Intergenerational Child Mortality Impacts of Deworming
KLPS deworming team · Kenya Life Panel Survey
Evaluator 1 (public-health lens): Shows great promise in providing evidence about the long-term benefits of deworming as a public health strategy. Several data-presentation questions raised.
Evaluator 2 (econometrics lens): A terrific paper documenting important results. Highly credible given the randomized design. One note on deviations from the pre-analysis plan — clarified by the authors.
Most of the summaries and plain-language explanations here were drafted by AI and then checked and edited by me (David Reinstein, The Unjournal). The evaluations, ratings, claims, and the authors’ response are the evaluators’ and authors’ own words.
Verbatim quote — evaluator or author's own words AI-drafted summary — checked and edited by D. Reinstein
Tell me what you think
This is a rough prototype — one of a few I'm trying (there's an overview with the others). Highlight anything and comment with the Hypothes.is sidebar (top-right), or react on the overview page. I'm mainly wondering whether this is easier to follow than the separate PubPub pages, and what would make it more useful. I'm not sure it's better, so I'd value your view.
1 · The paper
The research
This paper uses two decades of Kenyan panel data to ask whether deworming a child's parents — a generation earlier — lowers that child's risk of dying before age five. The answer seems to be yes, though the mechanisms are still being worked out.
AI-drafted summary
Kenya's Primary School Deworming Project (PSDP, 1998–2001) randomly assigned roughly 75 schools to earlier treatment and 25 to later treatment. The Kenya Life Panel Survey (KLPS) has tracked participants ever since. This paper asks a novel cross-generation question: did inducing a parent to receive more years of deworming, as a child, reduce infant and under-five mortality among their children?
Main finding: children of parents induced to receive more deworming years had significantly lower infant and under-5 mortality.
Both parent sexes: the effect holds for children of both female and male participants.
Mechanisms explored: parental health, education, living situation, fertility, and health-care-seeking behaviour.
Design: exploits the school-level random rollout of PSDP; KLPS data now spans more than 20 years.
Data and design
The identification strategy is an instrumental-variables design. Treatment assignment to early-deworming schools instruments for the number of deworming years received. Take-up in the treatment group was approximately 75%; the control group received minimal treatment (≈5%). The paper uses the KLPS panels (waves 1–4) to trace the long-run outcomes of the original participants and, crucially, the birth and survival outcomes of their offspring.
The evaluated version was a working-paper submitted under economics short-paper limits (≤5 exhibits, ≤6,000 words). Authors are expanding it to a full-length article.
Both evaluators independently highlighted the same two implications:
Higher returns to deworming than believed: standard cost-effectiveness analyses for deworming consider first-generation effects on the treated individuals. If benefits carry forward to the next generation — lowering infant and child mortality — then the true social return to deworming may be substantially higher, potentially changing prioritization calculations.
Health as a channel for intergenerational transmission: the finding adds to a growing body of evidence that early-life health is a conduit through which advantage (or disadvantage) propagates across generations. This has implications for the design of public-health and development interventions more broadly.
The paper is directly relevant to global-health cost-effectiveness analysis and cause prioritization, particularly for funders weighing deworming programs against other interventions.
3 · Evaluation
What the evaluators said
Two anonymous experts — one from a clinical/public-health background, one from econometrics and experimental design — reviewed the paper independently.
Overall verdict
"This working paper shows great promise in providing evidence about the long-term benefits of deworming as a public health strategy."Evaluator 1 — verbatim from evaluation report
Key questions raised (public comments to the authors)
Comparison population: The paper compares 2021 UNICEF child-mortality estimates for Kenya to the United States. Why not compare to other low- and middle-income countries (LMICs)?
Baseline characteristics table: No table summarises key demographic characteristics of the treatment groups. This is standard practice and would help readers evaluate comparability.
Dose-response description: The paper hints at a dose-response relationship between deworming and outcomes but does not describe deworming frequency.
Take-up differential: Take-up was approximately 75% in the treatment group but only approximately 5% in the control group. Why was the differential so pronounced?
