The Politics of Informal Redistribution in Latin America Appendix

Appendix I implemented a face-to-face public opinion survey of 900 voters in Bogotá, Colombia. The survey was administered by the Bogotá-based polling firm Cifras y Conceptos between August 5 and 29, 2013. On average, the survey interview lasted 25 minutes. A clustered random sample was generated within the city. Thirty-six polling stations were selected as the primary sampling units (PSUs), with 25 cases conducted in each PSU. To ensure sufficient power for the survey experiment, 12 polling stations were selected from lower class groups (Strata 1 and 2), 12 polling stations from lower-middle class groups (Strata 3), and 12 polling stations from middle and upper class groups (Strata 4, 5, and 6). Survey weights must be used to adjust for the oversampling of upper class groups to make population-representative statements. Interviewers began from a randomly selected corner in the PSU and proceeded in a clockwise direction. Interviewers used the different questionnaires in sequential order and rotated between asking for a male and female respondent in the household. Interviewers were part of the survey firm’s trained professionals. Training was conducted for interviewers on the logistics of the survey instrument and issues of respondent protection such as anonymity and privacy. On a separate sheet from the questionnaire, interviewers recorded the first name only and phone number of each respondent for the purposes of later supervision. Postsampling verification was conducted on a randomly-selected 30 percent of the sample by telephone, after which this information was destroyed. The response rate for the survey was 15.6 percent; the cooperation rate was 23.7 percent, the refusal rate 23.3 percent, and the contact rate was 36.4 percent. Table A1 presents summary statistics for the variables used in the analysis. Table A1. Summary Statistics, Bogotá Survey Variable Vending Squatting Decomission Eviction Electricity Tax Class Right Vendor past Squatter past Education (range) Female Age (in years) Unemployed Mean 0.512 0.251 0.197 0.419 0.106 0.075 0.434 0.540 0.228 0.112 0.638 0.491 47.636 0.055 Std. Dev. 0.382 0.332 0.308 0.381 0.218 0.181 0.307 0.301 0.42 0.315 0.304 0.5 16.049 0.228 Min. 0 0 0 0 0 0 0 0 0 0 0 0 19 0 Max. 1 1 1 1 1 1 1 1 1 1 1 1 90 1 N 895 894 897 888 895 895 899 763 895 894 899 890 899 889 To assess the claim of random assignment to treatment, Table A2 presents a balance check, comparing groups that received each treatment on measured covariates such as class, education, gender, age, political ideology, and past history working as a street vendor. All variables, except for age (which is measured in years), are rescaled from 0 to 1 for ease of interpretation. Table A2. Balance Check Candidate 1 Class Education Female Age Right Vendor Past Redistribution (A) 0.433 0.635 0.484 48.32 0.545 0.225 Business (B) 0.433 0.639 0.499 47.00 0.535 0.231 Candidate 2 Forbearance (C) 0.434 0.632 0.524 47.35 0.560 0.225 Enforcement (D) 0.433 0.642 0.458 47.98 0.520 0.231 Balance Tests Difference of Means (A-B) (C-D) 0.000 0.000 -0.003 -0.010 -0.015 0.065 1.32 -0.630 0.090 0.040 -0.006 -0.006 p-value (A-B) (C-D) 1.00 0.965 0.869 0.622 0.662 0.052 0.891 0.278 0.671 0.068 0.831 0.820 The p-values in the final column give the probability of observing a t-statistic as large in absolute value as the observed value, if the groups are drawn from the same distribution. N = 899 respondents total. An omnibus F-test also found no imbalance (the first candidate pair p < 0.9492, the second candidate pair, p < 0.5498). While the paper presented the results for each candidate, difference-of-means tests produce similar results. To do this, I create comparisons between candidate pairs: 1) enforcement-forbearance and 2) redistribution-business. My key hypothesis is that there should be a significant difference between the first pair of candidates, but not the second. In other words, enforcement positions are more polarizing than standard social policy ones. Gender and political ideology do come close to statistical significance, so I also corroborate my results using regression analyses in which the experimental treatment is included as an independent variable with and without covariates, namely gender, age, education, political ideology and vending past in Tables A3. The results are substantively unchanged. In the campaign platform database, the rich districts analyzed are Jesús María, La Molina, Lince, Magdalena del Mar, Miraflores, Pueblo Libre, San Borja, San Isidro, San Miguel, and Santiago del Surco. The poor districts are Ate, Ancón, Carabayllo, Comas, El Augustino, Lurigancho, Lurín, Puente Piedra, San Juan de Lurigancho, San Luis, Santa Anita, Villa El Salvador, Villa María del Triunfo, and Ventanilla. Table A3. Treatment Effects with Covariates Redistribution Treatment Forbearance Treatment Class Female Education Age Ideology Vendor Past R2 (1) (2) Benefits the Poor -0.007 (0.02) -0.342ú (0.02) 0.018 0.124ú (0.05) (0.05) 0.001 0.049ú (0.02) (0.02) -0.024 -0.087 (0.05) (0.05) 0.000 0.001 (0.00) (0.00) -0.006 -0.060 (0.04) (0.04) 0.017 -0.023 (0.03) (0.03) 0.007 0.229 (3) (4) Benefits Me 0.001 (0.02) -0.206ú (0.02) -0.114ú 0.016 (0.05) (0.05) 0.035 -0.002 (0.02) (0.02) 0.060 -0.016 (0.05) (0.05) -0.000 -0.000 (0.00) (0.00) -0.031 -0.085ú (0.04) (0.04) 0.033 0.001 (0.03) (0.03) 0.014 0.100 (5) (6) Vote Intention -0.045 (0.03) -0.237ú (0.03) 0.017 0.083 (0.05) (0.05) 0.039 -0.005 (0.03) (0.03) 0.031 -0.036 (0.05) (0.05) 0.000 0.001 (0.00) (0.00) -0.015 -0.055 (0.05) (0.05) 0.029 0.018 (0.03) (0.03) 0.009 0.114 (7) (8) Ideological Placement 0.081ú (0.03) 0.136ú (0.03) -0.048 -0.010 (0.05) (0.06) -0.008 -0.017 (0.03) (0.03) 0.087 0.073 (0.06) (0.06) -0.001 0.000 (0.00) (0.00) 0.110ú 0.040 (0.05) (0.05) -0.023 -0.014 (0.03) (0.04) 0.029 0.039 The redistribution treatment reflects the difference between the pro-redistribution and pro-business candidate information; the forbearance treatment shows the difference between the forbearance and enforcement candidate information. All dependent variables are rescaled from 0 to 1 and run from “very unlikely” to “highly likely”, or in the case of ideological placement, from the far Right to Left.