Candidate Faces and Election Outcomes: Is the Face-Vote Correlation Caused by Candidate Selection? Corrigendum

Quarterly Journal of Political Science, 2010, 5: 99 105 Corrigendum Candidate Faces and Election Outcomes: Is the Face-Vote Correlation Caused by Candidate Selection? Corrigendum Matthew D. Atkinson, Ryan D. Enos and Seth J. Hill Department of Political Science, University of California, Los Angeles, USA matthewa@ucla.edu renos@ucla.edu sjhill@ucla.edu This note corrects the analysis presented in Candidate Faces and Election Outcomes: Is the Face-Vote Correlation Caused by Candidate Selection? (Atkinson et al., 2009). A coding error assigned incumbent Slade Gorton s (R-WA) facial competence score to his opponent, Maria Cantwell, and vice versa. The error does not substantively affect the results, and all claims made in the article but for those about these two candidates stand. We claimed that Cantwell s face had a negative effect on her margin of victory. In fact, her face score is above median and, by the estimates of our model, led to her victory over Gorton. Thus the statement in our abstract (Atkinson et al., 2009, p. 229) we find that the challenging candidate s face is never the difference between a challenger and incumbent victory in all 99 elections in our study should instead state is the difference in one of 99 elections in our study. The discussion of Cantwell and Gorton in the third full paragraph on page 240 is also inaccurate. We produce below updated article tables and figures reflecting the corrected coding error. These tables and figures are produced using the updated replication archive housed with the Quarterly Journal of Political Science, or available from the authors. Supplementary Material available from: http://dx.doi.org/10.1561/100.000080621_supp ISSN 1554-0626; DOI 10.1561/100.000080621 2010 M. D. Atkinson, R. D. Enos and S. J. Hill

100 Atkinson, Enos and Hill Table 1. Predicting candidate facial competence with district competitiveness. challengers challengers incumbents incumbents Intercept 0.427 0.480 0.611 0.395 (0.306) (0.399) (0.251) (0.289) Cook incumbent risk 0.304 0.287 0.103 0.011 (0.110) (0.098) (0.090) (0.071) 1994 Fixed effect 0.337 0.035 (0.450) (0.327) 1996 Fixed effect 0.408 0.227 (0.469) (0.341) 1998 Fixed effect 0.033 0.442 (0.448) (0.325) 2000 Fixed effect 0.393 0.323 (0.416) (0.302) 2002 Fixed effect 0.001 0.103 (0.416) (0.302) Fixed effect 0.215 0.248 (0.425) (0.309) 2006 Fixed effect 0.203 0.203 (0.423) (0.307) N 148 145 167 145 R 2 0.049 0.078 0.008 0.042 Adjusted R 2 0.043 0.024 0.002 0.014 Std. error of regression 0.989 1.169 0.852 0.849 Ordinary least squares regression coefficients with standard errors in parentheses. models are for candidates from, models for candidates from 1992 to 2006. Dependent variable is facial competence, Cook incumbent risk is coded increasing from low to high risk.

Is the Face-Vote Correlation Caused by Candidate Selection? Corrigendum 101 Table 2. The Effect of Candidate Facial Competence and Partisanship on Incumbent Vote Choice. Intercept 0.336 0.293 1.236 0.785 (0.671) (0.694) (1.217) (1.559) Cook incumbent risk 0.139 0.116 0.106 0.111 (0.061) (0.065) (0.025) (0.025) Respondent shares challenger party 1.311 1.313 1.039 1.041 (0.081) (0.081) (0.044) (0.042) Respondent shares incumbent party 1.392 1.393 1.079 1.079 (0.090) (0.090) (0.036) (0.035) Challenger facial competence 0.119 0.080 0.077 0.076 (0.095) (0.098) (0.025) (0.027) Incumbent facial competence 0.024 0.018 0.014 0.011 (0.128) (0.128) (0.031) (0.036) Incumbent tenure 0.009 0.005 0.014 0.013 (0.016) (0.018) (0.009) (0.012) Tenure squared 0.000 0.000 0.000 0.000 (0.001) (0.001) (0.000) (0.000) Incumbent age 0.020 0.021 0.055 0.039 (0.030) (0.031) (0.041) (0.052) Age squared 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Challenger expenditures (logged) 0.022 0.018 (0.019) (0.018) State population (millions) 0.007 (0.008) 1994 Fixed effect 0.033 (0.093) 1996 Fixed effect 0.057 (0.077) 1998 Fixed effect 0.069 (0.078) 2000 Fixed effect 0.047 (0.079) 2002 Fixed effect 0.124 (0.114) 2006 Fixed effect 0.008 (0.134) N 4250 4250 26454 26454 AIC 3372.005 3372.261 25310.858 25294.115 Probit regression coefficients with standard errors in parentheses. Dependent variable is respondent vote for incumbent candidate. Cook Incumbent Risk is coded from 0 for contests classified as safe to 3 for contests classified as tossup. Robust standard errors clustered on state/district.