Cross-cutting experiment exclusions: The paper excludes participants in other cross-cutting experiments from the treatment group but not the control group. Why this asymmetry? (Note: the authors clarify the related PAP point in their inline reply; see Evaluator 2's full evaluation.)
Age at baseline: What proportion of the baseline sample (ages 8–15 at the time) were near the age of aging out of primary school when the programme began?
Verbatim · evaluator's own words · public section only
This working paper shows great promise in providing evidence about the long term benefits of deworming as a public health strategy. I am not sure why the authors compare the 2021 UNICEF estimates of child mortality in Kenya to that of the United States. I would expect the comparison of estimates to be with other LMIC countries.
The authors did not provide a description/table summarizing the key demographic characteristics of the treatment groups (e.g., age, proportion of respondents that were female/male) although they refer to differences in impact of deworming in child mortality and economic living standards.
The authors also hint of a dose-response relationship between deworming treatment and child mortality outcomes but do not provide a description of how frequently deworming took place (e.g. once a year versus once every 6 months).
If the only difference between the treatment groups was time of receipt of the intervention (deworming), why was the difference in take up of the intervention so wide among the groups (75% in the treatment group compared to only 5% in the control group)?
The authors also mention excluding from the treatment group participants who were also involved in other cross-cutting experiments. However the authors did not do similarly for the control group. Why?
In view of the authors mention of study participants not continuing to receive deworming treatments when they aged out of primary school, what proportion of the baseline sample (8–15 years) of both treatment and control groups were closer to the age/grade when they aged out of primary school?
E1's "Confidential comments to the Associate Editor" are excluded from this public presentation per Unjournal policy. The published evaluation on PubPub contains the full public report.
Overall verdict
"This is a terrific paper documenting important results. The results are highly credible given the randomized design."Evaluator 2 — verbatim from evaluation report
Two key implications (from this report) AI-drafted summary
Deworming interventions may have higher returns than previously believed, because benefits extend across generations.
Health is an important channel driving the intergenerational transmission of advantage.
Major comment — apparent deviations from the pre-analysis plan
The paper cites AEA-registered plan AEARCTR-0001191. E2 flagged several apparent deviations:
Second-generation child-health outcomes were pre-registered as secondary (to a primary fertility analysis). Only under-5 mortality — not infant mortality — was in the PAP. The pre-registered "child health index" does not appear in the manuscript.
The PAP proposed including participants in a later cost-sharing RCT and using an F-test of joint significance. The manuscript excludes cost-sharing-treatment schools, up-weights controls, and omits pre-specified multiple-hypothesis-adjusted p-values.
E2 adds: "I believe the approach taken in the manuscript… is superior to what was pre-specified" (simpler; avoids multi-arm biases, citing Goldsmith-Pinkham, Hull & Kolesár, AER 2024) — "However, it would have been nice for the authors to note that they made this deviation from the original PAP."
The authors respond directly to this point. Their reply appears as a margin note beside the relevant passage inside the full evaluation below (open "Full evaluation"), and is summarised in the inline-reply pointer.
Minor comments
First-stage statistics: Present formal first-stage statistics (how much more deworming the treated actually received). With take-up ≈75%, the effective additional deworming years may be approximately 1.7–1.8, rather than the stated 2.41.
Attrition table: Show baseline characteristics of respondents versus non-respondents by treatment arm. E2 notes Table A1 attrition is impressively low.
Marriage quality / assortative matching: Consider this as an additional mechanism, given that both parents' health outcomes matter.
Verbatim · evaluator's own words · complete report
Intergenerational Child Mortality Impacts of Deworming: Experimental Evidence from Two Decades of the Kenya Life Panel Survey
This manuscript examines the intergenerational consequences of a public health deworming intervention conducted at the school level at 75 schools in Kenya between 1998 and 2001 called the Primary School Deworming Project (PSDP). The PSDP gave free de-worming drugs to an essentially randomly-selected 50 of the 75 schools earlier (in 1998 and 1999), with the remaining 25 receiving them later (in 2001). As a result, the 50 early schools received more years of the deworming intervention than the later treated schools.
