Poor Voters vs. Poor Places Cesar Zucco Princeton University Oxford, December 2010
Outline The electoral shift (2006 Elections) A partial explanation (Bolsa Familia) The lurking pattern Neo-governismo or realignments? What s new?
A Puzzling Shift Lula s Vote Share and Level of Development Lula 2002 and before Poor voters don t vote for a poor candidate
A Puzzling Shift Lula s Vote Share and Level of Development Lula 2002 and before Lula 2006 and Dilma 2010
A simple (and incomplete) explanation The largest CCT in the world Reaches 11-13m families Costs 0.4% GDP 35% say fight against poverty most successful policy 67% approved of poverty reduction policies
Electoral Effects of CCTs Best estimates from aggregate electoral data OLS Serra 2002 Lula 2006 Dilma 2010 Scope 0.210 0.215 0.216 SE 0.021 0.012 0.013 N 5456 5466 5475
Electoral Effects of CCTs Best estimates from aggregate electoral data OLS Serra 2002 Lula 2006 Dilma 2010 Scope 0.210 0.215 0.216 SE 0.021 0.012 0.013 N 5456 5466 5475 Propensity Score Matching (Average Treatment Effects) Serra 2002 Lula 2006 Dilma 2010 Scope 0.124 0.196 0.101 SE 0.042 0.031 0.026
Electoral Effects of CCTs Best estimates from aggregate electoral data OLS Serra 2002 Lula 2006 Dilma 2010 Scope 0.210 0.215 0.216 SE 0.021 0.012 0.013 N 5456 5466 5475 Propensity Score Matching (Average Treatment Effects) Serra 2002 Lula 2006 Dilma 2010 Scope 0.124 0.196 0.101 SE 0.042 0.031 0.026
Electoral Effects of CCTs Best estimates from aggregate electoral data OLS Serra 2002 Lula 2006 Dilma 2010 Scope 0.210 0.215 0.216 SE 0.021 0.012 0.013 N 5456 5466 5475 Propensity Score Matching (Average Treatment Effects) Serra 2002 Lula 2006 Dilma 2010 Scope 0.124 0.196 0.101 SE 0.042 0.031 0.026
Electoral Effects of CCTs Best estimates from aggregate electoral data OLS Serra 2002 Lula 2006 Dilma 2010 Scope 0.210 0.215 0.216 SE 0.021 0.012 0.013 N 5456 5466 5475 Propensity Score Matching (Average Treatment Effects) Serra 2002 Lula 2006 Dilma 2010 Scope 0.124 0.196 0.101 SE 0.042 0.031 0.026
Electoral Effects of CCTs 2006 LAPOP Family Income N (Original N) Prob. Vote Lula Risk Ratio (in Min. Wages) Not CCT CCT Not CCT CCT Less than 1 156 (214) 156 (163) 0.59 0.79 1.35 1 5 234 (913) 234 (249) 0.52 0.74 1.43 5 10 22 (273) 22 (23) 0.45 0.68 1.52 10 20 4 (118) 4 (4) 0.35 0.54 1.76 More than 20 (41) (0)
Electoral Effects of CCTs 2006 LAPOP Family Income N (Original N) Prob. Vote Lula Risk Ratio (in Min. Wages) Not CCT CCT Not CCT CCT Less than 1 156 (214) 156 (163) 0.59 0.79 1.35 1 5 234 (913) 234 (249) 0.52 0.74 1.43 5 10 22 (273) 22 (23) 0.45 0.68 1.52 10 20 4 (118) 4 (4) 0.35 0.54 1.76 More than 20 (41) (0)
Electoral Effects of CCTs 2006 LAPOP Family Income N (Original N) Prob. Vote Lula Risk Ratio (in Min. Wages) Not CCT CCT Not CCT CCT Less than 1 156 (214) 156 (163) 0.59 0.79 1.35 1 5 234 (913) 234 (249) 0.52 0.74 1.43 5 10 22 (273) 22 (23) 0.45 0.68 1.52 10 20 4 (118) 4 (4) 0.35 0.54 1.76 More than 20 (41) (0)
Electoral Effects of CCTs 2006 LAPOP Family Income N (Original N) Prob. Vote Lula Risk Ratio (in Min. Wages) Not CCT CCT Not CCT CCT Less than 1 156 (214) 156 (163) 0.59 0.79 1.35 1 5 234 (913) 234 (249) 0.52 0.74 1.43 5 10 22 (273) 22 (23) 0.45 0.68 1.52 10 20 4 (118) 4 (4) 0.35 0.54 1.76 More than 20 (41) (0) Beneficiaries 40% more likely to vote for incumbent!
