klar, and Liviatan 2006; Roccas et al. 2008), Shnabel et al. (2009), and others. One difference to Study 1 is that in Study 2, outgroup antipathy was measured by all nine items (Study 1: just three items). The summary score created from all nine items was recoded to range from 0 to 1. 15 To do a manipulation check, the study measured infrahumanization, supposedly the same way as in Study 1: respondents rated the extent to which immigrants are likely to feel two secondary positive emotions: admiration and love. One difference here was that infrahumanization was measured both in the first and the second (after exposure to humanization vignette, see Gubler et al., 2022, p.2162-3) measurement wave. There were two experimental conditions: legal (also documented) and illegal (also undocumented) immigration. After exposure to humanization vignette, the respondents were informed that immigrants shown “have come to this country illegally/legally”. Next, the dissonance (dependent variable for H2a and H2b) was elicited via standard “induced compliance” framework by Festinger and Carlsmith (1959). Respondents were to answer questions about the immigrants on scales with positive responses only. After that, dissonance was measured as how they felt emotions identified by Elliot and Devine (1994) and Haslam (2006) as indicators of dissonance: uncomfortable, uneasy, bothered, tense, and concerned. Empathic concern (dependent variable for H1) was measured by Batson’s six-item measure (Batson et al. 1997, 2002). The summary score of all six items was recorded to range from 0 to 1. Attitudes towards immigrants (also political attitudes) were operationalized through “harm index” (dependent variable for H3) and measured by eight items (seven items in Study 1) that measured an attitude towards real or hypothetical law norms that may harm the immigrants. The summary score of all eight items was recoded to range from 0 to 1. In analyses, authors also controlled for gender, age, and political party preference (1–7, 1 = Strong Democrat, 7 = Strong Republican, rescaled to 0-1). The Data cleaning section provides more details on our success in replicating the calculations. 2.2.3. First independent attempt to reproduce the results In SPSS, we tried to compute the mean scores for all above described variables that were used for testing the hypotheses. According to the Supplemental material of the original paper (p. 5), the measure of group antipathy had 9 items and the item nr. 4 was reverse-coded. In the data file, we found 10 antipathy items (icb1-icb10) without specific labels. According to the R syntax, item 8 was reverse coded. It was not possible to identify if there was an extra item in the questionnaire or 16 if there is a mistake in the dataset. We tried to follow the R syntax for creating the antipathy index (log file p. 14). According to the syntax, the item icb7 is missing in the index. In a follow-up email conversation, the authors specify that icb7 was added “for exploratory reasons; it was not meant to be part of the 9-item scale”. We also found a variable that seemed to be an antipathy index calculated by the authors (icb_measure), but we were not able to calculate the same values using various combinations of antipathy items. As it was not possible to calculate the original antipathy index or to create a new antipathy index (because we don’t know the meaning of individual items icb1-icb10), the author of this reproduction attempt considered Study 2 to be not reproducible and did not try to reproduce the hypotheses testing. We would need further information from the authors which would allow us to computationally reproduce this part of the analysis. 2.2.4 Second independent attempt to reproduce the results In MATLAB, we tried to reproduce the original results. However, we found the same issues as highlighted in the first attempt. Despite some further difficulties, we successfully computed policy harm index. However, because full wording of questions were not provided, we could not verify the computed antipathy score and policy harm. Section B.5 in Supplemental material of the original paper mentions seven and ten policy items used for measuring attitude towards immigrants in study 1 and study 2, respectively. Nevertheless, we followed the same syntax as provided to calculate regression models. We successfully reproduced the table with regression of dissonance on treatments (table 5 here, G.12 in supplementary file). We found a weak effect suggesting that participants with high levels of outgroup antipathy reported higher dissonance (β = 0.05, p < .001). We notice that the authors have swapped the labels (titles) of policy harm tables in Study 2 in supplementary file: Table G14 and G16 with each other. Alternatively, the authors may have swapped actual table values, while the labels were correct. For example, the authors mention G14’s label as having antipathy variable as dichotomous. However, we get this table when we take antipathy as continuous variable. The vice-versa is true for G16. Here, we report tables keeping labels same as original authors but replacing table values. We also found some evidence that, relative to the legal condition, humanizing messages in the illegal condition decreased support for policy harm. The effect is quite small, β = -0.05, p < 0.01, while controlling for other variables. This is not in line with the 17 hypothesis that humanizing messages may not move policy attitudes in substantively significant way. Tab. 5: Regression of Dissonance on Treatments × Antipathy (Dichotomous), Controls, Study 2 (1) (2) (3) B SE B SE B SE Intercept 0.26*** 0.01 0.24*** 0.01 0.30*** 0.02 Illegal Condition -0.05*** 0.01 0.03† 0.01 0.03* 0.01 Outgroup Antipathy 0.05*** 0.01 0.06*** 0.01 Ill. Con. × Antipathy 0.05* 0.02 0.05* 0.02 Gender - 0.02* 0.01 Age - 0.00*** 0.00 Party ID (0–1) 0.02 0.02 N 1963 1961 1945 R2 0.01 0.04 0.05 adj. R2 0.01 0.04 0.05 Note. † p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001. Tab. 6: Regression of Policy Harm on Treatments × Antipathy (Continuous), Controls, Study 2 (1) (2) (3) B SE B SE B SE Intercept 0.62*** 0.01 0.25*** 0.01 0.19*** 0.02 Illegal Condition -0.03** 0.01 -0.05** 0.02 - 0.05** 0.02 Outgroup Antipathy 0.85*** 0.03 0.79*** 0.03 Ill. Con. × Antipathy 0.04 0.04 0.04 0.04 Gender - 0.02* 0.01 Age 0.00*** 0.00 Party ID (0–1) 0.09*** 0.01 N 19561 1959 1944 R2 0.00 0.52 0.53 adj. R2 0.00 0.52 0.53 Note. † p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001. 3. Robustness Checks We did a robustness replication of Study 1 using the same data, different software (SPSS) and different type of analyses. 18 The focus of the H1 is the moderation effect of outgroup antipathy on the relation between the humanization message and the empathic concern. The original study supported the hypothesis by examining the regression coefficient of the interaction terms "humanization x antipathy" and "combined x antipathy". To assess the practical significance of the moderation effect, we computed a new regression model in which we entered the predictors in several steps. First, we entered control variables and outgroup antipathy. In the second step, we inserted all treatments. In the third step, we added the interaction term "humanization x antipathy" and in the fourth step we inserted the interaction term "combined x antipathy". We assessed how the proportion of explained variance in empathic concern increased with each step. As can be seen in Tab. 7, the control variables and antipathy explained 9.6% of the variance in empathic concern. Adding treatments in the second step significantly improved the model (ΔR2 = 0,371, p < 0,001). Adding both the "humanization x antipathy" interaction (ΔR2 = 0,003, p < 0,001) and the "combined x antipathy" interaction (ΔR2 = 0,008, p < 0,001) led to a statistically significant but only marginal (+1,1% of explained variance in attitude) improvement of the model. This analysis provided statistical support for H1, but also showed that the moderation effect of antipathy is marginal and, that initial antipathy is not very contributing for understanding the effect of humanizing video clips on empathic concern. Tab. 7: Regression of Empathic Concern on Treatments x Antipathy and Controls (1) (2) (3) (4) B SE B B B SE B SE Intercept 0.55 0.03 0.34 0.02 0.32 0.02 0.28 0.02 Gender (1 = Male) - 0.03 0.01 - 0.03 0.01 - 0.03 0.01 - 0.03 0.01 Age 0.001 0.00 0.00 0,00 0.00 0.00 0.02** 0.00 Party ID (0 - 1) 0.06** 0.02 0.05 0.02 0.05** 0.02 0.05 0.02 Antipathy - 0.35 0.02 - 0.34 0.02 - 0.30 0.02 - 0.22 0.02 Humanization 0.36 0.01 0.44 0.02 0.48 0.02 Information 0.09 0.01 0.09 0.01 0.09 0.01 Combined 0.34 0.01 0.34 0.01 0.47 0.02 Hum. x Antipathy - 0.16 0.04 - 0.24 0.04 Comb. x Antipathy - 0.25 0.04 N 3255 3255 3255 3255 R2 0.096*** 0.413*** 0.416*** 0.424*** adj. R2 0.094 0.411 0.415 0.422 Note. *p < 0.05; **p < 0.01; *** p < 0.001. 19 The essence of H3 is the assumption that there is no indirect effect of humanization treatment message on attitudes (harm index) through empathic concern. The authors tested this assumption through the direct effect of humanization treatment on the harm index (see Table 1 in the original manuscript). As they found no significant relation, they concluded that “neither of the conditions with humanizing messages had any discernible effect on policy attitudes” (p. 2168). This conclusion is probably based on the assumption that an indirect effect (ie., humanization → empathic concern → attitude) can only exist if there is a significant relation between the independent and dependent variables (ie., humanization → attitude). Such an assumption is in line with the recommendations for mediation analyses from Baron and Kenny (1986) and others. However, we followed more recent recommendations (see e.g., Zhao et al., 2010) and tested whether there can be a significant indirect effect even if there is no correlation between the independent variable and the outcome . Therefore, we did a mediation analysis with 5000 bootstrap samples using the PROCESS v4.2 plugin for SPSS (Hayes, 2017, Model 4), to test the indirect effect of humanization treatment on attitudes towards immigrants operationalized as the harm index. We estimated two models in which the independent variable was humanization treatment and combined treatment, respectively. The mediator was always empathic concern, and the dependent variable was harm index. Two other treatments, gender, age, and political preference, were controlled as covariates. The analyses showed that empathic concern significantly mediates the relationship between the humanization treatment and the harm index (see Figure 1) and also between the combined treatment and the harm index (see Figure 2). Both mediation effects were rather weak (partially standardized indirect effects were –0,242 for humanization treatment and – 0,231 for combined treatment) but not marginal. This finding supports the H3. 20 Figure 4: Mediation analysis for Humanization treatment Figure 5: Mediation analysis for Combined treatment According to the mediation analysis, the total effects of treatments on the harm index were very weak and insignificant, which is in line with the results of the original study. However, besides the negative indirect effects of treatments through empathic concern, there were also positive direct effects of both treatments on the harm index. Both humanization and combined treatments worsened participants' attitudes toward immigrants (i.e., increased the harm index) when the effect of empathic concern was controlled (see Appendix 2 for detailed results). 