Week 4 - Miguel and Kremer (2004) - Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities.

What is a summary of this paper?

Aim: This paper evaluates the impact of a large-scale deworming intervention on health and education outcomes in rural regions of Kenya.

Empirical strategy: The authors run a randomized control trial (RCT) where randomization into treatment occurred at the school level and was phased in over three different stages (due to resource constraints). Treatment was rolled out to Group 1 in the first year, Group 2 in the second year, and Group 3 in the fourth year. The main comparisons are between Group 1 and Group 2 with outcomes measured during the year when only Group 1 had been treated.

Findings: The authors’ findings suggest that the school-based deworming program caused, on average, a 7.5 percentage point increase in school attendance for treated schools. These effects had significant positive spillovers by lowering infection rates in untreated schools. Bolstered by these positive externalities, the authors conduct a welfare analysis that supports the subsidization of deworming treatment. Their calculations suggest that, compared to other types of interventions aimed at increasing school participation, de-worming is among the most cost-effective in areas with high incidence of intestinal parasites. Despite the positive effects on school attendance and health outcomes, the authors do not find evidence that deworming improved academic test scores.

In your own words, explain why we expect deworming drugs to have sizable health externalities, especially in Western Kenya.

The fact that intestinal parasites are easily spread leads us to expect health externalities from deworming drugs. That is, the individual deciding whether to take the drug does not take into account the fact that eliminating worms in their body will reduce the probability that their classmates get infected. The externalities are likely sizeable in these regions of Kenya because of the unusually high worm prevalence (due to poor sanitation). In addition, the focus is on children, who are more susceptible to these parasites than adults, and who interact with each other closely and regularly while at school.

In your own words, explain what empirical variation allows the authors to separately estimate: the direct effect of the deworming drug, the health externality within schools, and the health externality across schools.

Randomization of deworming treatment across schools makes it possible to separately identify the overall effect of the program (direct effect plus the within-school externality effect) and cross-school externalities.

In estimating the latter, the regression includes the total number of pupils randomly assigned to treatment in primary schools at a distance d from the school in question. That is, the random variation in density of treated individuals across locations allows the authors to estimate externalities due to reduced transmission of the parasites across schools.

To separately identify the direct effect and the within-school externality effect, the authors exploit the fact that not all individuals in treated schools take-up treatment and impose some additional assumptions.

These assumptions are needed because randomization did not take place at the individual level, meaning that those who take-up treatment in the treated schools may self-select into treatment.

Specifically, the authors assume that the average error in net benefits from take-up of the untreated in Group 1 is no larger than that of the untreated in Group 2 (refer to equation (2) in the paper).

In Column (2) of Table VII, which coefficients capture the direct effect of the deworming drug, the health externality within schools, and the health externality across schools?

Direct effect: coefficient for Group 1 interacted with the indicator for receiving treatment when offered (seventh line: b₂ = −0.14). Externality within schools: coefficient for indicator for Group 1 (first line: β₁ = −0.12). Note that the sum of these two coefficients basically adds up to the first line in Column (1). Externality across schools: coefficient for Group 1 interacted with distance dummies (second and third line: γ₀–₃ = −0.26 and γ₃–₆ = −0.13). Note that these are essentially unchanged from Column (1). Refer to equation (3) in the paper for the full regression underlying these parameters.

In Column (2) of Table VII, which coefficient informs us regarding the sign and magnitude of a selection concern that pupils who actually received the treatment in treatment schools differ from other pupils?

The sixth line of the table informs us on the sign and magnitude of selection bias (b₁ = −0.06), which gives the estimate on the indicator for whether individuals received treatment when offered. The interpretation is that, even after controlling for all the covariates in equation (3), taking the treatment when offered (1998 for Group 1; 1999 for Group 2) implies lower levels of parasite infections. Note that those in Group 2 who eventually take the drugs have not been offered treatment yet. This suggests that those participants who are eventually treated in the treated schools are at baseline less likely to become infected. In other words, pupils that take-up treatment are not random, they are self-selecting into treatment.

What is the main takeaway from this paper?

The main takeaway of this paper is that externalities matter. Excluding externalities from the welfare analysis would have led us to believe that deworming drugs are not as effective as other interventions aimed at increasing school participation and student outcomes.

What is the main limitation of this paper?

The main limitation is that the authors include school- and pupil-level covariates to control for pre-treatment differences across schools that were present despite the randomized nature of the experiment. This means that the authors’ estimates can only be interpreted as causal if one is willing to assume that the controls in their regression capture all factors determining baseline differences in the treatment and control groups (this is sometimes called the “conditional independence” assumption). Other limitations you might have mentioned include concerns about external validity. In particular, we do not know the extent to which these results might hold up in other environments where initial school participation levels and infection rates are different. Given that the direct benefits of deworming may be proportional to the severity of infection, the externality benefits are likely to vary non-linearly with the worm burden. Hence, a different prevalence of worms in a region and alternative coverage levels could produce very different results.

Given the evidence provided in this paper, do you think that there should be a government intervention to increase the take-up of deworming drugs and, if yes, what specific government intervention would you suggest?

Even though the intervention generated no improvement in test scores, the paper makes a very convincing case for the existence of positive externalities of deworming drugs. Given that the benefits seem to far outweigh the costs, the specific form of government intervention should be fully subsidized provision at conventional access points (e.g. schools, village centers, etc.) similar to national vaccination campaigns. Free and convenient take-up opportunities would avoid unwanted behaviours (e.g. missing work), and direct provision would prevent abuses such as overconsumption or secondary resale markets.