End of story? Summarizing many hours of research Direct beneficiaries voted massively for Lula (80%) Non beneficiaries voted more for Lula in poorer places Recipients were 40% more likely to vote for Lula than similar non-recipients Direct CCT effects hold accounting for economic growth Possible indirect economic stimulus effect in 2006 No clear evidence of redistribution backlash (Diego?)
End of story? Summarizing many hours of research Direct beneficiaries voted massively for Lula (80%) Non beneficiaries voted more for Lula in poorer places Recipients were 40% more likely to vote for Lula than similar non-recipients Direct CCT effects hold accounting for economic growth Possible indirect economic stimulus effect in 2006 No clear evidence of redistribution backlash (Diego?) Not quite...
Is this a pattern? The Main Subject of the Talk PT 1994 1998 2002 2006 2010 1 st 2 nd 1 st 2 nd 1 st 2 nd PSDB
Is this a pattern? The Main Subject of the Talk PT PSDB 1994 1998 2002 2006 2010 1 st 2 nd 1 st 2 nd 1 st 2 nd Lula Lula Lula Lula Dilma 0.04 0.30 0.34 0.29 0.74 0.71 0.71 0.67 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 FHC FHC Serra Alckmin Serra 0.36 0.16 0.03 0.29 0.69 0.71 0.62 0.67 <0.01 <0.01 0.07 <0.01 <0.01 <0.01 <0.01 <0.01
Is this a pattern? The Main Subject of the Talk PT PSDB 1994 1998 2002 2006 2010 1 st 2 nd 1 st 2 nd 1 st 2 nd Lula Lula Lula Lula Dilma 0.04 0.30 0.34 0.29 0.74 0.71 0.71 0.67 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 FHC FHC Serra Alckmin Serra 0.36 0.16 0.03 0.29 0.69 0.71 0.62 0.67 <0.01 <0.01 0.07 <0.01 <0.01 <0.01 <0.01 <0.01 Incumbent Party Candidate in red
PT Voting Patterns 1989 2010 (President) Municipal Level Pooled Data with State Random Effects In Government N=10942 In Oppostion N=19664 Vote Share 0.0 0.2 0.4 0.6 0.8 Vote Share 0.0 0.2 0.4 0.6 0.8 Lower Level of Development (HDI) Higher Lower Level of Development (HDI) Higher
PSDB Voting Patterns 1989 2010 (President) Municipal Level Pooled Data with State Random Effects In Government N=15304 In Oppostion N=15302 Vote Share 0.0 0.2 0.4 0.6 0.8 Vote Share 0.0 0.2 0.4 0.6 0.8 Lower Level of Development (HDI) Higher Lower Level of Development (HDI) Higher
PMDB Voting Patterns 1974 1986 (Lower House) State Level Results 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 20 30 40 50 60 70 In Government N=26 Vote Share Level of Development (HDI) PI PR 0.3 0.4 0.5 0.6 0.7 10 20 30 40 50 60 70 In Opposition N=104 Vote Share Level of Development (HDI) Still working on these data
Defining the Empirical Regularity Neo-Governismo A built in advantage incumbents have in reaching certain places Government party does better in poorer places Does not mean government always wins Does not mean government gets support from poorest voters
The observational equivalence problem Two stories compatible with the data Lula Centered Lula was stronger in richer places But mostly among poorer people New in 2006, the rural poor voted for Lula Incumbent Centered Incumbents always perform better in poorer places But not among the poor in the richer places New in 2006 the urban poor voter for the incumbent
The observational equivalence problem Two stories compatible with the data Lula Centered Lula was stronger in richer places But mostly among poorer people New in 2006, the rural poor voted for Lula Incumbent Centered Incumbents always perform better in poorer places But not among the poor in the richer places New in 2006 the urban poor voter for the incumbent... And there seems to be two patterns: one across places and another within places