4. Conclusion Using the same data and the same software, we were able to reproduce the analyses presented in the original paper. We also found support for the hypothesis (H1) that outgroup antipathy moderates the effect of humanization media messages on empathetic concern for immigrants when we analyzed data from the Study 1 using different software. Nevertheless, the individual effect estimates were slightly different due to different procedure of data cleaning and minor coding 21 errors. The most relevant difference is the opposite effect of gender than reported in the original paper. We also point out that the moderation effect of initial outgroup antipathy is negligible and lacks practical significance. While treatments explain 31.7% of the variance in empathic concern, adding initial antipathy as a moderator helps explain only another 1.1% of the variance of the dependent variable. This means that although high initial antipathy weakens the effect of humanizing messages, this effect is negligible. Regardless of the level of initial antipathy, humanizing messages have a similar positive effect on people with different levels of initial outgroup antipathy. The robustness check provided important conclusions regarding the third hypothesis concerning the indirect effect of humanization messages on attitudes towards immigrants. In contrast to the original study, we provide suggestive evidence that humanization messages weaken negative attitudes toward immigrants through empathic concern. At the same time, however, these messages also directly reinforce the negative attitude through a yet unclear mechanism. Our analysis showed that the mediation effect might exist. However, the effect is not evident in the correlation or regression analysis because humanization improves attitudes toward immigrants through empathic concern and, at the same time, worsens them through other potential mechanisms. Our mediation analysis is not definitive evidence of a mediation effect and we cannot explain the nature of the effect with certainty based on the available data. We had issues to reproduce the results of the second study using different software. For instance, there are differences between how the questionnaires are presented in the Supplemental material of the original study and the number and order of items in the dataset and log file. We completed the reproduction attempt after obtaining supplemental information from the authors of the original study. We were able to reproduce the main conclusions. . 22 References Baron, R. M., & Kenny, D. A. 1986. The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173-82. Bar-Tal, Daniel, Amiram Raviv, Alona Raviv, and Adi Dgani-Hirsh. 2009. The Influence of the Ethos of Conflict on Israeli Jews’ Interpretation of Jewish-Palestinian Encounters. Journal of Conflict Resolution 53 (1): 94–118. Bar-Tal, Daniel, Keren Sharvit, Eran Halperin, and Anat Zafran. 2012. Ethos of Conflict: The Concept and Its Measurement. Peace and Conflict: Journal of Peace Psychology 18 (1): 40–61. Batson, C. Daniel, Johee Chang, Ryan Orr, and Jennifer Rowland. 2002. Empathy, Attitudes, and Action: Can Feeling for a Member of a Stigmatized Group Motivate One to Help the Group? Personality and Social Psychology Bulletin 28 (12): 1656–66. Batson, C. Daniel, Marina P. Polycarpou, Eddie Harmon-Jones, Heidi J. Imhoff, Erin C. Mitchener, Lori L. Bednar, Tricia R. Klein, and Lori Highberger. 1997. Empathy and Attitudes: Can Feeling for a Member of a Stigmatized Group Improve Feelings toward the Group? Journal of Personality and Social Psychology 72 (1): 105–18. Elliot, Andrew J., and Patricia G. Devine. 1994. On the Motivational Nature of Cognitive Dissonance: Dissonance as Psychological Discomfort. Journal of Personality and Social Psychology 67 (3): 382–94. Festinger, Leon, and James M. Carlsmith. 1959. Cognitive Consequences of Forced Compliance. Journal of Abnormal and Social Psychology 58 (1): 203–10. Gubler, Joshua R.; Karpowitz, Christopher F.; Monson, J. Quin; Romney, David; South, Mikle. 2022. Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail. The Journal of Politics 84:4, 2156-2171. Gubler, Joshua R.; Karpowitz, Christopher F.; Monson, J. Quin; Romney, David; South, Mikle. 2021, Replication Data for "Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail”, https://doi.org/10.7910/DVN/FUCDTT, Harvard Dataverse, V1, UNF:6:jCDq9q0Fqss6N/zJZNjZsQ== [fileUNF] Haslam, Nick. 2006. Dehumanization: An Integrative Review. Personality and Social Psychology Review 10 (August): 252–64. Hayes, A. F. 2017. Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford publications. McGuire, W. J., & Papageorgis, D. 1961. The relative efficacy of various types of prior beliefdefense in producing immunity against persuasion. The Journal of Abnormal and Social Psychology, 62(2), 327-337. Papageorgis, D., & McGuire, W. J. 1961. The generality of immunity to persuasion produced by pre-exposure to weakened counterarguments. The Journal of Abnormal and Social Psychology, 62(3), 475-481. 5 Balance This section provides summaries of balance on covariates between treatment groups. Omnibus balance statistics are provided in Table E.6, while figures showing standardized differences for individual covariates are found in Figure E.7. These results indicate imbalances for gender and age for some treatments, but omnibus balance tests for the treatments indicate that we cannot reject the null of a balanced sample. ## Overall statistics (reported in the table) > baltest_1a$overall chisquare df p.value unstrat 11 14 0.72 > baltest_1b$overall chisquare df p.value unstrat 9.9 14 0.77 > baltest_1c$overall chisquare df p.value unstrat 20 14 0.14 6 Balance in Studies 1 and 2. Regression of Humanization on Treatments × Antipathy (Dichotomous), Controls, Study 1 % & (1) & (2) & (3) \\ Intercept & 0.51 ^{***} & 0.59 ^{***} & 0.67 ^{***} \\ & (0.01) & (0.01) & (0.03) \\ Humanization & 0.13 ^{***} & 0.10 ^{***} & 0.09 ^{***} \\ & (0.01) & (0.02) & (0.02) \\ Information & -0.03 ^* & -0.05 ^{**} & -0.05 ^{**} \\ & (0.01) & (0.02) & (0.02) \\ Combined & 0.13 ^{***} & 0.10 ^{***} & 0.10 ^{***} \\ & (0.01) & (0.02) & (0.02) \\ Outgroup Antipathy & & -0.15 ^{***} & -0.15 ^{***} \\ & & (0.02) & (0.02) \\ 7 Humanization $\times$ Antipathy & & 0.08 ^{**} & 0.08 ^{**} \\ & & (0.02) & (0.03) \\ Information $\times$ Antipathy & & 0.05 ^* & 0.05 ^\dagger \\ & & (0.02) & (0.02) \\ Combined $\times$ Antipathy & & 0.05 ^* & 0.06 ^* \\ & & (0.02) & (0.03) \\ Gender (1 = Female) & & & -0.04 ^{***} \\ & & & (0.01) \\ Age & & & -0.00 ^{***} \\ & & & (0.00) \\ Party ID (0--1) & & & 0.02 \\ & & & (0.02) \\ $N$ & 3309 & 3305 & 3134 \\ $R^2$ & 0.08 & 0.12 & 0.13 \\ adj. $R^2$ & 0.08 & 0.12 & 0.13 \\ Resid. sd & 0.26 & 0.25 & 0.25 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001  8 Regression of Empathic Concern on Treatments × Antipathy (Continuous), Controls, Study 1 % & (1) & (2) & (3) \\ Intercept & 0.27 ^{***} & 0.30 ^{***} & 0.20 ^{***} \\ & (0.01) & (0.02) & (0.03) \\ Humanization & 0.35 ^{***} & 0.56 ^{***} & 0.56 ^{***} \\ & (0.01) & (0.02) & (0.02) \\ Information & 0.09 ^{***} & 0.24 ^{***} & 0.24 ^{***} \\ & (0.01) & (0.02) & (0.02) \\ Combined & 0.34 ^{***} & 0.55 ^{***} & 0.55 ^{***} \\ & (0.01) & (0.02) & (0.02) \\ Outgroup Antipathy & & -0.05 ^\dagger & -0.07 ^* \\ & & (0.03) & (0.03) \\ Humanization $\times$ Antipathy & & -0.40 ^{***} & -0.40 ^{***} \\ & & (0.04) & (0.04) \\ Information $\times$ Antipathy & & -0.29 ^{***} & -0.29 ^{***} \\ & & (0.04) & (0.04) \\ Combined $\times$ Antipathy & & -0.40 ^{***} & -0.40 ^{***} \\ & & (0.04) & (0.04) \\ Gender (1 = Female) & & & -0.03 ^{***} \\ & & & (0.01) \\ Age & & & 0.00 ^{***} \\ & & & (0.00) \\ Party ID (0--1) & & & 0.05 ^{**} \\ & & & (0.02) \\ $N$ & 3439 & 3433 & 3239 \\ $R^2$ & 0.32 & 0.42 & 0.43 \\ adj. $R^2$ & 0.32 & 0.42 & 0.43 \\ Resid. sd & 0.23 & 0.21 & 0.21 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 9 Regression of Empathic Concern on Treatments × Antipathy (Dichotomous), Controls, Study 1 % & (1) & (2) & (3) \\ Intercept & 0.27 ^{***} & 0.28 ^{***} & 0.21 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Humanization & 0.35 ^{***} & 0.43 ^{***} & 0.44 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Information & 0.09 ^{***} & 0.14 ^{***} & 0.14 ^{***} \\ & (0.01) & (0.01) & (0.01) \\ Combined & 0.34 ^{***} & 0.42 ^{***} & 0.42 ^{***} \\ & (0.01) & (0.01) & (0.01) \\ Outgroup Antipathy & & -0.02 & -0.02 \\ & & (0.01) & (0.02) \\ Humanization $\times$ Antipathy & & -0.16 ^{***} & -0.16 ^{***}\\ & & (0.02) & (0.02) \\ Information $\times$ Antipathy & & -0.11 ^{***} & -0.11 ^{***}\\ & & (0.02) & (0.02) \\ Combined $\times$ Antipathy & & -0.16 ^{***} & -0.16 ^{***}\\ & & (0.02) & (0.02) \\ Gender (1 = Female) & & & -0.04 ^{***}\\ & & & (0.01) \\ Age & & & 0.00 ^{***} \\ & & & (0.00) \\ Party ID (0--1) & & & 0.01 \\ & & & (0.02) \\ $N$ & 3439 & 3433 & 3239 \\ $R^2$ & 0.32 & 0.38 & 0.40 \\ adj. $R^2$ & 0.32 & 0.38 & 0.39 \\ Resid. sd & 0.23 & 0.22 & 0.21 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 10 ChangingHearts,Study2 This section provides supporting statistics and tables for the figures shown in the Study 2 subsection of the section titled “Changing Hearts: Humanization and Empathy. Regression of Empathic Concern on Treatments × Antipathy (Dichotomous), Controls, Study 2 % & (1) & (2) & (3) \\ Intercept & 0.63 ^{***} & 0.69 ^{***} & 0.65 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Illegal Condition & -0.04 ^{***} & -0.02 ^* & -0.02 ^* \\ & (0.01) & (0.01) & (0.01) \\ Outgroup Antipathy & & -0.14 ^{***} & -0.14 ^{***}\\ & & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & & -0.05 ^{**} & -0.05 ^{**} \\ & & (0.02) & (0.02) \\ Gender (1 = Female) & & & 0.04 ^{***} \\ & & & (0.01) \\ Age & & & 0.00 ^{***} \\ & & & (0.00) \\ Party ID (0--1) & & & -0.03 \\ & & & (0.02) \\ $N$ & 1977 & 1977 & 1962 \\ $R^2$ & 0.01 & 0.16 & 0.17 \\ adj. $R^2$ & 0.01 & 0.16 & 0.17 \\ Resid. sd & 0.22 & 0.20 & 0.20 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001  11 DissonanceasaMechanism This section provides supporting statistics and a table for the figure shown in the section titled “Dissonance as a Mechanism.” Regression of Dissonance on Treatments × Antipathy (Dichotomous), Controls, Study 2 % & (1) & (2) & (3) \\ Intercept & 0.27 ^{***} & 0.24 ^{***} & 0.30 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Illegal Condition & 0.04 ^{***} & 0.02 ^\dagger & 0.02 ^* \\ & (0.01) & (0.01) & (0.01) \\ Outgroup Antipathy & & 0.05 ^{***} & 0.06 ^{***} \\ & & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & & 0.05 ^* & 0.05 ^* \\ & & (0.02) & (0.02) \\ Gender (1 = Female) & & & -0.03 ^{**} \\ & & & (0.01) \\ Age & & & -0.00 ^{***} \\ & & & (0.00) \\ Party ID (0--1) & & & -0.01 \\ & & & (0.02) \\ $N$ & 1982 & 1982 & 1966 \\ $R^2$ & 0.01 & 0.04 & 0.05 \\ adj. $R^2$ & 0.01 & 0.04 & 0.05 \\ Resid. sd & 0.22 & 0.22 & 0.22 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 12 ChangingMindsaboutPolicy This section provides supporting tables for the results in the section of the paper titled “Changing Minds about Policy.” Tables in the following, provide an estimation of the models with control variables in addition to what is shown in the paper. Tables G.15 and G.16, on the other hand, show the same models with our dichotomous measure of outgroup antipathy. Regression of Policy Harm on Treatments × Antipathy (Continuous), Controls, Study 1 % & (1) & (2) & (3) \\ Intercept & 0.71 ^{***} & 0.57 ^{***} & 0.39 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Humanization & -0.01 & -0.00 & -0.00 \\ & (0.01) & (0.01) & (0.01) \\ Information & 0.01 & 0.03 ^* & 0.03 ^* \\ & (0.01) & (0.01) & (0.01) \\ Combined & -0.01 & 0.01 & 0.00 \\ & (0.01) & (0.01) & (0.01) \\ Outgroup Antipathy & & 0.27 ^{***} & 0.25 ^{***} \\ & & (0.01) & (0.01) \\ Humanization $\times$ Antipathy & & -0.01 & -0.01 \\ & & (0.02) & (0.02) \\ Information $\times$ Antipathy & & -0.03 ^\dagger & -0.03 ^\dagger \\ & & (0.02) & (0.02) \\ Combined $\times$ Antipathy & & -0.02 & -0.02 \\ & & (0.02) & (0.02) \\ Gender (1 = Female) & & & 0.00 \\ & & & (0.01) \\ Age & & & 0.00 ^* \\ & & & (0.00) \\ Party ID (0--1) & & & 0.19 ^{***} \\ & & & (0.01) \\ $N$ & 3489 & 3482 & 3281 \\ $R^2$ & 0.00 & 0.33 & 0.36 \\ adj. $R^2$ & 0.00 & 0.33 & 0.36 \\ Resid. sd & 0.22 & 0.18 & 0.18 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 13 Regression of Policy Harm on Treatments × Antipathy (Continuous), Controls, Study 2 % & (1) & (2) & (3) \\ Intercept & 0.62 ^{***} & 0.52 ^{***} & 0.36 ^{***}\\ & (0.01) & (0.01) & (0.02) \\ Illegal Condition & -0.03 ^{**} & -0.03 ^{**} & -0.03 ^{**}\\ & (0.01) & (0.01) & (0.01) \\ Outgroup Antipathy & & 0.25 ^{***} & 0.22 ^{***}\\ & & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & & 0.02 & 0.01 \\ & & (0.02) & (0.02) \\ Gender (1 = Female) & & & -0.02 ^{**}\\ & & & (0.01) \\ Age & & & 0.00 ^{***}\\ & & & (0.00) \\ Party ID (0--1) & & & 0.20 ^{***}\\ & & & (0.02) \\ $N$ & 1982 & 1982 & 1966 \\ $R^2$ & 0.00 & 0.30 & 0.35 \\ adj. $R^2$ & 0.00 & 0.30 & 0.35 \\ Resid. sd & 0.23 & 0.19 & 0.19 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 14 Regression of Policy Harm on Treatments × Antipathy (Dichotomous), Controls, Study 1 % & (1) & (2) & (3) \\ Intercept & 0.71 ^{***} & 0.36 ^{***} & 0.26 ^{***} \\ & (0.01) & (0.01) & (0.02) \\ Humanization & -0.01 & -0.00 & -0.01 \\ & (0.01) & (0.02) & (0.02) \\ Information & 0.01 & 0.04 ^* & 0.04 ^* \\ & (0.01) & (0.02) & (0.02) \\ Combined & -0.01 & -0.00 & -0.00 \\ & (0.01) & (0.02) & (0.02) \\ Outgroup Antipathy & & 0.67 ^{***} & 0.65 ^{***} \\ & & (0.02) & (0.02) \\ Humanization $\times$ Antipathy & & -0.01 & -0.01 \\ & & (0.03) & (0.03) \\ Information $\times$ Antipathy & & -0.06 ^\dagger & -0.06 ^\dagger \\ & & (0.03) & (0.03) \\ Combined $\times$ Antipathy & & -0.01 & -0.01 \\ & & (0.03) & (0.03) \\ Gender (1 = Female) & & & -0.00 \\ & & & (0.01) \\ Age & & & 0.00 \\ & & & (0.00) \\ Party ID (0--1) & & & 0.12 ^{***} \\ & & & (0.01) \\ $N$ & 3489 & 3482 & 3281 \\ $R^2$ & 0.00 & 0.51 & 0.52 \\ adj. $R^2$ & 0.00 & 0.51 & 0.52 \\ Resid. sd & 0.22 & 0.15 & 0.15 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 21 Figure showing the marginal effects of the treatments on empathic concern and policy harm, by Party ID, for Study 2. Rug plot of Party ID included; bars represent 95% confidence intervals.  