The Effect of Cash Transfers on Political Support

Manacorda, Miguel & Vigorito (2011) investigate whether Uruguay's PANES anti-poverty programme — a cash and food-card transfer to the poorest households — causally increased political support for the incumbent centre-left government. This replication uses a similar dataset covering ±0.02 around the eligibility cutoff and reconstructs the full empirical pipeline from scratch in Stata.

PANES (Plan de Atención Nacional a la Emergencia Social) ran from 2005 to 2007 and targeted households below a predicted income score threshold. Participants were surveyed at programme start (2005), 18 months into the programme (2007), and one year after it ended (2008). Self-reported political support was measured as a preference comparison of the incumbent government relative to the previous one.

The key question: does receiving a government transfer causally shift political loyalty toward the distributing government? This has first-order implications for how democratic governments deploy social spending and whether welfare programmes function as electoral tools.

The core identification challenge is that PANES eligibility is correlated with poverty and pre-existing political preferences — a simple comparison of recipients and non-recipients would be biased. The Regression Discontinuity Design (RDD) solves this by exploiting the sharp eligibility rule: households below the predicted income score cutoff (set to 0 after centring) received treatment, while those just above did not.

Near the cutoff, assignment is as good as random — households just below and just above the threshold are virtually identical in all baseline characteristics. This local quasi-randomisation yields an unbiased estimate of the Local Average Treatment Effect (LATE) for households around the poverty threshold.

The full analysis pipeline in Stata:

Visual Discontinuity Check

A preliminary plot of binned mean government support by income score (Figure 1) immediately reveals a visible jump at the cutoff — the first piece of evidence that PANES eligibility affects political support. This visual inspection motivates the full regression analysis.

McCrary Density Test

The McCrary test (McCrary, 2008) formally tests whether the density of the running variable is continuous at the cutoff. A significant discontinuity would indicate that households strategically manipulated their predicted income score to gain PANES eligibility — which would invalidate local randomisation.

A valid RDD requires that only the outcome — not pre-treatment characteristics — jumps at the cutoff. The table below regresses each baseline covariate on treatment status (being PANES-eligible). Significant coefficients would threaten the randomisation assumption.

Six of seven covariates show no significant jump at the cutoff, supporting the validity of the RDD. Mean household size is the exception (coefficient -0.273**, suggesting PANES-eligible households are smaller), which may reflect the income score prediction formula's weighting of household composition. To be conservative, all subsequent models include household size as a control variable.

The main regressions estimate the causal effect of PANES eligibility on self-reported government support. Four specifications are reported: OLS linear (1), second-order polynomial (2), OLS linear with household-size control (3), and polynomial with control (4).

The effect is consistent and significant across all four specifications. The treatment effect ranges from 9.3–13%. Tighter standard errors in linear models reflect a better fit near the cutoff. Notably, the effect persists into 2008 (post-PANES), suggesting that political loyalty induced by cash transfers does not immediately dissipate once the programme ends.

Beyond overall government support, I estimated the effect on confidence in the current president, confidence in the Ministry of Social Development (MIDES, the agency that administered PANES), and self-reported interest in politics. These provide a richer picture of how welfare transfers shape political behaviour.

PANES eligibility raised confidence in the Ministry of Social Development by 13–15% and in the president personally by 8–12%. The significant effect on interest in politics (linear model) suggests that cash transfers may also activate political participation — not merely shift support toward the incumbent. If recipients become more interested in politics, they may be more likely to vote, translating self-reported support into actual electoral outcomes.

Donut-Hole Regression (±0.005 exclusion)

Observations within ±0.005 of the cutoff are excluded to test whether the result depends on observations most susceptible to near-cutoff heaping or measurement error.

Results remain significant and the linear coefficients are slightly larger than in the main analysis. The inflated polynomial coefficients (0.32–0.45) suggest that near-cutoff observations slightly suppress the quadratic estimate — a key finding of this extension. The core result of a 10%+ treatment effect is unaffected by removing near-cutoff observations.

Adjusted Donut Regression (±0.002 exclusion)

A narrower donut of ±0.002 balances robustness to near-cutoff heaping with preserving statistical power.

With the adjusted donut, all four specifications achieve 1% significance for both years — stronger evidence than the main results. Polynomial coefficients settle at 16–21%, closer to the main estimates than the ±0.005 donut. This suggests the ±0.002 donut provides the best balance between robustness and precision, and that the true treatment effect is conservatively around 10–17%.

Local Polynomial Regressions (rdrobust, Epanechnikov kernel)

Nonparametric local polynomial regression using the rdrobust Stata package with bias-corrected inference. This approach uses data-driven bandwidth selection (MSERD and MSETWO) and an Epanechnikov kernel that down-weights observations further from the cutoff.

The 2007 local polynomial results are consistent with the OLS findings: a 12.5–17.7% effect depending on bandwidth. The mixed 2008 local polynomial results warrant caution — the post-PANES persistence found in OLS is less robust to the nonparametric approach, particularly with narrower bandwidths.

The final graphical analysis uses cmogram and rdplot Stata commands to present the treatment effect with alternative visualisation approaches. Cmograms show confidence intervals; rdplot focuses on the discontinuity at the cutoff without assuming a constant standard deviation on each side.

Methodological Limitations

---

Policy Implications

The finding that cash transfers increase political support, confidence in the implementing ministry, and interest in politics has several implications for policy design:

• Well-designed anti-poverty programmes with visible government branding can generate political dividends beyond their direct welfare impact
• The persistence of the effect into 2008 (post-PANES) suggests that loyalty effects from welfare transfers are not purely transactional — they may reflect genuine shifts in political attitudes
• The large effect on MIDES confidence (13–15%) implies that how transfers are delivered matters: personal contact with government personnel amplifies political effects beyond the monetary benefit alone
• Policymakers in other countries should not assume these magnitudes generalise: Uruguay in 2005–07 was a distinctive context — a newly elected left-wing government implementing its first major social programme

The original Manacorda et al. (2011) paper does not include a bandwidth sensitivity analysis — a significant gap, given that bandwidth selection is one of the most consequential analytical choices in RDD. This report fills that gap by systematically reporting treatment effects at four bandwidths (0.005, 0.01, 0.015, 0.02) for three estimators: OLS, second-order polynomial, and local polynomial (rdrobust).

All three estimators produce bandwidth-stable results for 2007, with coefficients consistently in the 9–13% range and confidence intervals excluding zero throughout. This provides the strongest possible evidence that the treatment effect is not a statistical artifact of a particular bandwidth choice. The conservative estimate of ~10% political support gain from PANES eligibility stands as the primary conclusion.

The 2008 local polynomial results are more mixed, raising legitimate questions about the long-run persistence of the effect. The OLS evidence for persistence is stronger, but future work with larger datasets and wider bandwidths would be needed to resolve this definitively.