Lula and the urban poor Valid Vote for LULA 2002 (Citywide: 58.1 %) Less than 33% 33 47% 47 54% 54 58% 58 63% More than 63% Valid Vote for LULA 2006 (Citywide: 44.5 %) Less than 33% 33 47% 47 54% 54 58% 58 63% More than 63%
Lula and the urban poor 0 2 4 6 km scale approx 1:280,000 Valid Vote for LULA 2006 (Citywide: 33.7 %) Less than 21% 21 36% 36 47% 47 55% 55 62% More than 62% 0 2 4 6 km scale approx 1:280,000 Valid Vote for LULA 2002 (Citywide: 55.1 %) Less than 21% 21 36% 36 47% 47 55% 55 62% More than 62%
Lula and the urban poor Valid Vote for LULA 2002 (Citywide: 46 %) Less than 23% 23 30% 30 39% 39 45% 45 50% More than 50% Valid Vote for LULA 2006 (Citywide: 29.7 %) Less than 23% 23 30% 30 39% 39 45% 45 50% More than 50%
Electoral swings when incumbency status changes PSDB 1989 and 1994 (becoming the incumbent) FHC 1994 Covas 1989 0.0 0.2 0.4 0.6 0.8 0.4 0.5 0.6 0.7 0.8 Human Development Index
Electoral swings when incumbency status changes Lula 2002 and 2006 (becoming incumbent) Lula 2006 Lula 2002 0.4 0.2 0.0 0.2 0.4 0.6 0.8 0.5 0.6 0.7 0.8 0.9 Human Development Index
Electoral swings when incumbency status changes PSDB 2002 and 2006 (becoming the challenger) Alckmin 2006 Serra 2002 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.5 0.6 0.7 0.8 0.9 Human Development Index
Electoral swings when incumbency status changes Serra 2002 and 2010 (becoming the challenger) Serra 2010 Serra 2002 0.6 0.4 0.2 0.0 0.2 0.4 0.5 0.6 0.7 0.8 0.9 Human Development Index
Thinking About the Neo-Govenismo A few conjectures (and not more than that!) What could be the mechanism? Not ideology Reliance on government State (and only state) delivers benefits What and how it is delivered? Geographical logic of government Not party machines in the peronist sense Cartorialismo revisited
Neo-Governismo vs. CCTs 2006 in historical perspective: If incumbents always perform better in poorer places, Lula s voting patterns would have changed between 2002 and 2006 even without Bolsa Familia
Neo-Governismo vs. CCTs 2006 in historical perspective: If incumbents always perform better in poorer places, Lula s voting patterns would have changed between 2002 and 2006 even without Bolsa Familia Did Bolsa Familia change things or not? Did Bolsa Familia not have any relevant effect?
Neo-Governismo vs. CCTs 2006 in historical perspective: If incumbents always perform better in poorer places, Lula s voting patterns would have changed between 2002 and 2006 even without Bolsa Familia Did Bolsa Familia change things or not? Did Bolsa Familia not have any relevant effect? Back to the data...
Poor Voters vs. Poor Places How new is the new policy? Estimated Vote for the Incumbent Among Poor Voters Vote Share 0.0 0.2 0.4 0.6 0.8 fhc.1994 fhc.1998 serra.2002 lula.2006 dilma.2010 HDI M (Poor) Highest (14.7m) High (8.5m) Middle (8.6m) Low (12m) Lowest (10.9m)
What s Old, What s New? If the incumbency story holds Bolsa Familia merely magnified the bias in poorer places
What s Old, What s New? If the incumbency story holds Bolsa Familia merely magnified the bias in poorer places Bolsa Familia altered voting patterns in the richer places
What s Old, What s New? If the incumbency story holds Bolsa Familia merely magnified the bias in poorer places Bolsa Familia altered voting patterns in the richer places Urban poor voted massively for the incumbent (new!) Potentially, many important implications!