22 Study2Resultswith3‐ItemAntipathyMeasure This section provides results from study 2 with a 3-item antipathy measure and compares them to the original 9-item measure in Tables H.21, H.22, and H.23. Results are almost identical with either measure. Regression of Empathic Concern on Pre-Treatment Antipathy and Treatments, Study 2, 3vs. 9Item Antipathy Measure % & 3-Item & 9-Item \\ Intercept & 0.69 ^{***} & 0.69 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition & -0.02 ^\dagger & -0.02 ^* \\ & (0.01) & (0.01) \\ Outgroup Antipathy & -0.14 ^{***} & -0.14 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & -0.05 ^* & -0.05 ^{**} \\ & (0.02) & (0.02) \\ $N$ & 1977 & 1977 \\ $R^2$ & 0.15 & 0.16 \\ adj. $R^2$ & 0.14 & 0.16 \\ Resid. sd & 0.20 & 0.20 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 Regression of Dissonance on Pre-Treatment Antipathy and Treatments, Study 2, 3vs. 9-Item Antipathy Measure % & 3-Item & 9-Item \\ Intercept & 0.25 ^{***} & 0.24 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition & 0.02 ^\dagger & 0.02 ^\dagger \\ & (0.01) & (0.01) \\ Outgroup Antipathy & 0.03 ^* & 0.05 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & 0.04 ^* & 0.05 ^* \\ & (0.02) & (0.02) \\ $N$ & 1982 & 1982 \\ $R^2$ & 0.03 & 0.04 \\ adj. $R^2$ & 0.02 & 0.04 \\ Resid. sd & 0.22 & 0.22 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 Regression of Policy Harm on Pre-Treatment Antipathy and Treatments, Study 2, 3vs. 9-Item Antipathy Measure % & 3-Item & 9-Item \\ Intercept & 0.52 ^{***} & 0.52 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition & -0.04 ^{***} & -0.03 ^{**} \\ & (0.01) & (0.01) \\ Outgroup Antipathy & 0.24 ^{***} & 0.25 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & 0.02 & 0.02 \\ & (0.02) & (0.02) \\ $N$ & 1982 & 1982 \\ $R^2$ & 0.27 & 0.30 \\ adj. $R^2$ & 0.27 & 0.30 \\ Resid. sd & 0.20 & 0.19 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 23 ResultsforSeparatePolicyOutcomes  This section break down our Minds about Policy” results by the different policy components of the outcome measure in the following table: % & Law (English) & Law (Tuition) & Law (Welfare) & Law (Hire) & Imm. Opinion & AZ Law & State Bill Harm \\ Intercept & 0.58 ^{***} & 0.56 ^{***} & 0.62 ^{***} & 0.57 ^{***} & 0.37 ^{***} & 0.53 ^{***} & 0.78 ^{***} \\ & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) \\ Humanization & -0.00 & -0.04 ^\dagger & -0.00 & 0.01 & 0.01 & 0.01 & 0.01 \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.01) \\ Information & -0.01 & 0.02 & 0.02 & 0.04 ^* & 0.04 ^* & 0.05 ^* & 0.03 ^* \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.01) \\ Combined & -0.03 & -0.01 & 0.02 & 0.01 & 0.01 & 0.01 & 0.03 ^\dagger \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.01) \\ Outgroup Antipathy & 0.27 ^{***} & 0.32 ^{***} & 0.26 ^{***} & 0.26 ^{***} & 0.28 ^{***} & 0.34 ^{***} & 0.16 ^{***} \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.01) \\ Antipathy $\times$ Humanization & -0.03 & -0.01 & -0.01 & 0.00 & -0.02 & -0.00 & -0.01 \\ & (0.03) & (0.03) & (0.03) & (0.03) & (0.03) & (0.03) & (0.02) \\ Antipathy $\times$ Information & -0.00 & -0.02 & -0.06 ^* & -0.03 & -0.04 & -0.04 & -0.02 \\ & (0.03) & (0.03) & (0.03) & (0.03) & (0.02) & (0.03) & (0.02) \\ Antipathy $\times$ Combined & 0.01 & -0.02 & -0.05 ^\dagger & -0.01 & -0.03 & - 0.02 & -0.02 \\ & (0.03) & (0.03) & (0.03) & (0.03) & (0.03) & (0.03) & (0.02) \\ $N$ & 3477 & 3476 & 3477 & 3476 & 3409 & 3478 & 3474 \\ $R^2$ & 0.18 & 0.23 & 0.15 & 0.17 & 0.21 & 0.27 & 0.12 \\ adj. $R^2$ & 0.18 & 0.23 & 0.15 & 0.17 & 0.20 & 0.27 & 0.12 \\ Resid. sd & 0.29 & 0.28 & 0.27 & 0.27 & 0.26 & 0.27 & 0.20 \\ \hline 24 Regression of Separate Policy Outcomes on Antipathy and Treatments, Study 2 % & Law (English) & Law (Tuition) & Law (Welfare) & Law (Hire) & Imm. Opinion & Aid Illegal & Take Resources & Deny Rights \\ Intercept & 0.39 ^{***} & 0.48 ^{***} & 0.55 ^{***} & 0.48 ^{***} & 0.38 ^{***} & 0.74 ^{***} & 0.39 ^{***} & 0.33 ^{***} \\ & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) & (0.01) \\ Illegal & -0.02 & -0.02 & -0.03 ^\dagger & -0.04 ^{**} & -0.05 ^{**} & -0.04 ^{**} & -0.05 ^{***} & -0.01 \\ & (0.02) & (0.01) & (0.01) & (0.01) & (0.02) & (0.01) & (0.01) & (0.01) \\ Antipathy & 0.21 ^{***} & 0.22 ^{***} & 0.16 ^{***} & 0.19 ^{***} & 0.28 ^{***} & 0.16 ^{***} & 0.27 ^{***} & 0.27 ^{***} \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) & (0.02) \\ Illegal $\times$ Antipathy & 0.03 & -0.01 & 0.04 ^\dagger & 0.03 & 0.02 & 0.03 & 0.03 & -0.03 \\ & (0.02) & (0.02) & (0.02) & (0.02) & (0.03) & (0.02) & (0.02) & (0.02) \\ $N$ & 1981 & 1981 & 1982 & 1981 & 1972 & 1977 & 1978 & 1977 \\ $R^2$ & 0.15 & 0.14 & 0.12 & 0.16 & 0.20 & 0.13 & 0.26 & 0.19 \\ adj. $R^2$ & 0.15 & 0.14 & 0.12 & 0.16 & 0.20 & 0.13 & 0.26 & 0.19 \\ Resid. sd & 0.26 & 0.25 & 0.23 & 0.23 & 0.28 & 0.23 & 0.24 & 0.25 \\ \hline 25 ResultsUsingCommonPolicyOutcomes This section replicates the “Changing Minds about Policy” results while only using the five survey questions contained in both surveys. As can be seen in Tables H.26 and H.27, the results are almost identical. Regression of Policy Harm on Antipathy and Treatments, Study 1, Common Items vs. Full Scale % & Common Policy Items & Full Policy Scale \\ Intercept & 0.54 ^{***} & 0.57 ^{***} \\ & (0.01) & (0.01) \\ Humanization & -0.01 & -0.00 \\ & (0.01) & (0.01) \\ Information & 0.02 & 0.03 ^* \\ & (0.01) & (0.01) \\ Combined & 0.00 & 0.01 \\ & (0.01) & (0.01) \\ Outgroup Antipathy & 0.28 ^{***} & 0.27 ^{***} \\ & (0.01) & (0.01) \\ Humanization $\times$ Antipathy & -0.01 & -0.01 \\ & (0.02) & (0.02) \\ Information $\times$ Antipathy & -0.03 ^\dagger & -0.03 ^\dagger \\ & (0.02) & (0.02) \\ Combined $\times$ Antipathy & -0.02 & -0.02 \\ & (0.02) & (0.02) \\ $N$ & 3481 & 3482 \\ $R^2$ & 0.31 & 0.33 \\ adj. $R^2$ & 0.31 & 0.33 \\ Resid. sd & 0.20 & 0.18 \\ \hline 26 Regression of Policy Harm on Antipathy and Treatments, Study 2, Common Items vs. Full Scale % & Common Policy Items & Full Policy Scale \\ Intercept & 0.53 ^{***} & 0.52 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition & -0.03 ^{**} & -0.03 ^{**} \\ & (0.01) & (0.01) \\ Outgroup Antipathy & 0.23 ^{***} & 0.25 ^{***} \\ & (0.01) & (0.01) \\ Illegal Condition $\times$ Antipathy & 0.02 & 0.02 \\ & (0.02) & (0.02) \\ $N$ & 1982 & 1982 \\ $R^2$ & 0.25 & 0.30 \\ adj. $R^2$ & 0.25 & 0.30 \\ Resid. sd & 0.21 & 0.19 \\ \hline Standard errors in parentheses † significant at p< .10; ∗p< .05; ∗∗p< .01; ∗∗∗p< .001 27 RelationshipbetweenEmpathicConcernandSupportforHarmfulPolicies Following figures show the correlation between empathic concern and support for harmful policies in studies 1 and 2, respectively. Note the strong negative correlation across treatments and across low vs. high antipathy, with the exception of the control condition in study 1. Relationship between post-treatment empathic concern and post-treatment support for harmful policies, with regression lines, study 1. Relationship between post-treatment empathic concern and post-treatment support for harmful policies, with regression lines, study 2. 28 LinearityandBinningofMarginalEffects The results of our analyses are very reliant on the presence of heterogeneous treatment effects. For the sake of simplicity in interpretation, we usually opt to present these heterogeneous by binning participants into low and high antipathy groups in the paper. However, recent research (Hainmueller et al. 2019) indicates that estimates from multiplicative interaction models like ours can, at times, be highly dependent on binning choices. For this reason, we use the interflex package in R to examine what our main marginal effects would look like with a kernel estimate, two bins (the analysis used in the paper), and three bins. As seen in Figures H.14, H.15, and H.16, the moderating effect of outgroup antipathy is highly linear in nature and the choice of number of bins has little effect on the substantive conclusions drawn from our analyses. 29 Marginal effects on empathic concern from study 1, kernel estimates and tests with two and three bins Marginal effects on empathic concern from study 2, kernel estimates and tests with two and three bins. Marginal effects on dissonance from study 2, kernel estimates and tests with two and three bins 1 Appendix 2: Computational replication in SPSS SPSS syntax and outputs of reproduction and robustness checks for Study 1 Syntax for SPSS analyses *Recoding items to numeric values. Set "99" as a missing value. RECODE i_admire i_love i_resent i_shame i_excite i_plea i_fear i_anger e_sym e_moved e_com e_warm e_soft e_tender d_uncom d_angry d_shame d_uneasy d_friend d_disgust d_emba d_bother d_opti d_annoy d_tense d_disa d_happy d_ener d_concern d_good ('NA'='99'). EXECUTE. *Now, e_sym to d_good were changed to "Numeric" manually. RECODE m_less ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree nor '+ 'Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO OD1. VARIABLE LABELS OD1 'In general, illegal immigrants...'. RECODE m_learn ('Strongly Disagree'=7) ('Disagree'=6) ('Somewhat Disagree'=5) ('Neither Agree nor Disagree'=4) ('Somewhat Agree'=3) ('Agree'=2) ('Strongly Agree'=1) (ELSE=99) INTO IG1_recoded. VARIABLE LABELS IG1_recoded 'Illegal immigrants have moral...'. RECODE m_suffer ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree '+ 'nor Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO IVO1. VARIABLE LABELS IVO1 'Legal residents...'. RECODE law_english ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree or '+ 'Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO law1_english. VARIABLE LABELS law1_english '...documents in English only...'. RECODE law_tuition ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree or '+ 'Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO law2_tuition. VARIABLE LABELS law2_tuition '...pay out-of-state tuition...'. RECODE law_welfare ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree or '+ 'Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO law3_welfare. VARIABLE LABELS law3_welfare '...restricting welfare support...'. RECODE law_hire ('Strongly Disagree'=1) ('Disagree'=2) ('Somewhat Disagree'=3) ('Neither Agree or '+ 'Disagree'=4) ('Somewhat Agree'=5) ('Agree'=6) ('Strongly Agree'=7) (ELSE=99) INTO law4_hire. VARIABLE LABELS law4_hire '...increasing the penalties...who hire...'. RECODE immig_opinion ('Illegal immigrants should be required to go home immediately.'=1) ('Most illegal immigrants should be required to go home, but some should be allowed to remain in the U.S. under a temporary guest worker program.'=2) ('Most illegal immigrants should be allowed to stay in the U.S. but only as temporary workers who must eventually return home.'=3) ('Illegal immigrants should be allowed to stay permanently in the U.S.'=4) (ELSE=99) INTO immig_opinion_nr. 8 Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Chan g e Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,276a ,076 ,075 ,25491 ,076 86,466 33146 <,001 2 ,347b ,120 ,118 ,24891 ,044 39,373 43142 <,001 3 ,361c ,131 ,128 ,24757 ,010 12,427 33139 <,001 a. Predictors: ( Constant ) , Treatment: Humanization + Information, Treatment: humanization, Treatment: Information b. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, Antipathy dichotomous, comxantd, humxantd, infxantd c. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, Antipathy dichotomous, comxantd, humxantd, infxantd, Male=1, Other=0, a g e, part_id 9 A NOVAa Model Sum of Squares d f Mean Square F Si g . 