Assignment 3: PROGRESA conditional cash transfers

Assignment 3: PROGRESA conditional cash transfers and
secondary school enrollment
October 31, 2022
The data come from Bobonis and Finan (2009, Review of Economics and Statistics) who looked
at the effect of PROGRESA conditional cash transfers on school enrollment and the role of peer
effects. Data was collected in 1997 (baseline) and by 1998 (endline/post=1) only some villages
(treat=1) had received the transfers. Treatment across villages was randomized. Within each vil lages, only households with a low enough welfare index were considered poor and therefore eligible
(eligible=1) to receive transfers; other households were deemed ineligible (even in the treatment
villages). The data include the following variables for students and heads of households (collected
at baseline unless otherwise specified):
folio – household ID
numero – individual ID
year – year
village – village ID
sc – school enrollment
eligible – PROGRESA eligible
welf are – eligibility welfare index (lower for the poor)
post – =1 for 1998, =0 for 1997
treat – PROGRESA treatment village (=1), control (=0)
gender – (=1 Male, =0 if Female)
indig – (=1 indigenous, =0 otherwise)
hohedu – Head of household schooling attainment
hohsex – Head of household gender (=1 Male, =0 Female)
hohage – Head of household age
f amn – household size
distsec – minimum distance to secondary school
mindist – minimum distance between the village and large urban center
meanvillsc – average school enrollment of eligible children in the village/year
1
(a) What can we say about the determinants of secondary school enrollment based on the
(cross-sectional) relationships in the data (for 1997)? [2pts]
(b) At what level should you cluster the standard errors in (a)? [1pt]
(c) Which characteristics are systematically (statistically) different between eligible (poor) and
ineligible (“rich”) households? [2pts]
(d) Provide evidence that the across-village randomized treatment was really randomized [2pts]
(e) Provide an estimate of the average treatment effect (ATE) of PROGRESA transfers (on
school enrollment) for the eligible children and the ATE for the ineligible children using the pro gram randomization and interpret the results [2pts]
(f) At what level should we cluster the standard errors in parts (d) and (e)? [1pt]
(g) We can also estimate the ATE of village-level PROGRESA treatment using a difference in-difference estimator – do this for the eligible children and discuss how these estimates relate to
your estimates in (e) as well as what we learn about overall trends in schooling [2pts]
(h) Suppose we only had data from treatment villages. Use the eligible and ineligible children
to estimate an average treatment effect using a difference-in-difference specification. Discuss the
potential bias from this estimate. [2pts]
(i) Suppose we wanted to estimate the causal effect of the village-level school enrollment rate
of eligible children on school enrollment of ineligible children. Provide an OLS and a IV (2SLS)
estimate of this relationship [hint: you can use the randomized treatment as the instrument, and
can calculate the IV coefficient in excel using the reduced form and first-stage coefficients], discuss
whether the IV conditions are satisfied and interpret the results. [2pts]
(j) Suppose we only had data from treatment villages in the post-period (1998). Noting that
eligibility for the program is related to the welfare (poverty) index, provide as close to a causal
estimate as one can estimate given the data limitations and discuss the intuition for why your
approach is superior to simply regressing school enrollment on program eligibility and available
control variables [hint: think in terms of the logic of a regression discontinuity design] [4pts]
2
3