REVIEW 3 major objections 6 minor 3 references
Unmasking inequility: socio-economic determinants and gender disparities in Maharashtra and India's health outcomes -- Insights from NFHS-5
T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read Women report nearly double the morbidity of men; in India marital status widens the gender gap while insurance, caste, urban residence, and wealth narrow it.
desk verdict The descriptive gender gap in morbidity is likely solid, but the Fairlie decomposition that the paper calls 'key drivers' is too small, partly artifactual, and overstated; worth engaging only as a revised descriptive update. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The mechanism that carries the argument is the Fairlie decomposition, a nonlinear extension of the Blinder-Oaxaca method for logit and probit models. It takes the difference in the mean predicted probability of reporting morbidity between the female and male samples and separates it into a part explained by the observed categorical covariates (age, residence, education, religion, caste, household size, wealth, insurance, and marital status) and a residual. The paper feeds this decomposition with gender-stratified logistic regressions, choosing base categories by variance inflation factors to limit multicollinearity, and uses ROC curves to confirm that the models have discriminatory power. The decomposition's per-variable contributions are the numbers that carry the conclusions about which factors widen or narrow the gender gap.
What would settle it
Re-run the same logistic regressions and Fairlie decomposition using NFHS-5 survey weights and a common age range of 15-49 for both sexes; if the weighted results no longer show women reporting roughly double the male morbidity rate, or if the largest positive contribution is no longer marital status in India and the explained share moves far from the reported -6.4% (India) and 2.9% (Maharashtra), then the central decomposition claims do not survive.
Extended reading notes
Core claim
The paper's central discovery is a quantified decomposition of the gender difference in self-reported morbidity. In India the predicted probability of reporting morbidity is 0.066 for men and 0.124 for women, a gap of -0.058; the explained component is small (0.0037, about -6.4 percent of the gap), and within that explained component marital status makes a positive contribution that widens the gap, while insurance coverage, urban residence, caste, and wealth make negative contributions that narrow it. In Maharashtra the gap is -0.069 (0.073 for men, 0.142 for women); urban residence and marital status widen it, while religion, caste, and insurance coverage narrow it. The logistic regressions underlying the decomposition show that age sharply raises reported morbidity, that insured individuals report more morbidity, and that ST/SC groups report less, which the paper interprets as reflecting reporting and access rather than disease prevalence alone.
Load-bearing premise
The analysis assumes that the pooled NFHS-5 sample of 724,115 women and 101,839 men can be treated as directly representative without survey weights or correction for the survey's stratified, clustered design, and that the two very unequal samples are comparable; if that assumption fails, the reported morbidity rates and decomposition shares could be biased.
Editorial extensions
If this is right
- If the decomposition is right, expanding health insurance coverage is a concrete policy lever that should narrow the gender gap in reported morbidity, since insurance has a negative contribution in both India and Maharashtra.
- Because marital status is the largest widening factor in India, policies aimed at never-married, widowed, and divorced women—who face different health-seeking constraints—could reduce part of the gender disparity.
- The divergence between national and Maharashtra results (urban residence widens the gap in Maharashtra but narrows it nationally) implies that state-level policy design is necessary rather than a single national template.
- The small explained share of the gap means that the measured socio-economic variables account for only a little of the gender difference; closing the gap will require attention to unmeasured factors such as health-seeking behaviour and reporting culture.
- The paper's own conclusion extends these findings to awareness campaigns: because insurance and treatment rates are linked, targeted illness-specific awareness and treatment subsidies could reduce underreporting and improve treatment, as the paper argues happened for tuberculosis.
Reading between the lines
- An extension the paper does not report would re-run the decomposition with the NFHS-5 survey weights and with both sexes restricted to ages 15-49; the male sample includes ages 50-54, so part of the age contribution is a pure sampling composition effect.
- The near-zero explained share suggests that the gender gap in self-reported morbidity may be driven less by measured socio-economic position and more by gendered differences in illness perception, health-seeking behaviour, or survey response; modelling treatment-seeking as the outcome would test this.