1 Re g ression 16,856 35,619 86,466 <,001b Residual 204,431 3146 ,065 Total 221,287 3149 2 Re g ression 26,614 73,802 61,364 <,001c Residual 194,673 3142 ,062 Total 221,287 3149 3 Regression 28,899 10 2,890 47,152 <,001d Residual 192,388 3139 ,061 Total 221,287 3149 a. Dependent Variable: pos_em b. Predictors: ( Constant ) , Treatment: Humanization + Information, Treatment: humanization, Treatment: Information c. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, Antipathy dichotomous, comxantd, humxantd, infxantd d. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, Antipathy dichotomous, comxantd, humxantd, infxantd, Male=1, Other=0, age, part_id 10 Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 ( Constant ) ,515 ,009 55,650 <,001 Treatment: humanization ,131 ,013 ,212 10,017 <,001 Treatment: Information -,026 ,013 -,043 -2,024 ,043 Treatment: Humanization + Information ,132 ,013 ,215 10,172 <,001 2 ( Constant ) ,588 ,013 46,922 <,001 Treatment: humanization ,095 ,018 ,153 5,348 <,001 Treatment: Information -,046 ,017 -,077 -2,656 ,008 Treatment: Humanization + Information ,100 ,017 ,163 5,769 <,001 Antipathy dichotomous -,153 ,018 -,289 -8,483 <,001 humxantd ,075 ,026 ,090 2,926 ,003 infxantd ,046 ,025 ,059 1,838 ,066 comxantd ,060 ,025 ,071 2,368 ,018 3 ( Constant ) ,676 ,027 25,166 <,001 Treatment: humanization ,096 ,018 ,155 5,438 <,001 Treatment: Information -,046 ,017 -,077 -2,653 ,008 Treatment: Humanization + Information ,097 ,017 ,158 5,624 <,001 Antipath y dichotomous -,148 ,018 -,279 -8,220 <,001 humxantd ,074 ,025 ,089 2,916 ,004 infxantd ,045 ,025 ,058 1,821 ,069 comxantd ,059 ,025 ,069 2,309 ,021 Male=1, Other=0 -,035 ,009 -,064 -3,821 <,001 a g e-,002 ,000 -,077 -4,556 <,001 part_id ,016 ,021 ,013 ,776 ,438 11 a. Dependent Variable: pos_em Excluded Variablesa Model Beta In tSi g .Partial Correlation Collinearity Statistics Tolerance 1 Antipathy dichotomous -,204b-12,147 <,001 -,212 ,999 humxantd -,095b-4,270 <,001 -,076 ,591 infxantd -,138b-6,190 <,001 -,110 ,581 comxantd -,110b-5,074 <,001 -,090 ,619 Male=1, Other=0 -,076b-4,466 <,001 -,079 ,999 age -,098b-5,718 <,001 -,101 ,996 part_id -,023b-1,329 ,184 -,024 1,000 2 Male=1, Other=0 -,067c-4,012 <,001 -,071 ,997 a g e-,079c-4,709 <,001 -,084 ,986 part_id ,007c,424 ,671 ,008 ,978 a. Dependent Variable: pos_em b. Predictors in the Model: ( Constant ) , Treatment: Humanization + Information, Treatment: humanization, Treatment: Information c. Predictors in the Model: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, Antipathy dichotomous, comxantd, humxantd, infxantd Notes Output Created 14-JUL-2023 13:52:04 Comments Input Active Dataset DataSet1 Filter keep=4 (FILTER) Weight <none> Split File <none> 12 N of Rows in Working Data File 3514 Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA CHANGE /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT pos_em /METHOD=ENTER t_hum t_inf t_comb /METHOD=ENTER antipath humxant infxant comxant /METHOD=ENTER Male age part_id. Resources Processor Time 00:00:00,08 Elapsed Time 00:00:00,08 Memory Required 19568 bytes Additional Memory Required for Residual Plots 0 bytes Variables Entered/Removeda Model Variables Entered Variables Removed Method 1 Treatment: Humanization + Information, Treatment: humanization, Treatment: Informationb .Enter 2 antipath, comxant, humxant, infxantb .Enter 3 Male=1, Other=0, age, part_idb .Enter a. Dependent Variable: pos_em b. All requested variables entered. 13 Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change 1 ,276a ,076 ,075 ,25491 ,076 86,466 2 ,381b ,145 ,143 ,24536 ,069 63,475 3 ,394c ,155 ,152 ,24405 ,010 12,258 Model Summary Model Change Statistics df1 df2 Sig. F Change 1 3 3146 <,001 2 4 3142 <,001 3 3 3139 <,001 a. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information b. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, antipath, comxant, humxant, infxant c. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, antipath, comxant, humxant, infxant, Male=1, Other=0, age, part_id ANOVAa Model Sum of Squares df Mean Square F Sig. 1 Regression 16,856 35,619 86,466 <,001b Residual 204,431 3146 ,065 Total 221,287 3149 2 Regression 32,141 74,592 76,272 <,001c Residual 189,146 3142 ,060 Total 221,287 3149 3 Regression 34,331 10 3,433 57,642 <,001d Residual 186,956 3139 ,060 Total 221,287 3149 a. Dependent Variable: pos_em b. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information c. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, 14 antipath, comxant, humxant, infxant d. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, antipath, comxant, humxant, infxant, Male=1, Other=0, age, part_id Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t B Std. Error Beta 1 (Constant) ,515 ,009 55,650 Treatment: humanization ,131 ,013 ,212 10,017 Treatment: Information -,026 ,013 -,043 -2,024 Treatment: Humanization + Information ,132 ,013 ,215 10,172 2 (Constant) ,711 ,021 33,925 Treatment: humanization ,046 ,029 ,075 1,567 Treatment: Information -,074 ,029 -,124 -2,558 Treatment: Humanization + Information ,063 ,029 ,103 2,150 antipath -,381 ,037 -,346 -10,346 humxant ,163 ,052 ,154 3,145 infxant ,099 ,051 ,099 1,959 comxant ,128 ,052 ,119 2,440 3 (Constant) ,764 ,031 24,895 Treatment: humanization ,048 ,029 ,077 1,623 Treatment: Information -,073 ,029 -,123 -2,544 Treatment: Humanization + Information ,062 ,029 ,100 2,100 antipath -,377 ,037 -,342 -10,223 humxant ,162 ,052 ,153 3,133 infxant ,098 ,050 ,097 1,932 comxant ,124 ,052 ,115 2,372 Male=1, Other=0 -,033 ,009 -,060 -3,633 age -,001 ,000 -,070 -4,237 part_id ,047 ,021 ,039 2,300 Coefficientsa Model Sig. 15 1 (Constant) <,001 Treatment: humanization <,001 Treatment: Information ,043 Treatment: Humanization + Information <,001 2 (Constant) <,001 Treatment: humanization ,117 Treatment: Information ,011 Treatment: Humanization + Information ,032 antipath <,001 humxant ,002 infxant ,050 comxant ,015 3 (Constant) <,001 Treatment: humanization ,105 Treatment: Information ,011 Treatment: Humanization + Information ,036 antipath <,001 humxant ,002 infxant ,053 comxant ,018 Male=1, Other=0 <,001 age <,001 part_id ,022 a. Dependent Variable: pos_em Excluded Variablesa Model Beta In t Sig. Partial Correlation Collinearity Statistics Tolerance 1 antipath -,257b-15,569 <,001 -,267 ,999 humxant -,206b-5,763 <,001 -,102 ,229 infxant -,281b-7,863 <,001 -,139 ,225 comxant -,235b-6,565 <,001 -,116 ,226 Male=1, Other=0 -,076b-4,466 <,001 -,079 ,999 age -,098b-5,718 <,001 -,101 ,996 16 part_id -,023b-1,329 ,184 -,024 1,000 2 Male=1, Other=0 -,062c-3,755 <,001 -,067 ,995 age -,072c-4,323 <,001 -,077 ,984 part_id ,034c2,018 ,044 ,036 ,952 a. Dependent Variable: pos_em b. Predictors in the Model: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information c. Predictors in the Model: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information, antipath, comxant, humxant, infxant Regression Notes Output Created 14-JUL-2023 13:52:04 Comments Input Active Dataset DataSet1 Filter keep=4 (FILTER) Weight <none> Split File <none> N of Rows in Working Data File 3514 Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT e_conc /METHOD=ENTER t_hum t_inf t_comb. Resources Processor Time 00:00:00,08 Elapsed Time 00:00:00,07 Memory Required 13904 bytes Additional Memory Required for Residual Plots 0 bytes 17 Variables Entered/Removeda Model Variables Entered Variables Removed Method 1 Treatment: Humanization + Information, Treatment: humanization, Treatment: Informationb .Enter a. Dependent Variable: e_conc b. All requested variables entered. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,561a ,315 ,314 ,22798 a. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information ANOVAa Model Sum of Squares df Mean Square F Sig. 1 Regression 82,445 327,482 528,732 <,001b Residual 179,372 3451 ,052 Total 261,817 3454 a. Dependent Variable: e_conc b. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t B Std. Error Beta 1 (Constant) ,269 ,008 34,234 Treatment: humanization ,354 ,011 ,550 31,752 Treatment: Information ,086 ,011 ,139 7,928 Treatment: Humanization + Information ,342 ,011 ,535 30,844 Coefficientsa Model Sig. 24 Information, Treatment: humanization, Treatment: Informationb a. Dependent Variable: harm b. All requested variables entered. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change 1 ,048a ,002 ,001 ,22089 ,002 2,697 Model Summary Model Change Statistics df1 df2 Sig. F Change 1 3 3501 ,044 a. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information ANOVAa Model Sum of Squares df Mean Square F Sig. 1 Regression ,395 3,132 2,697 ,044b Residual 170,827 3501 ,049 Total 171,222 3504 a. Dependent Variable: harm b. Predictors: (Constant), Treatment: Humanization + Information, Treatment: humanization, Treatment: Information Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t B Std. Error Beta 1 (Constant) ,706 ,008 93,643 Treatment: humanization -,010 ,011 -,018 -,890 Treatment: Information ,014 ,010 ,028 1,342 Treatment: Humanization + Information -,013 ,011 -,025 -1,213 Coefficientsa 25 Model Sig. 1 (Constant) <,001 Treatment: humanization ,373 Treatment: Information ,180 Treatment: Humanization + Information ,225 a. Dependent Variable: harm Regression Notes Output Created 14-JUL-2023 13:52:05 Comments Input Active Dataset DataSet1 Filter keep=4 (FILTER) Weight <none> Split File <none> N of Rows in Working Data File 3514 Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA CHANGE /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT harm /METHOD=ENTER t_hum t_inf t_comb antipath humxant infxant comxant. Resources Processor Time 00:00:00,09 26 Elapsed Time 00:00:00,10 Memory Required 16624 bytes Additional Memory Required for Residual Plots 0 bytes Variables Entered/Removeda Model Variables Entered Variables Removed Method 1 comxant, antipath, Treatment: humanization, Treatment: Information, Treatment: Humanization + Information, humxant, infxantb .Enter a. Dependent Variable: harm b. All requested variables entered. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change 1 ,713a ,508 ,507 ,15522 ,508 515,783 Model Summary Model Change Statistics df1 df2 Sig. F Change 1 7 3490 <,001 a. Predictors: (Constant), comxant, antipath, Treatment: humanization, Treatment: Information, Treatment: Humanization + Information, humxant, infxant ANOVAa Model Sum of Squares df Mean Square F Sig. 1 Regression 86,989 712,427 515,783 <,001b Residual 84,086 3490 ,024 Total 171,076 3497 a. Dependent Variable: harm b. Predictors: (Constant), comxant, antipath, Treatment: humanization, Treatment: Information, Treatment: Humanization + 27 Information, humxant, infxant Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t B Std. Error Beta 1 (Constant) ,356 ,013 28,285 Treatment: humanization -,005 ,018 -,010 -,285 Treatment: Information ,035 ,018 ,071 2,004 Treatment: Humanization + Information ,000 ,018 ,000 -,011 antipath ,674 ,022 ,732 30,608 humxant -,010 ,031 -,011 -,305 infxant -,053 ,030 -,064 -1,747 comxant -,011 ,031 -,012 -,336 Coefficientsa Model Sig. 1 (Constant) <,001 Treatment: humanization ,776 Treatment: Information ,045 Treatment: Humanization + Information ,991 antipath <,001 humxant ,760 infxant ,081 comxant ,737 a. Dependent Variable: harm Regression Notes Output Created 14-JUL-2023 13:52:05 28 Comments Input Active Dataset DataSet1 Filter keep=4 (FILTER) Weight <none> Split File <none> N of Rows in Working Data File 3514 Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA CHANGE /CRITERIA=PIN(.05) POUT(.10) /NOORIGIN /DEPENDENT e_conc /METHOD=ENTER Male age part_id antipath /METHOD=ENTER t_hum t_inf t_comb /METHOD=ENTER humxant /METHOD=ENTER comxant. Resources Processor Time 00:00:00,11 Elapsed Time 00:00:00,13 Memory Required 18720 bytes Additional Memory Required for Residual Plots 0 bytes Variables Entered/Removeda Model Variables Entered Variables Removed Method 1 antipath, Male=1, Other=0, age, part_idb .Enter 2 Treatment: Information, Treatment: humanization, Treatment: Humanization + Informationb .Enter 29 3 humxantb .Enter 4 comxantb .Enter a. Dependent Variable: e_conc b. All requested variables entered. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change 1 ,309a ,096 ,094 ,26124 ,096 85,801 2 ,642b ,413 ,411 ,21064 ,317 584,053 3 ,645c ,416 ,415 ,20999 ,004 21,060 4 ,651d ,424 ,422 ,20865 ,008 42,927 Model Summary Model Change Statistics df1 df2 Sig. F Change 1 4 3250 <,001 2 3 3247 <,001 3 1 3246 <,001 4 1 3245 <,001 a. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id b. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information c. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information, humxant d. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information, humxant, comxant ANOVAa Model Sum of Squares df Mean Square F Sig. 1 Regression 23,423 45,856 85,801 <,001b Residual 221,804 3250 ,068 Total 245,226 3254 30 2 Regression 101,163 714,452 325,725 <,001c Residual 144,063 3247 ,044 Total 245,226 3254 3 Regression 102,091 812,761 289,402 <,001d Residual 143,135 3246 ,044 Total 245,226 3254 4 Regression 103,960 911,551 265,339 <,001e Residual 141,266 3245 ,044 Total 245,226 3254 a. Dependent Variable: e_conc b. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id c. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information d. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information, humxant e. Predictors: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information, humxant, comxant Coefficientsa Model Unstandardized Coefficients Standardized Coefficients t B Std. Error Beta 1 (Constant) ,552 ,026 21,394 Male=1, Other=0 -,031 ,010 -,055 -3,266 age ,001 ,000 ,063 3,753 part_id ,057 ,022 ,045 2,623 antipath -,353 ,020 -,308 -17,942 2 (Constant) ,338 ,022 15,458 Male=1, Other=0 -,032 ,008 -,056 -4,137 age ,002 ,000 ,080 5,916 part_id ,051 ,018 ,040 2,885 antipath -,340 ,016 -,297 -21,408 Treatment: humanization ,358 ,011 ,559 33,845 Treatment: Information ,091 ,010 ,146 8,801 Treatment: Humanization + Information ,341 ,011 ,535 32,343 3 (Constant) ,317 ,022 14,214 Male=1, Other=0 -,032 ,008 -,056 -4,187 age ,002 ,000 ,081 5,975 part_id ,051 ,018 ,040 2,893 31 antipath -,300 ,018 -,262 -16,550 Treatment: humanization ,442 ,021 ,690 20,885 Treatment: Information ,091 ,010 ,146 8,791 Treatment: Humanization + Information ,342 ,011 ,536 32,489 humxant -,164 ,036 -,149 -4,589 4 (Constant) ,276 ,023 11,980 Male=1, Other=0 -,033 ,008 -,058 -4,333 age ,002 ,000 ,081 6,030 part_id ,051 ,017 ,040 2,947 antipath -,221 ,022 -,192 -10,178 Treatment: humanization ,483 ,022 ,754 22,015 Treatment: Information ,090 ,010 ,145 8,775 Treatment: Humanization + Information ,468 ,022 ,733 21,397 humxant -,243 ,037 -,221 -6,486 comxant -,247 ,038 -,223 -6,552 Coefficientsa Model Sig. 1 (Constant) <,001 Male=1, Other=0 ,001 age <,001 part_id ,009 antipath <,001 2 (Constant) <,001 Male=1, Other=0 <,001 age <,001 part_id ,004 antipath <,001 Treatment: humanization <,001 Treatment: Information <,001 Treatment: Humanization + Information <,001 3 (Constant) <,001 Male=1, Other=0 <,001 age <,001 32 part_id ,004 antipath <,001 Treatment: humanization <,001 Treatment: Information <,001 Treatment: Humanization + Information <,001 humxant <,001 4 (Constant) <,001 Male=1, Other=0 <,001 age <,001 part_id ,003 antipath <,001 Treatment: humanization <,001 Treatment: Information <,001 Treatment: Humanization + Information <,001 humxant <,001 comxant <,001 a. Dependent Variable: e_conc Excluded Variablesa Model Beta In t Sig. Partial Correlation 1 Treatment: humanization ,337b21,573 <,001 ,354 Treatment: Information -,228b-14,061 <,001 -,240 Treatment: Humanization + Information ,305b19,247 <,001 ,320 humxant ,290b17,658 <,001 ,296 comxant ,256b15,519 <,001 ,263 2 humxant -,149c-4,589 <,001 -,080 comxant -,151c-4,681 <,001 -,082 3 comxant -,223d-6,552 <,001 -,114 Excluded Variablesa Model Collinearity Statistics Tolerance 33 1 Treatment: humanization ,999 Treatment: Information ,999 Treatment: Humanization + Information ,995 humxant ,945 comxant ,952 2 humxant ,170 comxant ,172 3 comxant ,154 a. Dependent Variable: e_conc b. Predictors in the Model: (Constant), antipath, Male=1, Other=0, age, part_id c. Predictors in the Model: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information d. Predictors in the Model: (Constant), antipath, Male=1, Other=0, age, part_id, Treatment: Information, Treatment: humanization, Treatment: Humanization + Information, humxant Matrix Notes Output Created 14-JUL-2023 13:54:36 Comments Input Active Dataset DataSet1 Filter keep=4 (FILTER) Weight <none> Split File <none> N of Rows in Working Data File 3514 Resources Processor Time 00:00:19,17 Elapsed Time 00:00:19,20 40 OUTCOME VARIABLE: harm Model Summary R R-sq MSE F df1 df2 p ,7377 ,5442 ,0220 484,4796 8,0000 3246,0000 ,0000 Model coeff se t p LLCI ULCI constant ,3175 ,0160 19,8930 ,0000 ,2862 ,3488 t_comb ,0436 ,0085 5,0958 ,0000 ,0268 ,0603 e_conc -,1483 ,0124 -12,0005 ,0000 -,1726 -,1241 Male -,0044 ,0055 -,8140 ,4157 -,0151 ,0063 age ,0005 ,0002 2,5101 ,0121 ,0001 ,0009 part_id ,1247 ,0124 10,0677 ,0000 ,1004 ,1489 antipath ,5788 ,0120 48,4090 ,0000 ,5554 ,6023 t_hum ,0424 ,0087 4,8946 ,0000 ,0254 ,0594 t_inf ,0225 ,0074 3,0630 ,0022 ,0081 ,0370 Standardized coefficients coeff t_comb ,1984 e_conc -,1855 Male -,0097 age ,0302 part_id ,1225 antipath ,6314 t_hum ,0828 t_inf ,0454 ************************** TOTAL EFFECT MODEL **************************** OUTCOME VARIABLE: harm Model Summary R R-sq MSE F df1 df2 p 41 ,7239 ,5240 ,0230 510,6275 7,0000 3247,0000 ,0000 Model coeff se t p LLCI ULCI constant ,2673 ,0157 16,9865 ,0000 ,2365 ,2982 t_comb -,0071 ,0076 -,9317 ,3516 -,0220 ,0078 Male ,0003 ,0056 ,0540 ,9569 -,0106 ,0112 age ,0003 ,0002 1,2504 ,2112 -,0001 ,0007 part_id ,1172 ,0126 9,2709 ,0000 ,0924 ,1419 antipath ,6293 ,0114 55,0226 ,0000 ,6069 ,6517 t_hum -,0107 ,0076 -1,4043 ,1603 -,0256 ,0042 t_inf ,0091 ,0074 1,2193 ,2228 -,0055 ,0236 Standardized coefficients coeff t_comb -,0322 Male ,0007 age ,0153 part_id ,1151 antipath ,6865 t_hum -,0209 t_inf ,0183 ************** TOTAL, DIRECT, AND INDIRECT EFFECTS OF X ON Y ************** Total effect of X on Y Effect se t p LLCI ULCI c_ps -,0071 ,0076 -,9317 ,3516 -,0220 ,0078 -,0322 Direct effect of X on Y Effect se t p LLCI ULCI c'_ps ,0436 ,0085 5,0958 ,0000 ,0268 ,0603 ,1984 Indirect effect(s) of X on Y: Effect BootSE BootLLCI BootULCI 42 e_conc -,0506 ,0048 -,0602 -,0415 Partially standardized indirect effect(s) of X on Y: Effect BootSE BootLLCI BootULCI e_conc -,2307 ,0218 -,2741 -,1888 *********************** ANALYSIS NOTES AND ERRORS ************************ Level of confidence for all confidence intervals in output: 95,0000 Number of bootstrap samples for percentile bootstrap confidence intervals: 5000 NOTE: Standardized coefficients for dichotomous or multicategorical X are in partially standardized form. ------ END MATRIX ----- 1 Appendix 3: Computational replication in Matlab The next table is not presented in results by original author however it is important Table G.8_2: Regression of Humanization on Treatments × Antipathy (Continuous), Controls, Study 1 M1M2M3 Intercept 0.512*** ( 0.01 ) 0.708*** ( 0.02 ) 0.762*** ( 0.03 ) Humanization 0.135*** ( 0.01 ) 0.044 ( 0.03 ) 0.046 ( 0.03 ) Information -0.023† ( 0.01 ) -0.071* ( 0.03 ) -0.07* ( 0.03 ) Combined 0.134*** ( 0.01 ) 0.075* ( 0.03 ) 0.063* ( 0.03 ) Out g roup Antipath y -0.378*** ( 0.04 ) -0.378*** ( 0.04 ) Humanization × Antipath y 0.173*** ( 0.05 ) 0.166** ( 0.05 ) Information × Antipath y 0.098* ( 0.05 ) 0.092† ( 0.05 ) Combined × Antipath y 0.106* ( 0.05 ) 0.122* ( 0.05 ) Gender ( 1 = Male ) -0.035*** ( 0.01 ) Age -0.001***( 0) Part y ID ( 0 – 1 ) 0.046* ( 0.02 ) N 3278 3273 3113 R-square 0.08 0.15 0.16 ad j . R-square 0.08 0.14 0.15 Resid. sd 0.25 0.25 0.24 % script to reproduce results of Gubler, J. R., Karpowitz, C. F., Monson, J. Q., Romney, D. A., & South, M. (2022). % Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail. The Journal of Politics, 84(4), 2156-2171. % Written by: Shubham Pandey % Indian Institute of Technology Bombay, Mumbai % contact: [email protected] %-------------------------------------------------------------------------- % ------to run the script, keep the data in same directory as script------- clear; close; rawData=readtable('stud01_deID.csv'); stud01=rawData; %drop unnecessary columns stud01.welcome1=[]; stud01.welcome2=[]; %total participants N.S1_Total=height(stud01); %drop participants who could not finish toDrop = stud01.finished == 0; 2 N.S1_notFinish= sum(toDrop == 1); stud01(toDrop,:) = []; %drop participants with video issues toDrop = stud01.vidscreen == "No"; N.S1_VidIssues= sum(toDrop == 1); stud01(toDrop,:) = []; %keep only white/Caucasian participants toDrop = stud01.ethnicity ~= "White / Caucasian"; N.S1_NonWhites= sum(toDrop == 1); stud01(toDrop,:) = []; %drop duplicate rows (entries) in the data stud01=sortrows(stud01,1); N.S1_BeforeUnique=height(stud01); %-unique(A,setOrder) returns the unique values of A in a specific order. setOrder can be 'sorted' (default) or 'stable'. [~, uniqueIdx, ~] = unique(stud01(:,"identifier"), 'rows', 'stable'); stud01 = stud01(uniqueIdx, :); % Find the number of rows deleted N.S1_FinalSample= height(stud01); N.S1_Duplicates = N.S1_BeforeUnique -N.S1_FinalSample; %make gender numeric, 1 for male, 0 for female stud01.gender=replace(stud01.gender, {'Male', 'Female'}, {'1', '0'}); stud01.gender=str2double(stud01.gender); %change year born column to age by substracting from 2012 stud01.year_born=2012-stud01.year_born; stud01 = renamevars(stud01, 'year_born', 'Age'); %change partyId stud01.partyid = replace(stud01.partyid, {'Strong Republican', 'Not so strong Republican', 'Independent leaning Republican',... 'Independent leaning Democrat','Independent', 'Not so strong Democrat', 'Strong Democrat', 'Other'}, {'7', '6', '5', '3', '4','2', '1', 'NaN'}); stud01.partyid = replace(stud01.partyid, "Don't know", "NaN"); stud01.partyid=str2double(stud01.partyid); %now lets change likert scale repsonse to digit for antipathy 3 likertCols = {'m_learn', 'm_less', 'm_suffer'}; % Specify the column names of the Likert scale responses likertScale = {'NA', 'Strongly Disagree', 'Disagree', 'Somewhat Disagree', 'Neither Agree nor Disagree', 'Somewhat Agree', 'Agree', 'Strongly Agree'}; replaceValues = [NaN 1 2 3 4 5 6 7]; for i = 1:numel(likertCols) colName = likertCols{i}; colData = stud01.(colName); % Access the column data [~, colData] = ismember(colData, likertScale); % Convert the responses to indices colData = replaceValues(colData); % Replace with values from 1 to 7 stud01.(colName) = colData'; % Assign the updated column data back to the table end %mlearn is reverse coded stud01.m_learn=max(stud01.m_learn)+ 1 - stud01.m_learn; %code to find policy harm index study 1 stud01.immig_opinion = str2double(replace(stud01.immig_opinion, {'NA', 'Illegal immigrants should be required to go home immediately.',... 'Most illegal immigrants should be required to go home, but some should be allowed to remain in the U.S. under a temporary guest worker program.',... 'Most illegal immigrants should be allowed to stay in the U.S. but only as temporary workers who must eventually return home.',... 'Illegal immigrants should be allowed to stay permanently in the U.S.'}, {'NaN', '1', '2', '3', '4'})); %reverse code the above response stud01.immig_opinion=5-stud01.immig_opinion; %below two policy questions were only asked in Study 1, 'arizona_law', 'st8_hb497' rated on 1 to 5 stud01.arizona_law = str2double(replace(stud01.arizona_law, {'NA', 'Strongly Oppose', 'Oppose', 'Neither Favor nor Oppose', 'Favor', 'Strongly Favor'}, {'NaN', '1', '2', '3', '4', '5'})); stud01.st8_hb497 = str2double(replace(stud01.st8_hb497, {'NA', 'Strongly Oppose', 'Oppose', 'Neither Favor nor Oppose', 'Favor', 'Strongly Favor'}, {'NaN', '1', '2', '3', '4', '5'})); %now similarly change other likert columns related to policy harm likertCols = {'law_english', 'law_tuition', 'law_welfare', 'law_hire'}; % Specify the column names of the Likert scale responses likertScale = {'NA', 'Strongly Disagree', 'Disagree', 'Somewhat Disagree', 'Neither Agree or Disagree', 'Somewhat Agree', 'Agree', 'Strongly Agree'}; replaceValues = [NaN 1 2 3 4 5 6 7]; for i = 1:numel(likertCols) colName = likertCols{i}; 4 colData = stud01.(colName); % Access the column data [~, colData] = ismember(colData, likertScale); % Convert the responses to indices colData = replaceValues(colData)'; % Replace with values from 1 to 7 stud01.(colName) = colData; % Assign the updated column data back to the table end %standardize all variables between zero to one var = {'partyid', 'm_learn', 'm_less', 'm_suffer', 'i_admire', 'i_love',... 'e_sym', 'e_moved', 'e_com', 'e_warm', 'e_soft', 'e_tender',... 'd_uncom', 'd_uneasy', 'd_bother', 'd_tense', 'd_concern',... 