The authors take advantage of this intervention to investigate whether the beneficial effects of de-worming for the parents' generation result in better health for their children. To measure these intergenerational effects, the authors use high-quality panel data collected via the Kenya Life Panel Survey (KLPS). The primary outcomes they consider are infant mortality and under age 5 mortality. The analysis shows that the children of parents induced to receive more years of deworming medication had children with significantly lower mortality rates. Lower mortality rates were observed both for the children of female parents and male parents. This result is surprising as we might hypothesize different mechanisms through which a mother versus a father might confer intergenerational health advantages.
The analysis then explores potential mechanisms for this reduced infant mortality, including measures of parental health, education, living situation, fertility, and health care, and finds some evidence that all of these may have moved, suggesting that the portfolio of improved outcomes resulting from the deworming program may be responsible for the improved child mortality outcomes observed.
This is a terrific paper documenting important results. The results are highly credible given the randomized design. The two most important implications are: first, that these de-worming interventions could have higher returns than previously believed, because the benefits extend across generations. Second, it is well known that higher socio-economic status parents tend to confer this advantage to their children. It is also well documented that early life health is predictive of adult health, educational attainment, and earnings. So, these results suggest that health may be an important channel driving the intergenerational transmission of advantage.
I had one major comment regarding the paper and a few minor points, which I make below.
Major comment
Deviations from pre-specified analysis. High-quality experiments typically publicly pre-specify their analysis plans in order to avoid the appearance of searching across specifications and outcomes for significant effects and selectively reporting those estimates. However, over time, researchers may change their opinion about the best way to approach a topic or the appropriate emphasis to place on certain estimates versus others. Researchers are, naturally, free to deviate from or amend pre-specified analysis plans over the course of their investigation. However, it is appropriate, when doing so, to make note of meaningful deviations somewhere in the manuscript, ideally with an explanation for why these deviations occurred.
This paper cites AEA-registered plan AEARCTR-0001191 as its pre-analysis plan (PAP). I read these analysis plan documents and noted several meaningful differences between this paper and the pre-specified analysis. The most meaningful deviations I noticed were:
Second generation child health outcomes were pre-registered as "secondary" outcomes related to a primary analysis of participants' fertility rather than focal outcomes themselves. Only under 5 mortality, and not infant mortality, was included as a possible outcome in the PAP. The first child health outcome listed in the PAP is the "child health index," described as "mean-effects index based on the respondent-reported subjective health of their child, and a reverse-coded sum of indicators for common health symptoms children can experience." But, this child health index does not appear in the paper.
The econometric specification in the PAP proposes including participants in a later randomized controlled trial (that randomized schools into a cost-sharing scheme) in the analysis and estimating the coefficient on an indicator for those at schools selected for this cost-sharing treatment. The PAP proposes a secondary hypothesis test that uses an F test to examine the joint significance of the original de-worming treatment and the cost-sharing treatment. In contrast, the manuscript excludes those from schools selected for the cost-sharing treatment and up-weights the control group participants from the cost-sharing experiment, including an indicator that the individual was in the cost-sharing control group. This is a different approach than pre-specified. In addition, the PAP states that the authors will report multiple hypothesis adjusted p-values, but the manuscript does not contain them.
I believe the approach taken in the manuscript (to exclude the second cost-sharing experiment treatment group from the analysis and focus on the main randomization) is superior to what was pre-specified both because it is simpler and because it avoids potential biases that can arise when there are multiple treatment arms (see Goldsmith-Pinkham, Hull, and Kolesar AER 2024). However, it would have been nice for the authors to note that they made this deviation from the original PAP.
