The Curious Case of Theresa May and the Public That Did Not Rally: Gendered Reactions to Terrorist Attacks Can Cause Slumps Not Bumps – CORRIGENDUM

This corrigendum corrects errors in the published version of the article “The Curious Case of Theresa May and the Public That Did Not Rally: Gendered Reactions to Terrorist Attacks Can Cause Slumps Not Bumps” (Holman, Merolla, and Zechmeister 2022). The authors acknowledge errors in the code, a set of errors in one of the datasets, and missing information from table and figure notes. The authors apologize for these errors and ask readers to use the corrected output and data. The authors thank Michael Jetter, Kieran Stockley, the Institute for Replication, and the team at the APSR for their efforts to identify errors that this corrigendum corrects.

Correction 1: The difference-in-difference analysis presented in Table 2 included waves that should not have been incorporated into the data. We have corrected this analysis in corrected Table 2. None of the central findings associated with the table and difference-in-difference results change. We further updated the table note to state that both models were run using OLS, rather than logistic regression as originally reported for “May best PM.” We have also corrected the supplementary Table A5 and D4 that accompany this table.

Table 2. (Corrected): Difference-in-Difference, with Fixed Effects

Note: Dependent variables are 11-point favorability scale and perceptions of May as the best PM (OLS regression). Favorability model includes waves 9 to 12, while May best PM models include waves 10 to 12. Results consistent with logistic regression instead of OLS for May best PM model. Standard errors in parentheses. Full controls include whether someone identifies as ethnically British, gender, Labour party membership, other party membership, income, and ideology. Constant is represented by wave 10 coefficent for May best PM model. See Appendix A (Table A5) for full models. *p < 0.05.

Correction 2: We identified errors in the completeness and accuracy of the data used to generate the results presented in Table 4. We have produced corrected results for Table 4 below, and in the supplementary materials in Figure H1 and Tables H1, H2, and H3.

Table 4. (Corrected): Effect of Terrorist Attacks on Executive Approval Ratings

Note: Linear regression using time-series cross-sectional data of country-quarters. Controls for the presence of a woman executive, real GDP growth (and lagged), inflation (logged and lagged), the left-center-right placement of the leader (vs. undefined ideology as the baseline), and election quarters (lagged), with country fixed effects. Panel corrected standard errors in parentheses. Dataset includes all countries in the executive approval database that also appear in the Global Terrorism Database (N countries = 44). Full results in Appendix H, Table H1. *p < 0.05.

The updated Table 4 leads us to update the text on page 260 to indicate that the threshold is 14 deaths (not 16 as previously noted). Specifically, the pooled dataset with relevant measures covers 1975 to 2017. The number of countries and country-quarters in the analysis is 44 and 4,612, respectively, and 22 of the countries have had a woman leader.

We define this as any attack that involved an international component (using the definition from the Global Terrorism Database) and had 14 or more deaths. The dataset contains 68 country-quarters with such an attack. As expected, there are a relatively low number of women national executives: of these quarters, only 11 total are under women leaders. The results in Table 4 show that men receive a statistically significant bump in the aftermath of an international terrorist attack, whereas the interaction with women executives is negative and significant. Post-estimation analyses of the marginal effects of an international terrorist attack yield a predicted positive, significant effect for men, and a predicted negative, insignificant effect for women.

We add to Footnote 19: “If we drop the case of the Manchester attack (by omitting May’s approval ratings for the 1 period after the attack), the interaction between a woman head of state and a large terrorist attack remains significant and negative. If we omit May entirely from the dataset, the interaction retains the same sign (negative) but is no longer significant. In each case, post-estimation analyses of the marginal effects of an international terrorist attack yield a positive, significant effect for men, and a negative, insignificant effect for women. Full results are in Appendix H, Table H4.”

Correction 3: We have also updated the table and figure notes for all figures and tables in the manuscript to indicate which wave we use of the British Election Study.

Table 1: Table note should also say “Wave 12, British Elections Study.”

Table 2: Table note should say “Favorability model (Waves 9-12, British Elections Study), Best Prime Minister (Waves 10-12, British Elections Study).”

Table 3: Table note should also say “Wave 12, British Elections Study.”

Figure 1: Figure note should say “Waves 12 (dependent variable) and Wave 10 (gender attitudes), British Elections Study; see Appendix C, Table C1.”

Figure 2: Figure note should also say “Wave 12, British Elections Study.”

Figure 3: Figure note should also say “Wave 12, British Elections Study.”

Correction 4: The models to produce C1 were inappropriately specified. We have updated them and the text on page 255 to say: “As we show in Appendix Figure C1, the pattern of a substantively larger reaction among those with negative views toward women holds across all four of questions.” Figure C1 is also updated.

Correction 5: The models for column 3 of Tables E4 and E5 were inappropriately specified. We update footnote 9 to say: “We evaluate the results with a variety of additional time-related specifications, including errors clustered on the day of the survey, errors clustered and controls for time, an interaction between the Manchester attack and time, a multilevel model with day of the survey serving as the second level of the model, and AFRIMA models (see Appendix E, Table E4 and E5, and Figure E2). Results are consistent in four of the eight models. At no point does the terrorist attack result in a significant positive effect on May’s evaluation or ratings.” Table E4 and E5 are also updated.

All data and materials to verify the reproducibility of the original and amended versions of the code have been posted to the American Political Science Review Dataverse.

Updated Supplementary Materials: Please see updated supplementary materials available on the American Political Science Review Dataverse, https://doi.org/10.7910/DVN/VHNPUO

Specifically, we have updated the following tables and figures in the supplementary materials:

Table A5: Appendix to Table 2: Difference-in-Difference, with Fixed Effects and Average Treatment Effects

Figure C1: Effect of All Gender Views on Liking May

Table D4: Differences-in-Differences Evaluation of All Leaders, Fixed Effects

Table E4: Effect of Manchester Attack on May Likability with Time-Related Controls

Table E5: Effect of Manchester Attack on May Best PM with Time-Related Controls

Figure H1: Effect of Terrorist Events on Executive Approval

Table H1: Gender-Revised Rally Effects

Table H2: Changing Thresholds and the Gender-revised Rally Effects

Table H3: Interactive Effects with Executive Ideological Placement

We have updated the table notes for Tables and Figures to reflect which waves of the British Elections Survey that we draw data from for the results and the modeling strategies for each table and figure.

We have added the following tables to the supplementary materials:

Tables to accompany gender attitude figures.

Table to accompany Figure H1.

Table H4: Global Analysis, without May.