'law_english', 'law_tuition', 'law_welfare', 'law_hire', 'arizona_law', 'st8_hb497', 'immig_opinion'}; % Apply standardizing function stud01{:, var} = zero_to_one(stud01{:, var}); clear var; %-------now caluclate scores -------- stud01.antipathy_score=(stud01.m_learn+ stud01.m_less+ stud01.m_suffer)/3; % make two groups of high and low antipaty stud01.antipathy_group= stud01.antipathy_score > 0.5; stud01.humanization_score=(stud01.i_admire+stud01.i_love)/2; stud01.empathy_score = (stud01.e_sym + stud01.e_moved + stud01.e_com + stud01.e_warm + stud01.e_soft + stud01.e_tender)/6; %calculated dissonance stud01.diss = mean(stud01{:, {'d_uncom', 'd_uneasy', 'd_bother', 'd_tense', 'd_concern'}}, 2, 'omitnan'); %calculate policy harm score stud01.harm = mean(stud01{:, {'law_english', 'law_tuition', 'law_welfare', 'law_hire', 'immig_opinion', 'arizona_law', 'st8_hb497'}}, 2, 'omitnan'); % Calculate row means and assign them to a new column 'harm' %--------------------------------- %------now applying regression models-------- % Main study 1 regression model for humanization ~ treatments * antipathy % Extract the variables from the table/dataset emp = stud01.empathy_score; %convert numeric to string stud01.treatment1 = str2double(stud01.treatment1); stud01.treatment2 = str2double(stud01.treatment2); stud01.treatment3 = str2double(stud01.treatment3); 5 stud01.treatment4 = str2double(stud01.treatment4); % Replace NaN values with zero stud01.treatment1(isnan(stud01.treatment1)) = 0; stud01.treatment2(isnan(stud01.treatment2)) = 0; stud01.treatment3(isnan(stud01.treatment3)) = 0; stud01.treatment4(isnan(stud01.treatment4)) = 0; icb_pre = stud01.antipathy_score; icb_pre_d = stud01.antipathy_group; possec=stud01.humanization_score; duplicateData=stud01; %Table G.8: Regression of Humanization on Treatments × Antipathy (Dichotomous), Controls, Study 1 % Get the variable names of the desired columns table2fit = stud01(:, {'treatment1', 'treatment2', 'treatment3', 'gender', 'Age', 'partyid','antipathy_group', 'humanization_score'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Humanization', 'Information', 'Combined', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy','humanization_score' }; % regression model G8(1) in appendix r.s1.hum.first = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined'); %regression model G8(2) in appendix r.s1.hum.second = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy'); %regression model G8(3) in appendix r.s1.hum.third = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a combined table with three models and save in excel for viewing purpose Table.G8 = create_table (r.s1.hum,'G8'); % The next table is not presented in results by original author however it is important % Table G.8_2: Regression of Humanization on Treatments × Antipathy (Continuosu), Controls, Study 1 table2fit = stud01(:, {'treatment1', 'treatment2', 'treatment3', 'gender', 'Age', 'partyid','antipathy_score', 'humanization_score'}); 6 %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Humanization', 'Information', 'Combined', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy','humanization_score' }; r.s1.hum.first = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined'); r.s1.hum.second = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy'); r.s1.hum.third = fitlm(table2fit,'humanization_score ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a combined table with three models and save in excel for viewing purpose Table.G8_2 = create_table (r.s1.hum,'G8_2'); %Table G.9: Regression of Empathic Concern on Treatments × Antipathy (Continuous), Controls, Study 1 table2fit = stud01(:, {'treatment1', 'treatment2', 'treatment3', 'gender', 'Age', 'partyid','antipathy_score', 'empathy_score'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Humanization', 'Information', 'Combined', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy','Empathetic_concern' }; % regression model G9(1) in appendix r.s1.emp.first = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined'); %regression model G9(2) in appendix r.s1.emp.second = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy'); %regression model G9(3) in appendix r.s1.emp.third = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a combined table with three models and save in excel for viewing purpose Table.G9 = create_table (r.s1.emp,'G9'); %Table G.10: Regression of Empathic Concern on Treatments × Antipathy (Dichotomous), Controls, Study 1 7 table2fit = stud01(:, {'treatment1', 'treatment2', 'treatment3', 'gender', 'Age', 'partyid','antipathy_group', 'empathy_score'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Humanization', 'Information', 'Combined', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy','Empathetic_concern' }; % regression model G10(1) in appendix r.s1.emp.first = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined'); %regression model G10(2) in appendix r.s1.emp.second = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined + Outgroup_Antipathy + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy'); %regression model G10(3) in appendix r.s1.emp.third = fitlm(table2fit,'Empathetic_concern ~ Humanization + Information + Combined + Outgroup_Antipathy + Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a combined table with three models and save in excel for viewing purpose Table.G10 = create_table (r.s1.emp,'G10'); %Table G.13: Regression of Policy Harm on Treatments × Antipathy (Continuous), Controls, Study table2fit = stud01(:, {'treatment1', 'treatment2', 'treatment3', 'gender', 'Age', 'partyid','antipathy_score', 'harm'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Humanization', 'Information', 'Combined', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy','Policy_Harm' }; % regression model G13(1) in appendix r.s1.harm.first = fitlm(table2fit,'Policy_Harm ~ Humanization + Information + Combined'); %regression model G13(2) in appendix r.s1.harm.second = fitlm(table2fit,'Policy_Harm ~ Humanization + Information + Combined + Outgroup_Antipathy+ Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy'); %regression model G13(3) in appendix r.s1.harm.third = fitlm(table2fit,'Policy_Harm ~ Humanization + Information + Combined + Outgroup_Antipathy+ Humanization*Outgroup_Antipathy + Information*Outgroup_Antipathy + Combined*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a combined table with three models and save in excel for viewing purpose 14 stud02(isnan(stud02.condition), :) = []; %Humanization Measures and Index myvars = {'post_hum1', 'post_hum2'}; stud02.Properties.VariableNames(startsWith(stud02.Properties.VariableNames, 'Q25_')) = ... strcat('post_hum', string(1:8)); stud02.Properties.VariableNames(startsWith(stud02.Properties.VariableNames, 'hum')) = ... strcat('pre_', stud02.Properties.VariableNames(startsWith(stud02.Properties.VariableNames, 'hum'))); % Calculate composite scores stud02.post_hum_measure = mean(stud02{:, myvars}, 2); stud02.post_hum_measure = (stud02.post_hum_measure - 1) / (7 - 1); % Alpha raw_alpha.hum = cronbach(stud02{:, myvars}); %dissonance: 10 items % Dissonance Measures and Index stud02.Properties.VariableNames(ismember(stud02.Properties.VariableNames, {'Q35_1', 'Q35_4', 'Q35_5', 'Q35_7', 'Q35_8', 'Q36_3', 'Q36_4', 'Q36_5', 'Q36_8', 'Q36_9'})) = ... strcat('diss', string(1:10)); myvars = {'diss1', 'diss2', 'diss5', 'diss6', 'diss7'}; stud02.diss_measure = mean(stud02{:, myvars}, 2); stud02.diss_measure = (stud02.diss_measure - 1) / (7 - 1); % Alpha raw_alpha.diss = cronbach(stud02{:, myvars}); % Calculate median of diss_measure with na.rm = TRUE my_med = median(stud02.diss_measure, 'omitnan'); % Calculate diss_hi based on diss_measure and my_med stud02.diss_hi = double(~(stud02.diss_measure <= my_med)); stud02.diss_hi(isnan(stud02.diss_measure)) = NaN; % in next few lines, a dichotomous dissonance measuer is calcuted based on median of dissonance of study 1. % I do not know the reason why this was done by original authors. The final generated varible is "diss_hi_alt". % however this variable was not used in any regression model % % Calculate median of stud01$diss where treatment is 4, with na.rm = TRUE % my_med = median(stud01.diss(stud01.treatment == 4), 'omitnan'); % % Calculate diss_hi_alt based on diss_measure and my_med from Study 1 data % stud02.diss_hi_alt = double(~(stud02.diss_measure <= my_med)); % stud02.diss_hi_alt(isnan(stud02.diss_measure)) = NaN; 15 %Empathy Measures and Index stud02.Properties.VariableNames(startsWith(stud02.Properties.VariableNames, 'Q38_')) = ... strcat('emp', string(1:6)); myvars = {'emp1', 'emp2', 'emp3', 'emp4', 'emp5', 'emp6'}; stud02.emp_index = mean(stud02{:, myvars}, 2); stud02.emp_index01 = (stud02.emp_index - 1) / (7 - 1); raw_alpha.emp = cronbach(stud02{:, myvars}); %Policy Measures and Index oldvars = {'Q40_1', 'Q40_2', 'Q41_14', 'Q41_21', 'Q41_22', 'Q41_16', 'Q41_20', ... 'Q42', 'Q43_6', 'Q43_7', 'Q44_1', 'Q44_2', 'Q44_3', 'Q44_4'}; newvars = {'pol1a', 'pol1b', 'pol2a', 'pol2b', 'pol2c', 'pol2d', 'pol2e', ... 'pol3', 'pol4a', 'pol4b', 'pol5a', 'pol5b', 'pol5c', 'pol5d'}; stud02.Properties.VariableNames(ismember(stud02.Properties.VariableNames, oldvars)) = newvars; stud02.pol1b_rev = abs(stud02.pol1b - 6); stud02.pol3_rev = abs(stud02.pol3 - 5); myvars = {'pol1b_rev', 'pol3_rev', 'pol4a', 'pol4b', 'pol5a', 'pol5b', 'pol5c', 'pol5d'}; % stud02.pol3_gohome = zeros(size(stud02, 1), 1); % stud02.pol3_gohome(stud02.pol3 == 1 | stud02.pol3 == 2) = 1; % stud02.pol3_gohome(isnan(stud02.pol3)) = NaN; stud02.policy_harm = mean(stud02{:, myvars}, 2); stud02.policy_harm = (stud02.policy_harm - 1) / (6.4 - 1); raw_alpha.emp = cronbach(stud02{:, myvars}); %Antipathy Measures and Index %Update: the author of original paper have infomred that they took icb7 as an extra exploratory variable and it should not be included in analysis. stud02.icb8_rev = abs(8 - stud02.icb8); myvars = ["icb1", "icb2", "icb3", "icb4", "icb5", "icb6", "icb8_rev", "icb9", "icb10"]; stud02.icb_measure = mean(stud02{:, myvars}, 2); raw_alpha.antipathy = cronbach(stud02{:, myvars}); %Fix the hi_icb measure stud02.hi_icb = double(~ (stud02.icb_measure < 4)); stud02.hi_icb(isnan(stud02.icb_measure)) = NaN; %standardize some variable stud02.icb_measure = zero_to_one(stud02.icb_measure); stud02.partyid = zero_to_one(stud02.partyid); 16 %now perform regression modeling %Table G.11: Regression of Empathic Concern on Treatments × Antipathy (Dichotomous), Controls, Study 2 table2fit = stud02(:, {'condition', 'gender', 'age', 'partyid','hi_icb', 'emp_index01'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Illegal_Condition', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy', 'Empathetic_concern' }; % regression model G11(1) in appendix r.s2.emp.first = fitlm(table2fit,'Empathetic_concern ~ Illegal_Condition'); %regression model G11(2) in appendix r.s2.emp.second = fitlm(table2fit,'Empathetic_concern ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy'); %regression model G11(3) in appendix r.s2.emp.third = fitlm(table2fit,'Empathetic_concern ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a table with these three models and save in excel Table.G11 = create_table (r.s2.emp,'G11'); %Table G.12: Regression of Dissonance on Treatments × Antipathy (Dichotomous), Controls, Study 2 table2fit = stud02(:, {'condition', 'gender', 'age', 'partyid','hi_icb', 'diss_measure'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Illegal_Condition', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy', 'Dissonance' }; % regression model G12(1) in appendix r.s2.diss.first = fitlm(table2fit,'Dissonance ~ Illegal_Condition'); %regression model G12(2) in appendix r.s2.diss.second = fitlm(table2fit,'Dissonance ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy'); %regression model G12(3) in appendix r.s2.diss.third = fitlm(table2fit,'Dissonance ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a table with these three models and save in excel Table.G12 = create_table (r.s2.diss,'G12'); % Table G.14: Regression of Policy Harm on Treatments × Antipathy (Continuous), Controls, Study 2 17 table2fit = stud02(:, {'condition', 'gender', 'age', 'partyid','icb_measure', 'policy_harm'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Illegal_Condition', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy', 'Policy_Harm' }; % regression model G14(1) in appendix r.s2.harm.first = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition'); %regression model G14(2) in appendix r.s2.harm.second = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy'); %regression model G14(3) in appendix r.s2.harm.third = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a table with these three models and save in excel Table.G14 = create_table (r.s2.harm,'G14'); %Table G.16: Regression of Policy Harm on Treatments × Antipathy (Dichotomous), Controls, Study 2 table2fit = stud02(:, {'condition', 'gender', 'age', 'partyid','hi_icb', 'policy_harm'}); %now change variable names for better