Minor comments
The authors do not present formal statistics on the "first stage"—i.e., how much more likely were participants who attended a group 1 or group 2 school to receive any de-worming medication compared to those who attended a group 3 school, and how many more years of de-worming medication did they actually receive. The paper states that group 1 and group 2 schools received "on average" 2.41 additional years of exposure to the PSDP, but take-up was about 75%, with some amount of treatment reaching the control group. So, we might think of the number of additional medication years was lower, perhaps closer to 1.7 or 1.8 years. It's also likely the difference in the likelihood of receiving any de-worming treatment during these adolescent years is small between the two groups. More formal analysis on this would be helpful in interpreting the magnitude of the reduced form differences in child health outcomes across the treatment and control groups.
Table A1, documenting the low attrition rates of the KLPS survey waves, is very impressive. Very few studies are able to achieve such a high follow-up rate over such a long period. One additional helpful piece of information would be to show differences in baseline characteristics across respondents and non-respondents by treatment arm.
In a footnote, the manuscript mentions that marriage quality may be a mechanism, and is an area for future research. Given that the results show that both mother's and father's health is equally important in generating intergenerational health benefits, this seems like a relevant outcome to consider for this paper. E.g., if fathers exposed to the deworming program matched with healthier mothers, it is possible that the (untreated) mother's health is the mechanism responsible for the improvement infant and child mortality outcomes.
Authors' replies are shown inline
The authors' point-by-point replies now appear beside the passage each one answers, inside Evaluator 2's full evaluation above. On a wide screen they render as green-ruled "Authors' reply" margin notes to the right of the relevant text; on a narrow screen they fold in directly under the passage. Two replies are anchored there:
Pre-analysis plan. In plain terms, E2 read the manuscript as excluding the cost-sharing schools and treated that as a PAP deviation. The authors clarify that only participants in the later, separate TVVP (vocational-training) and SCY (cash-grant) experiments were excluded — a different group — while the cost-sharing schools remain in the analysis as the PAP required, so the apparent discrepancy dissolves. See the reply.
Short-paper format. Many of the evaluators' requests for extra tables and analyses reflect the ≤5-exhibit / ≤6,000-word short-paper limits; the authors are expanding the paper to a full-length article. See the reply.
4 · Ratings
Qualitative verdict and ratings
Evaluator stances (qualitative)
Evaluator 1 (public-health lens): The paper shows great promise in documenting long-term deworming benefits. Key requests centre on clearer data presentation — comparison populations, baseline tables, dose-response description.
Evaluator 2 (econometrics lens): Highly credible results given the randomized design. One note on an apparent pre-analysis-plan deviation, which the authors clarified is not a deviation. Incremental improvements requested on first-stage statistics, attrition tables, and mechanisms.
Quantitative ratings and predictions for this package are available on the PubPub evaluation: unjournal.pubpub.org.
No numeric ratings were included in the source material used to build this page. This presentation focuses on the prose and the author-response dialogue. The PubPub evaluation pages contain the full structured ratings data.
5 · Author response
Authors' response — overview
Michael Walker, on behalf of the paper authors
"We thank the Unjournal editor and evaluators for helpful comments. We are currently in the process of revising the paper and will update this note to point readers to relevant parts of the revised paper when available."
The authors' two substantive replies are now shown inline, beside the passage each one answers, inside Evaluator 2's full evaluation above: the clarification of the pre-analysis-plan point, and the explanation of the short-paper format constraints. On a wide screen they appear as green-ruled "Authors' reply" margin notes; on a narrow screen they fold in under the passage. See the inline-reply pointer for jump links.
6 · Process & status
Transparency & what is next
Evaluator selection
Two evaluators were selected for complementary expertise: one with a clinical/public-health perspective and one with an econometrics and experimental-design perspective. Both reviewed the same working-paper version.
Anonymity
Both evaluators are anonymous in this public package. The Unjournal's standard practice is to offer evaluators the option of named or anonymous publication; identities remain confidential here.
Pre-analysis plan
The paper is registered as AEA trial AEARCTR-0001191. The apparent PAP-deviation issue flagged by E2 was clarified by the authors; see the centrepiece exchange.
Status at time of evaluation
The authors were revising the paper to a full-length article at the time of the author response. The author response note commits to updating once the revision is available. Revision in progress · author response posted