Strategic complementarities and social transfers: how do PROGRESA payments impact nonbeneficiaries?

Abstract

The article examines local economy effects of social transfers by focusing on food consumption and asset holdings of noneligible households in rural Mexico following the introduction of Programa de Educación, Salud y Alimentación (PROGRESA) in 1997. The quasi experimental nature of the evaluation data collected for the purposes of evaluating the impact of PROGRESA enables comparison of welfare indicators among noneligible households in treatment areas and control areas. The analysis finds that noneligible households in treatment areas show significantly higher levels of food consumption and asset holdings following the introduction of PROGRESA, compared to noneligible households in control areas. These results are interpreted to suggest that transfers in poor rural areas in Mexico enable agents to interact more strategically such that nonbeneficiaries, as well as beneficiaries, reap consumption and production advantages.

Acknowledgements

We gratefully acknowledge financial support from UK Government's Department for International Development, Social Protection Team, Policy Division. We also gratefully acknowledge advice on panel estimation from Ricardo Sabatés and help with some variable creation from Jen Leavy. The views and opinions expressed are those of the authors alone.

Notes

¹ See, for example, comparative studies of conditional transfer programmes (Morley and Coady, 2003; Das et al., 2005; Rawlings and Rubio, 2005).

² Because our empirical work focuses on nonbeneficiary households, we are not in a position to examine community multipliers. Several studies have identified private multipliers among direct beneficiaries of transfers. See Barrientos and Sabatés-Wheeler (2009) for a review.

³ Further evaluation survey was collected in November 2003, which included further 151 communities to act as the new control group. These locations were selected using propensity score matching techniques. This dataset was not used in this report.

⁴ The dataset for March 1998 included questions on expenditures, but these were formulated in a different way to the other follow up surveys and will not be used in the empirical work below.

⁵ The localities were randomly selected using proportionate to size probabilities from the full sample of communities in seven states in which the programme was implemented by November 1997 and from the full sample of communities in which the programme was planned to begin in by December 2000 (Skoufias, 2001).

⁶ In the last period, period four, PROGRESA was fully rolled over to all areas, which means that for this period we cannot differentiate between treatment and control areas. The justification for including this period data was to compare the impact on nontreated households in treatment areas that have received treatment for more than one period versus nontreated household in treatment areas who have just received the treatment.

⁷ We acknowledge that we could have estimated fixed effects linear probability models for our dichotomous outcome variables or fixed effects linear probability models for truncated outcome variables and ignore the structure of the outcome variable. We opted, however, for the correct modelling of the outcome variable and the inclusion of random effects to deal with time invariant heterogeneity.

⁸ Where only quantity of food purchased or consumed was reported, a monetary value was imputed using median prices.

⁹ The actual question in the survey instrument asks: how many plots of agricultural, livestock, or forestry land are owned or used by members of the household in the last 12 months? This is followed by questions on the hectares and use of each of these plots, up to five in the baseline 1997 Survey.

¹⁰ Due to limitations of space the table reports on the parameter of interest only, the full regression results are available from the authors (Barrientos and Sabatés-Wheeler, 2009). The estimated model includes individual (age, gender, language and work status); household (income, land holdings and shocks); and community level (poverty, marginalization and region) covariates.

¹¹ We tested for the robustness of these findings. We ran the same models eliminating consumption outliers, and we also re-estimated the equations using consumption values rather than logs. The sign and significance of the parameters were unchanged.

¹² The period covered by the data are characterized by structural changes in the rural economy in Mexico, showing a decline in agricultural production. PROGRESA was introduced to mitigate the impact of structural change on the poorest rural households.

¹³ A detailed discussion of this issue can be found in Barrientos and Sabatés-Wheeler (2009).