understanding of results table2fit.Properties.VariableNames = {'Illegal_Condition', 'Gender_1_male', 'Age', 'Party_ID', 'Outgroup_Antipathy', 'Policy_Harm' }; % regression model G16(1) in appendix r.s2.harm.first = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition'); %regression model G16(2) in appendix r.s2.harm.second = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy'); %regression model G16(3) in appendix r.s2.harm.third = fitlm(table2fit,'Policy_Harm ~ Illegal_Condition + Outgroup_Antipathy+ Illegal_Condition*Outgroup_Antipathy + Gender_1_male + Age + Party_ID'); %make a table with these three models and save in excel Table.G16 = create_table (r.s2.harm,'G16'); %--------------------------------------------------------------------------- % --------functions used in above script------------------------------ %----------function to add significance star---------- function stars_cell=findstar (array) pvalue=array.pValue; significant_stars= {'***';'**'; '*'; '†'}; 18 thresholds= [.001, .01, .05, .10]; % Initialize the cell array of stars stars_cell = cell(size(pvalue)); % Assign stars based on thresholds for i = 1:numel(pvalue) for j = 1:numel(thresholds) if pvalue(i) <= thresholds(j) stars_cell{i} = significant_stars{j}; break; end end end end %-------------------------------------------------------------------------- %------------------function to swap rows of regression table--------------- function coeficients = swapRows (array) % Matlab automatically keeps interaction effects at last of coeficieint table % however, the paper tables report them in middle so we need to swap rows here % Assuming you have the structs r.s2.emp.third and r.s1.emp.third coeficients=array.Coefficients; % find wheter this study 1 or 2 depending on no. of variables Numvariables=length(array.CoefficientNames); if Numvariables == 11 %if Study 1 % Define the rows to move rowsToMove = coeficients(5:7, :); % Shift rows 8 to 11 up coeficients(5:8, :) = coeficients(8:11, :); % Add rows 5, 6, 7 as the last three rows coeficients(9:11, :) = rowsToMove; % Update row names rowNames = coeficients.Properties.RowNames; tempRownames=rowNames(5:7); rowNames(5:8) = rowNames(8:11); rowNames(9:11) = tempRownames; coeficients.Properties.RowNames = rowNames; end if Numvariables == 7 %if study 2 % Define the rows to move rowsToMove = coeficients(3:5, :); % Shift rows 5 to 6 up 19 coeficients(3:4, :) = coeficients(6:7, :); % Add rows 2, 3, 4 as the last three rows coeficients(5:7, :) = rowsToMove; % Update row names rowNames = coeficients.Properties.RowNames; tempRownames=rowNames(3:5); rowNames(3:4) = rowNames(6:7); rowNames(5:7) = tempRownames; coeficients.Properties.RowNames = rowNames; end end %-------------------------------------------------------------------------- %---------------------function to fortmat axes of graphs------------------- function make_axis() set(gca,'box','off')%Removes right and upper axes set(gca,'FontSize',12); set(gca,'FontWeight','bold'); set(gca,'Ticklength',[0.01 0.01]); set(gca,'TickDir', 'out'); end %-------------------------------------------------------------------------- function [a,R,N]=cronbach(X) %downloaded from internet to calculate alpha value % Writen by: Frederik Nagel % Institute of Music Physiology and Musicians' Medicine % Hanover University of Music and Drama % Hannover % Germany % % e-mail: [email protected] % homepage: http://www.immm.hmt-hannover.de if nargin<1 || isempty(X) error('You shoud provide a data set.'); else % X must be at least a 2 dimensional matrix if size(X,2)<2 error('Invalid data set.'); end end % Items N=size(X,2); 20 % Entries of the upper triangular matrix e=(N*(N-1)/2); % Spearman's correlation coefficient R = corr(X,'rows','pairwise','type','spearman'); % Coefficients from upper triangular matrix R = triu(R,1); % Mean of correlation coefficients r = sum(sum(triu(R,1)))/e; % If there are columns with zero variance, these have to be excluded. if(isnan(r)) disp('There are columns with zero variance!'); disp('These columns have been excluded from the calculation of alpha!'); disp([num2str(sum(sum(isnan(R)))) ' coefficients of ' num2str(N*N) ' have been excluded.']); % Correct # of items e = e-sum(sum(isnan(R))); % corrected mean of correlation coefficients r = nansum(nansum(R))/e; % corrected number of items N = N - sum(isnan(R(1,:))); end % Formular for alpha (Cronbach 1951) a=(N*r)/(1+(N-1)*r); end %-------------------------------------------------------------------------- %---------------------function to create actual regression table----------- function Table = create_table (model,sheet) %function to create an excel sheet having three models along with their significance if model.first.NumCoefficients == 4 %then this model is from study 1 data0 = {'Intercept'; 'Humanization'; 'Information'; 'Combined'; 'Outgroup Antipathy';... 'Humanization × Antipathy'; 'Information × Antipathy'; 'Combined × Antipathy';... 'Gender (1 = Male)'; 'Age'; 'Party ID (0–1)';... 'N'; 'R-square'; 'adj. R-square'; 'Resid. sd'; 'pValue(model)'}; %these are rows to be shown in first column of results table TotalRows = 11; %total rows in shown table elseif model.first.NumCoefficients == 2 %then this model is from study 2 data0 = {'Intercept'; 'Illegal Condition'; 'Outgroup Antipathy'; 'Illegal Condition × Antipathy';... 'Gender'; 'Age'; 'Party ID (0–1)';... 21 'N'; 'R-square'; 'adj. R-square'; 'Resid. sd'; 'pValue(model)'}; %these are rows to be shown in first column of results table TotalRows = 7; %total rows in shown table exluding four summary rows else disp ('error'); end % Concatenate "estimate" and "SE" columns into a new column with brackets data1 = cellstr(strcat(num2str(round(model.first.Coefficients.Estimate,3)), findstar(model.first.Coefficients),'(', num2str(round(model.first.Coefficients.SE,2)), ')')); data2 = cellstr(strcat(num2str(round(model.second.Coefficients.Estimate,3)), findstar(model.second.Coefficients), '(', num2str(round(model.second.Coefficients.SE,2)), ')')); %Matlab automatically keeps interaction effects at last of coeficieint table so bring them up to align with table presnted in paper Coefficients=swapRows(model.third); data3 = cellstr(strcat(num2str(round(Coefficients.Estimate,3)), findstar(Coefficients), '(', num2str(round(Coefficients.SE,2)), ')')); %caluclated 4 keys results like N, R square, adjusted R square, Residual SD fourValues.s2.emp.first = {round(model.first.NumObservations,2); round(model.first.Rsquared.Ordinary,2); round(model.first.Rsquared.Adjusted,2); round(nanstd(model.first.Residuals.Raw),2)}; fourValues.s2.emp.second = {round(model.second.NumObservations,2); round(model.second.Rsquared.Ordinary,2); round(model.second.Rsquared.Adjusted,2); round(nanstd(model.second.Residuals.Raw),2)}; fourValues.s2.emp.third = {round(model.third.NumObservations,2); round(model.third.Rsquared.Ordinary,2); round(model.third.Rsquared.Adjusted,2); round(nanstd(model.third.Residuals.Raw),2)}; %find p values to add in separate column----- p1=cellfun(@(x) sprintf('%.4f', x), num2cell(model.first.Coefficients.pValue), 'UniformOutput', false); p2=cellfun(@(x) sprintf('%.4f', x), num2cell(model.second.Coefficients.pValue), 'UniformOutput', false); p3=cellfun(@(x) sprintf('%.4f', x), num2cell(Coefficients.pValue), 'UniformOutput', false); % Pad the shorter columns with empty strings p1= [p1; repmat("", TotalRows-length(data1)+5, 1)]; p2= [p2; repmat("", TotalRows-length(data2)+5, 1)]; p3= [p3; repmat("", TotalRows-length(data3)+5, 1)]; 22 % Pad the shorter columns with empty strings data1 = [data1; repmat("", TotalRows-length(data1), 1); fourValues.s2.emp.first]; data2 = [data2; repmat("", TotalRows-length(data2), 1); fourValues.s2.emp.second]; data3 = [data3; repmat("", TotalRows-length(data3), 1); fourValues.s2.emp.third]; %add regression model p value at the end of data column, i.e., last row data1= [data1; sprintf('%.4f', model.first.ModelFitVsNullModel.Pvalue)]; data2= [data2; sprintf('%.4f', model.second.ModelFitVsNullModel.Pvalue)]; data3= [data3; sprintf('%.4f', model.third.ModelFitVsNullModel.Pvalue)]; % % Create the table % Table= table(data0, data1, data2, data3, 'VariableNames', {' ','M1', 'M2', 'M3'}); Table= table(data0, data1, p1, data2, p2, data3, p3, 'VariableNames',{' ','M1', 'p1', 'M2', 'p2', 'M3', 'p3'}); % Write the table to an Excel file writetable(Table, 'regression.xls', 'Sheet', sheet); end 23 Appendix 4: Comparison of main results, original results and replication results TableG.8:RegressionofHumanizationonTreatments×Antipathy(Dichotomous),Controls,Study1 Replication Originalstudy R1:Comp.Repl.InR R2:Comp.Repl.InSPSS R3:Comp.Repl.InMatlab Equation (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intercept 0.51*** (0.01) 0.59*** (0.01) 0.67*** (0.03) 0.51*** (0.01) 0.59*** (0.01) 0.67*** (0.03) 0.52*** (0.09) 0.71*** (0.02) 0.76*** (0.02) 0.512*** (0.01) 0.587*** (0.01) 0.675*** (0.03) Humanization 0.13*** (0.01) 0.10*** (0.02) 0.10*** (0.02) 0.13*** (0.01) 0.10*** (0.02) 0.09*** (0.02) 0.21*** (0.13) 0.07(0.03) 0.08(0.03) 0.135*** (0.01) 0.096*** (0.02) 0.096*** (0.02) Information ‐0.03* (0.01) ‐0.05** (0.02) ‐0.05** (0.02) ‐0.03* (0.01) ‐0.05** (0.02) ‐0.05** (0.02) ‐0.04* (0.01) ‐0.12* (0.03) ‐0.12* (0.03) ‐0.023† (0.01) ‐0.046** (0.02) ‐0.045** (0.02) Combined 0.13*** (0.01) 0.10*** (0.02) 0.10*** (0.02) 0.13*** (0.01) 0.10*** (0.02) 0.10*** (0.02) 0.22*** (0.1) 0.10* (0.03) 0.10* (0.03) 0.134*** (0.01) 0.104*** (0.02) 0.098*** (0.02) OutgroupAntipathy  ‐0.15*** (0.02) ‐0.15*** (0.02)  ‐0.15*** (0.02) ‐0.15*** (0.02)  ‐0.35*** (0.04) ‐0.34*** (0.04)  ‐0.152*** (0.02) ‐0.146*** (0.02) HumanizationxAntipathy  0.08** (0.02) 0.07** (0.03)  0.08** (0.02) 0.08** (0.03)  0.15** (0.05) 0.15** (0.05)  0.077** (0.03) 0.074** (0.03) InformationxAntipathy  0.05* (0.02) 0.05† (0.02)  0.05* (0.02) 0.05† (0.02)  0.10* (0.05) 0.10† (0.05)  0.05* (0.02) 0.045† (0.02) CombinedxAntipathy  0.05* (0.02) 0.06* (0.03)  0.05* (0.02) 0.06* (0.03)  0.12* (0.05) 0.12* (0.05)  0.052* (0.03) 0.056* (0.03) Gender(1=Male)   ‐0.04*** (0.01)   ‐0.04*** (0.01)   ‐0.06*** (0.01)   ‐0.038*** (0.01) Age   ‐0.00*** (0.00)   ‐0.00*** (0.00)   ‐0.07*** (0.00)   ‐0.001*** (0.00) PartyID(0‐1)   0.02(0.02)   0.02(0.02)   0.04* (0.02)   0.013 (0.02) N 3309 3305 3134 3309 3305 3134 3149 3149 3149 3278 3278 3117 R2 0.08 0.12 0.13 0.08 0.12 0.13 0.08 0.15 0.16 0.08 0.12 0.13 adj.R2 0.08 0.12 0.13 0.08 0.12 0.13 0.08 0.14 0.15 0.08 0.12 0.13 Resid.SD 0.26 0.25 0.25 0.26 0.25 0.25    0.25 0.25 0.25 Standarderrorsinparentheses  †significantatp<.10;*p<.05;**p<.01;***p<.001  30 TableG.15:RegressionofPolicyHarmonTreatments×Antipathy(Dichotomous),Controls,Study1 Replication Originalstudy R1:Comp.Repl.InR R2:Comp.Repl.InSPSS R3:Comp.Repl.InMatlab Equation (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intercept 0.71***(0.01) 0.36*** (0.01) 0.26*** (0.02) 0.71*** (0.01) 0.36*** (0.01) 0.26*** (0.02)    0.706*** (0.01) 0.574*** (0.01) 0.392*** (0.02) Humanization ‐0.01(0.01) ‐0.00 (0.02) ‐0.01 (0.02) ‐0.01 (0.01) ‐0.00 (0.02) ‐0.01 (0.02)    ‐0.011 (0.01) ‐0.003 (0.01) ‐0.006 (0.01) Information 0.01(0.01) 0.04* (0.02) 0.04* (0.02) 0.01(0.01) 0.04* (0.02) 0.04* (0.02)    0.012 (0.01) 0.025* (0.01) 0.027* (0.01) Combined ‐0.01(0.01) ‐0.00 (0.02) ‐0.00 (0.02) ‐0.01 (0.01) ‐0.00 (0.02) ‐0.00 (0.02)    ‐0.014 (0.01) 0.008 (0.01) 0.004 (0.01) OutgroupAntipathy  0.67*** (0.02) 0.65*** (0.02)  0.67*** (0.02) 0.65*** (0.02)     0.267*** (0.01) 0.25*** (0.01) HumanizationxAntipathy  ‐0.01 (0.03) ‐0.01 (0.03)  ‐0.01 (0.03) ‐0.01 (0.03)     ‐0.011 (0.02) ‐0.007 (0.02) InformationxAntipathy  ‐0.06† (0.03) ‐0.06† (0.03)  ‐0.06† (0.03) ‐0.06† (0.03)     ‐0.03† (0.02) ‐0.032† (0.02) CombinedxAntipathy  ‐0.01 (0.03) ‐0.01 (0.03)  ‐0.01 (0.03) ‐0.01 (0.03)     ‐0.021 (0.02) ‐0.016 (0.02) Gender(1=Male)   ‐0.00 (0.01)   ‐0.00 (0.01) 0.003 (0.01) Age   0.00(0.00)   0.00(0.00)      0.001* (0.00) PartyID(0‐1)   0.12*** (0.01)   0.12*** (0.01) 0.192*** (0.01) N 3489 3482 3281 3489 3482 3281    3478 3478 3292 R2 0.00 0.51 0.52 0.00 0.51 0.52    0.00 0.32 0.36 adj.R2 0.00 0.51 0.52 0.00 0.51 0.52    0.00 0.32 0.36 Resid.SD 0.22 0.15 0.15 0.22 0.15 0.15    0.22 0.18 0.18 Standarderrorsinparentheses  †significantatp<.10;*p<.05;**p<.01;***p<.001  31 TableG.16:RegressionofPolicyHarmonTreatments×Antipathy(Continuous)1,Controls,Study2 Replication Originalstudy R1:Comp.Repl.InR R2:Comp.Repl.InSPSS R3:Comp.Repl.InMatlab Equation (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intercept 0.62*** (0.01) 0.25*** (0.01) 0.19*** (0.02) 0.616*** (0.01) 0.247*** (0.01) 0.189*** (0.02) IllegalCondition ‐0.03** (0.01) ‐0.04* (0.02) ‐0.04* (0.02) ‐0.03** (0.01) ‐0.047** (0.02) ‐0.046** (0.02) OutgroupAntipathy  0.85*** (0.03) 0.80*** (0.03) 0.846*** (0.03) 0.796*** (0.03) IllegalConditionxAntipathy  0.03(0.03) 0.03(0.04)        0.036(0.04)0.036(0.04) Gender   ‐0.02* (0.01) ‐0.017* (0.01) Age   0.00** (0.00) 0.001** (0.00) PartyID(0‐1)   0.09*** (0.01) 0.089*** (0.01) N 1982 1982 1966       1961 1959 1944 R2 0.00 0.53 0.54       0 0.520.54 adj.R2 0.00 0.52 0.53       0 0.520.53 Resid.SD 0.23 0.16 0.16       0.23 0.160.16 Standarderrorsinparentheses  †significantatp<.10;*p<.05;**p<.01;***p<.001 1Pleasenotethatthistableislabelled,intheoriginalmanuscript,asAntipathy(DICHOTOMOUS)  MUNI Econ Working Paper Series (since 2018) 2024-02 Prochazka, J., Pandey, S., Castek, O., Firouzjaeiangalougah, M. (2024). Replication of Changing Hearts and Minds? Why Media Messages Designed to Foster Empathy Often Fail (Gubler et al., 2022). MUNI ECON Working Paper n. 2024-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2024-02 2024-01 Marini, M. M., Ulivieri, G. 2024. Meta-analyses in Economic Psychology: A sustainable approach to cross-cultural differences. MUNI ECON Working Paper n. 2024-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2024-01 2023-09 Levi, E., Ramalingam, A. 2023. Absolute vs. relative poverty and wealth: Cooperation in the presence of between-group inequality. MUNI ECON Working Paper n. 2023-09. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-09 2023-08 Fumarco, L., Harrell, B., Button, P., Schwegman, D., Dils, E. 2023. Gender Identity, Race, and Ethnicity-based Discrimination in Access to Mental Health Care: Evidence from an Audit Correspondence Field Experiment. MUNI ECON Working Paper n. 2023-08. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-08 2023-07 Levi, E., Bayerlein, M., Grimalda, G., Reggiani, T. 2023. Narratives on migration and political polarization: How the emphasis in narratives can drive us apart. MUNI ECON Working Paper n. 2023-07. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-07 2023-06 Fumarco, L., Gaddis, S. M., Sarracino, F., Snoddy, I. 2023. sendemails: An automated email package with multiple applications. MUNI ECON Working Paper n. 2023-06. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-06 2023-05 Harrell, B., Fumarco, L., Button, P., Schwegman, D., Denwood, K. 2023. The Impact of COVID-19 on Access to Mental Healthcare Services. MUNI ECON Working Paper n. 2023-05. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-05 2023-04 Friedhoff, T., Au, C., Krahnhof, P. 2023. Analysis of the Impact of Orthogonalized Brent Oil Price Shocks on the Returns of Dependent Industries in Times of the Russian War. MUNI ECON Working Paper n. 2023-04. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-04 2023-03 Mikula, Š., Reggiani, T., Sabatini, F. 2023. The long-term impact of religion on social capital: lessons from post-war Czechoslovakia. MUNI ECON Working Paper n. 2023-03. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-03 2023-02 Clò, S., Reggiani, T., Ruberto, S. 2023. onsumption feedback and water saving: An experiment in the metropolitan area of Milan. MUNI ECON Working Paper n. 2023-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-02 2023-01 Adamus, M., Grežo, M. 2023. Attitudes towards migrants and preferences for asylum and refugee policies before and during Russian invasion of Ukraine: The case of Slovakia. MUNI ECON Working Paper n. 2023-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2023-01 2022-12 Guzi, M., Kahanec, M., Mýtna Kureková, L. 2022. The Impact of Immigration and Integration Policies On Immigrant-Native Labor Market Hierarchies. MUNI ECON Working Paper n. 2022-12. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-12 2022-11 Antinyan, A., Corazzini, L., Fišar, M., Reggiani, T. 2022. Mind the framing when studying social preferences in the domain of losses. MUNI ECON Working Paper n. 2022-11. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-11 2022-10 Corazzini, L., Marini, M. 2022. Focal points in multiple threshold public goods games: A single-project meta-analysis. MUNI ECON Working Paper n. 2022-10. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-10 2022-09 Fazio, A., Scervini, F., Reggiani, T. 2022. Social media charity campaigns and pro-social behavior. Evidence from the Ice Bucket Challenge.. MUNI ECON Working Paper n. 2022-09. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-09 2022-08 Coufalová, L., Mikula, Š. 2022. The Grass Is Not Greener on the Other Side: The Role of Attention in Voting Behaviour.. MUNI ECON Working Paper n. 2022-08. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-08 2022-07 Fazio, A., Reggiani, T. 2022. Minimum wage and tolerance for inequality.. MUNI ECON Working Paper n. 2022-07. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-07 2022-06 Mikula, Š., Reggiani, T. 2022. Residential-based discrimination in the labor market. MUNI ECON Working Paper n. 2022-06. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-06 2022-05 Mikula, Š., Molnár, P. 2022. Expected Transport Accessibility Improvement and House Prices: Evidence from the Construction of the World’s Longest Undersea Road Tunnel. MUNI ECON Working Paper n. 2022-05. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-05 2022-04 Coufalová, L., Mikula, Š., Ševčík, M. 2022. Homophily in Voting Behavior: Evidence from Preferential Voting. MUNI ECON Working Paper n. 2022-04. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-04 2022-03 Kecskésová, M., Mikula, Š. 2022. Malaria and Economic Development in the Short-term: Plasmodium falciparum vs Plasmodium vivax. MUNI ECON Working Paper n. 2022-03. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-03 2022-02 Mladenović, D., Rrustemi, V., Martin, S., Kalia, P., Chawdhary, R. 2022. Effects of Sociodemographic Variables on Electronic Word of Mouth: Evidence from Emerging Economies. MUNI ECON Working Paper n. 2022-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-02 2022-01 Mikula, Š., Montag, J. 2022. Roma and Bureaucrats: A Field Experiment in the Czech Republic. MUNI ECON Working Paper n. 2022-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2022-01 2021-14 Abraham, E. D., Corazzini, L., Fišar, M., Reggiani, T. 2021. Delegation and Overhead Aversion with Multiple Threshold Public Goods. MUNI ECON Working Paper n. 2021-14. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-14 2021-13 Corazzini, L., Cotton, C., Longo, E., Reggiani, T. 2021. The Gates Effect in Public Goods Experiments: How Donations Flow to the Recipients Favored by the Wealthy. MUNI ECON Working Paper n. 2021-13. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-13 2021-12 Staněk, R., Krčál, O., Mikula, Š. 2021. Social Capital and Mobility: An Experimental Study. MUNI ECON Working Paper n. 2021-12. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-12 2021-11 Staněk, R., Krčál, O., Čellárová, K. 2021. Pull yourself up by your bootstraps: Identifying procedural preferences against helping others in the presence. MUNI ECON Working Paper n. 2021-11. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-11 2021-10 Levi, E., Sin, I., Stillman, S. 2021. Understanding the Origins of Populist Political Parties and the Role of External Shocks. MUNI ECON Working Paper n. 2021-10. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-10 2021-09 Adamus, M., Grežo, M. 202. Individual Differences in Behavioural Responses to the Financial Threat Posed by the COVID-19 Pandemic. MUNI ECON Working Paper n. 2021-09. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-09 2021-08 Hargreaves Heap, S. P., Karadimitropoulou, A., Levi, E. 2021. Narrative based information: is it the facts or their packaging that matters?. MUNI ECON Working Paper n. 2021-08. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-08 2021-07 Hargreaves Heap, S. P., Levi, E., Ramalingam, A. 2021. Group identification and giving: in-group love, out-group hate and their crowding out. MUNI ECON Working Paper n. 2021-07. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-07 2021-06 Medda, T., Pelligra, V., Reggiani, T. 2021. Lab-Sophistication: Does Repeated Participation in Laboratory Experiments Affect Pro-Social Behaviour?. MUNI ECON Working Paper n. 2021-06. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-06 2021-05 Guzi, M., Kahanec, M., Ulceluse M., M. 2021. Europe’s migration experience and its effects on economic inequality. MUNI ECON Working Paper n. 2021-05. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-05 2021-04 Fazio, A., Reggiani, T., Sabatini, F. 2021. The political cost of lockdown´s enforcement. MUNI ECON Working Paper n. 2021-04. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-04 2021-03 Peciar, V. Empirical investigation into market power, markups and employment. MUNI ECON Working Paper n. 2021-03. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-03 2021-02 Abraham, D., Greiner, B., Stephanides, M. 2021. On the Internet you can be anyone: An experiment on strategic avatar choice in online marketplaces. MUNI ECON Working Paper n. 2021-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-02 2021-01 Krčál, O., Peer, S., Staněk, R. 2021. Can time-inconsistent preferences explain hypothetical biases?. MUNI ECON Working Paper n. 2021-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2021-01 2020-04 Pelligra, V., Reggiani, T., Zizzo, D.J. 2020. Responding to (Un)Reasonable Requests by an Authority. MUNI ECON Working Paper n. 2020-04. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2020-04 2020-03 de Pedraza, P., Guzi, M., Tijdens, K. 2020. Life Dissatisfaction and Anxiety in COVID-19 pandemic. MUNI ECON Working Paper n. 2020-03. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2020-03 2020-02 de Pedraza, P., Guzi, M., Tijdens, K. 2020. Life Satisfaction of Employees, Labour Market Tightness and Matching Efficiency. MUNI ECON Working Paper n. 2020-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2020-02 2020-01 Fišar, M., Reggiani, T., Sabatini, F., Špalek, J. 2020. a. MUNI ECON Working Paper n. 2020-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2020-01 2019-08 Fišar, M., Krčál, O., Špalek, J., Staněk, R., Tremewan, J. 2019. A Competitive Audit Selection Mechanism with Incomplete Information. MUNI ECON Working Paper n. 2019-08. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-08 2019-07 Guzi, M., Huber, P., Mikula, M. 2019. Old sins cast long shadows: The Long-term impact of the resettlement of the Sudetenland on residential migration. MUNI ECON Working Paper n. 2019-07. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-07 2019-06 Mikula, M., Montag, J. 2019. Does homeownership hinder labor market activity? Evidence from housing privatization and restitution in Brno. MUNI ECON Working Paper n. 2019-06. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-06 2019-05 Krčál, O., Staněk, R., Slanicay, M. 2019. Made for the job or by the job? A lab-in-the-field experiment with firefighters. MUNI ECON Working Paper n. 2019-05. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-05 2019-04 Bruni, L., Pelligra, V., Reggiani, T., Rizzolli, M. 2019. The Pied Piper: Prizes, Incentives, and Motivation Crowding-in. MUNI ECON Working Paper n. 2019-04. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-04 2019-03 Krčál, O., Staněk, R., Karlínová, B., Peer, S. 2019. Real consequences matters: why hypothetical biases in the valuation of time persist even in controlled lab experiments. MUNI ECON Working Paper n. 2019-03. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-03 2019-02 Corazzini, L., Cotton, C., Reggiani, T., 2019. Delegation And Coordination With Multiple Threshold Public Goods: Experimental Evidence. MUNI ECON Working Paper n. 2019-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-02 2019-01 Fišar, M., Krčál, O., Staněk, R., Špalek, J. 2019. The Effects of Staff-rotation in Public Administration on the Decision to Bribe or be Bribed. MUNI ECON Working Paper n. 2019-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2019-01 2018-02 Guzi, M., Kahanec, M. 2018. Income Inequality and the Size of Government: A Causal Analysis. MUNI ECON Working Paper n. 2018-02. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2018-02 2018-01 Geraci, A., Nardotto, M., Reggiani, T., Sabatini, F. 2018. Broadband Internet and Social Capital. MUNI ECON Working Paper n. 2018-01. Brno: Masaryk University. https://doi.org/10.5817/WP_MUNI_ECON_2018-01 ISSN electronic edition 2571-130X MUNI ECON Working Paper Series is indexed in RePEc: https://ideas.repec.org/s/mub/wpaper.html