570 Panel Data Methods
intervention aimed at children. A program aimed at the eradication of hookworm is
shown to lead to a long-term gain in the income of beneficiaries. Areas with greater
scope for benefiting, due to higher levels of hookworm infection, show greater
contemporaneous increases in school enrollment and attendance and in literacy
among children. The natural experiment is the Rockefeller Sanitary Commission’s
(RSC’s) funding of treatment and education programs to eradicate hookworm in
the Southern US, which took place around 1910–15. This policy intervention
was implemented over a well-defined and relatively short period. Geographic
differences in infection rates prior to the intervention can be used to formulate
and identify a treatment/control design, estimated using difference-in-differences.
Data on long-term consequences for the cohorts exposed to the eradication pro-
gram are obtained from the US Census available through the Integrated Public Use
Micro Sample (IPUMS).
In 1966 the Ceau ̧sescu regime in Romania banned abortion and family plan-
ning. Birth rates doubled the following year. In Pop-Eleches (2006) this provides
an interesting contrast to studies that have examined moves in the opposite direc-
tion in the United States. The raw data show an improvement in educational
attainment and labor market outcomes associated with the ban, but these results
are reversed by allowing for compositional changes in the type of families hav-
ing children. The findings are explained by the fact that affluent urban women
were more likely to have abortions and use contraception before the ban. Data
are drawn from a 15% sample of the 1992 Romanian census and focus on chil-
dren born between January and October 1967. There is a spike in births between
July and October due to the ban, but all of these children entered school in
the same year and experienced the same overcrowding effect. Although it is not
labeled as such, the paper uses a discontinuity design estimating a simple dif-
ference equation that includes a dummy variable for the period after the policy.
Additional covariates are included, but there is no control group and identification
relies on any sudden changes in outcomes for those born just before or just after
the ban.
Lakdawallaet al.(2006) present a careful application of the method of instrumen-
tal variables, based on state-level variation in Medicaid eligibility in the US, which
shows that an unintended consequence of highly active antiretroviral therapy
(HAART) is to increase risky sexual behavior among patients who are HIV+. Sim-
ple correlations show lower sexual activity among those who are HIV+, but this is
because of the debilitating effects of the disease and does not show the causal effect
of treatment. Panel data on HIV+patients in care are taken from the HIV Costs and
Sevices Utilization Study (HCSUS) for the period 1996–98. The outcome of inter-
est is the number of sex partners of the previous six months and the treatment
is HAART, which is inferred from records of medications. Simple, unconditional
estimates do not show a difference in sexual activity. But when treatment is instru-
mented by variation in the eligibility rules for Medicaid across states a positive
effect emerges. The validity of these instruments is checked by examining the
reduced form association between Medicaid eligibility and sexual activity prior to