Some possible implications Conjectures and more conjectures... If government has established a direct channel with urban poor Opposition looses its natural constituency Less reliance on public goods Middle class probably loses out
The End
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Lula s Constituency Electoral Data Lula 1989 1994 Lula 1994-1998
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Lula s Constituency Electoral Data Lula 1998 2002 Lula 2002-2006
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Lula and the PT Electoral Data PT Vote (% of Lula s) PT (%) & GDP-PC 1994 34.83 0.30 <0.001 1998 40.93 0.29 <0.001 2002 40.81 0.34 <0.001 2006 29.99 0.11 <0.001 Correlations (5500 municipalities) PT (%) & Lula (%) 0.56 <0.001 0.60 <0.001 0.54 <0.001 0.02 0.263 Lula (%) & GDP-PC 0.06 <0.001 0.27 <0.001 0.28 <0.001 0.65 <0.001
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Back of the Envelope Calculations Based on Individual Level Results: In a poor municipality If recipients vote for Lula with Prob=0.86 If non-recipients vote with Prob=66 Expanding coverage from 80% to 81% Increases vote for Lula in 0.22 pct points Estimated aggregate effects approx. 0.2
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Back of the Envelope Calculations Based on Individual Level Results: In a poor municipality If recipients vote for Lula with Prob=0.86 If non-recipients vote with Prob=66 Expanding coverage from 80% to 81% Increases vote for Lula in 0.22 pct points Estimated aggregate effects approx. 0.2
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Back of the Envelope Calculations Based on Individual Level Results: In a poor municipality If recipients vote for Lula with Prob=0.86 If non-recipients vote with Prob=66 Expanding coverage from 80% to 81% Increases vote for Lula in 0.22 pct points Estimated aggregate effects approx. 0.2 Assuming: 20 million voters in BF range With BF: 16.1 million Without BF: 11.6 million Swing: 4.5 million + indirect effect
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Inferring Individual Behavior from Aggregate Data For each municipality Lula Not Lula Total BF Not BF Total
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Inferring Individual Behavior from Aggregate Data For each municipality BF Not BF Total Lula Known Not Lula Known Total Known Known Marginals are known
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Inferring Individual Behavior from Aggregate Data For each municipality BF Not BF Total Lula Known Not Lula Known Total Known Known Marginals are known Marginals provide upper an lower bounds for inner cells
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Inferring Individual Behavior from Aggregate Data For each municipality BF Not BF Total Lula Known Not Lula Known Total Known Known Marginals are known Marginals provide upper an lower bounds for inner cells Surveys provide prior estimates
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Inferring Individual Behavior from Aggregate Data For each municipality BF Not BF Total Lula Known Not Lula Known Total Known Known Marginals are known Marginals provide upper an lower bounds for inner cells Surveys provide prior estimates EI combines all this information
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Is there redistribution backlash? Effects of CCTs on non-beneficiaries Region N (Original N) Prob. Vote Lula Risk Low Income High Income Low Income High Income Ratio Nordeste 53 (263) 53 (53) 0.43 0.36 0.86 Sudeste 220 (443) 220 (220) 0.31 0.26 0.83 Sul 71 (146) 71 (71) 0.29 0.24 0.83 Centro-Oeste 35 (71) 35 (35) 0.34 0.29 0.84 Norte 17 (77) 17 (17) 0.58 0.52 0.89 No Region 396 (1000) 396 (396) 0.33 0.28 0.84
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Is there redistribution backlash? Effects of CCTs on non-beneficiaries Region N (Original N) Prob. Vote Lula Risk Low Income High Income Low Income High Income Ratio Nordeste 53 (263) 53 (53) 0.43 0.36 0.86 Sudeste 220 (443) 220 (220) 0.31 0.26 0.83 Sul 71 (146) 71 (71) 0.29 0.24 0.83 Centro-Oeste 35 (71) 35 (35) 0.34 0.29 0.84 Norte 17 (77) 17 (17) 0.58 0.52 0.89 No Region 396 (1000) 396 (396) 0.33 0.28 0.84 Differences between income levels are not significant! (p-values > 0.1)
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Is there redistribution backlash? Effects of CCTs on non-beneficiaries Region N (Original N) Prob. Vote Lula Risk Low Income High Income Low Income High Income Ratio Nordeste 53 (263) 53 (53) 0.43 0.36 0.86 Sudeste 220 (443) 220 (220) 0.31 0.26 0.83 Sul 71 (146) 71 (71) 0.29 0.24 0.83 Centro-Oeste 35 (71) 35 (35) 0.34 0.29 0.84 Norte 17 (77) 17 (17) 0.58 0.52 0.89 No Region 396 (1000) 396 (396) 0.33 0.28 0.84 Differences between income levels are not significant! (p-values > 0.1) Differences between regions are larger
More Figures Lula and the PT Back of the Envelope Ecological Inference Backlash Is there redistribution backlash? Effects of CCTs on non-beneficiaries Region N (Original N) Prob. Vote Lula Risk Low Income High Income Low Income High Income Ratio Nordeste 53 (263) 53 (53) 0.43 0.36 0.86 Sudeste 220 (443) 220 (220) 0.31 0.26 0.83 Sul 71 (146) 71 (71) 0.29 0.24 0.83 Centro-Oeste 35 (71) 35 (35) 0.34 0.29 0.84 Norte 17 (77) 17 (17) 0.58 0.52 0.89 No Region 396 (1000) 396 (396) 0.33 0.28 0.84 Differences between income levels are not significant! (p-values > 0.1) Differences between regions are larger