102 Atkinson, Enos and Hill 10 Effects Effects 5 5 0 10 Independent Challenger Incumbent Independent Percentage Point Effect on Challenger Vote Challenger Incumbent Figure 1. Estimated effect on challenger vote probability of increasing challenger facial competence, by respondent partisan affiliation with 95 percent confidence intervals. Each point represents the estimated difference in challenger vote probability moving the challenger s face from the 25th percentile to the 75th percentile, holding incumbent age and tenure at their means, the Cook report at likely going to the incumbent, and the incumbent s face at the chamber median. The points represent the average estimate across 500 samples from the clustered coefficient distribution, and the lines the 2.5th to the 97.5th percentile of the sampled effects.

Is the Face-Vote Correlation Caused by Candidate Selection? Corrigendum 103 SD 02 MA 94 MO 98 PA 94 NV 98 NY 94 WI 92 FL 98 WA 00 TN 94 TX 96 DE 02 TX 94 OR 02 MI 02 UT 94 UT 00 IA 96 NM 00 CT 98 ME 00 OK 92 DE 00 RI 00 MN 00 MD 92 KY 96 VA 96 NC 96 MD 94 KY 02 GA 98 AR 02 NM 02 PA 00 GA 92 NY 98 KY 92 SC 98 MO 02 VT 00 AR 92 VT 92 VA 00 IL 02 MO 00 ID 02 IA 02 AL 02 WI 98 IA 98 OH 92 WA 94 OK 02 IL 98 WV 02 LA 02 MI 00 MT 96 SD 96 ND 00 LA 98 ME 02 IN 00 VA 94 ID 96 MD 00 NE 02 VT 94 CO 98 PA 92 GA 02 WV 00 CO 02 MN 96 OR 92 WA 98 MA 00 KS 98 WI 94 NV 94 OH 00 OR 98 DE 96 OK 96 MT 02 NM 94 TN 00 WY 02 IA 92 WI 00 FL 94 WY 00 TX 00 RI 02 VT 98 MS 96 DE 94 7 6 5 4 3 2 1 0 0 1 2 3 4 5 6 7 Effect on Challenger Vote Figure 2. Estimated effect on challenger vote share moving challenger face from median senate candidate face to actual challenger face, by senate contest. Each line represents the estimated difference in vote share between an election with the actual challenger s face and a hypothetical election with a challenger face at the median of all candidates. Each election outcome is estimated using the results from the probit model in Table 2, column 3, weighted by the statewide partisan proportions estimated by Wright et al. (N.d.).

104 Atkinson, Enos and Hill Table A1. Using candidate facial competence and partisanship to predict individual-level vote choice. Intercept 0.605 0.593 0.218 0.334 (0.155) (0.159) (0.330) (0.418) Cook incumbent risk 0.032 0.026 0.028 0.030 (0.014) (0.015) (0.007) (0.007) Respondent shares challenger party 0.431 0.432 0.371 0.371 (0.025) (0.025) (0.015) (0.015) Respondent shares incumbent party 0.379 0.378 0.324 0.323 (0.025) (0.025) (0.013) (0.012) Challenger facial competence 0.024 0.014 0.020 0.020 (0.021) (0.022) (0.006) (0.007) Incumbent facial competence 0.005 0.004 0.004 0.003 (0.028) (0.028) (0.008) (0.010) Incumbent tenure 0.001 0.001 0.004 0.003 (0.004) (0.004) (0.003) (0.003) Tenure squared 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Incumbent age 0.005 0.005 0.014 0.010 (0.007) (0.007) (0.011) (0.014) Age squared 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Challenger expenditures (logged) 0.005 0.005 (0.004) (0.005) State population (millions) 0.002 (0.002) 1994 Fixed effect 0.009 (0.025) 1996 Fixed effect 0.017 (0.021) 1998 Fixed effect 0.019 (0.022) 2000 Fixed effect 0.009 (0.021) 2002 Fixed effect 0.031 (0.031) 2006 Fixed effect 0.000 (0.037) N 4250 4250 26454 26454 Adjusted R 2 0.496 0.496 0.372 0.373 Std. error of regression 0.351 0.351 0.393 0.393 Ordinary least squares regression coefficients with standard errors in parentheses. Dependent variable is respondent vote for incumbent candidate. Cook Incumbent Risk is coded from 0 for contests classified as safe to 3 for contests classified as tossup. Robust standard errors clustered on state/district.

Is the Face-Vote Correlation Caused by Candidate Selection? Corrigendum 105 Replication of Results Replication of Results Predicted Proportion Dem More Competent Given Our Scores 1.0 0.8 0.6 0.4 0.2 0.0 r = 0.83 Predicted Proportion Dem More Competent Given Our Scores 1.0 2000 2002 0.8 0.6 0.4 0.2 0.0 r = 0.73 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Proportion Todorov et al. Respondents Picking Dem More Competent Proportion Todorov et al. Respondents Picking Dem More Competent Figure 3. Replication of Todorov et al. experimental results. REFERENCES Atkinson, M. D., R. D. Enos, and S. J. Hill. 2009. Candidate Faces and Election Outcomes: Is the Face Vote Correlation Caused by Candidate Selection? Quarterly Journal of Political Science 4(3): 229 249. Wright, G. C., J. P. McIver, and R. S. Erikson. N.d. Aggregated CBS News/New York Times National Polls. Electronic File.