- Insurance's negative contribution is plausibly a selection effect rather than a health effect: insured individuals are more likely to be diagnosed and thus to report chronic conditions, so 'insurance narrows the gap' may mean differential diagnosis between men and women; using clinical measurement sub-samples, if available, would separate these mechanisms.
- Because the paper pools nine conditions and reports that diabetes dominates in India while hypertension dominates in Maharashtra, a natural next step is to decompose each condition separately; gender gaps and their socio-economic drivers may differ across diseases.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper uses NFHS-5 (2019-21) data to study socio-economic determinants of self-reported morbidity in India and Maharashtra, with a focus on gender disparities. The authors construct a binary morbidity indicator from chronic conditions and disabilities, estimate gender-stratified logit regressions, and apply a Fairlie decomposition to separate the male-female morbidity gap into explained and unexplained components. The paper reports that about one in nine Indians and one in eight Maharashtrians report morbidity, that women report nearly double the male rate, and that the decomposition identifies marital status as widening the gap and insurance, caste, urban residence, and wealth as narrowing it.
Significance. The topic is important: NFHS-5 is a recent, large, and policy-relevant data source, and gender disparities in self-reported morbidity are under-studied relative to mortality. The parallel India/Maharashtra analysis is a useful design, and the Fairlie decomposition is an appropriate tool for a binary outcome. The paper also carries out several diagnostics (VIF, link test, ROC curves) and provides a detailed morbidity classification. However, the central quantitative claim—that the decomposition identifies 'key drivers' of the gender gap—is not supported by the paper's own Table 4: the explained component is negligible in magnitude and, for India, has the opposite sign from the observed gap. The largest single decomposition contribution in both samples is an artifact of non-overlapping age eligibility for men and women. These issues, together with the absence of survey weights and design controls, make the headline conclusions unreliable in their current form.
major comments (3)
- [Decomposition Analysis, Table 4] The decomposition results contradict the paper's central claim that the analysis identifies key drivers of the gender gap. For India, the raw difference is Pr(Male)-Pr(Female) = -0.0576, while the total explained component is +0.00369, i.e., -6.4% of the gap. For Maharashtra, the total explained component is -0.001997, only 2.9% of the -0.0692 gap. Thus the included covariates explain essentially none of the female excess, and in the India specification they predict a male excess. The abstract and conclusion state that marital status, insurance, caste, urban residence, and wealth are 'key drivers,' but the sum of all these contributions is far smaller than the gap; the paper should either fundamentally reframe this conclusion or revisit the specification (for example, by allowing interactions or a richer set of covariates) before claiming that the decomposition identifies actionable drivers.
- [Decomposition Analysis, Table 5] The single largest contribution in both India and Maharashtra is the Age 50-54 category, with coefficients of 0.0147 and 0.0101, respectively. This category exists only for men because NFHS-5 samples women up to age 49 and men up to age 54, as the authors themselves note in the text. This component is therefore an artifact of sample eligibility, not a socio-economic driver of the gender gap, and it is larger than the total explained component in Maharashtra and about four times larger in India. The decomposition should be re-run on a common age range (e.g., 15-49) for both sexes, or age should be handled in a way that does not create a category present only for one group.
- [Data and Methodology] The analysis does not use survey weights or account for the complex stratified, clustered design of NFHS-5, despite the paper's own description of the data as collected through multiple stratified sampling phases. Unweighted analysis of a survey with vastly different sampling probabilities—here 724,115 women versus 101,839 men—can bias prevalence estimates, logit coefficients, and decomposition contributions. Since the paper makes population-level claims such as 'one in nine Indians' and reports national and state prevalence rates, the authors should either apply survey weights (e.g., Stata svy commands) or clearly justify and perform a robustness check demonstrating that unweighted estimates are not materially affected.
minor comments (6)
- [Title] The title contains a typo: 'inequility' should be 'inequality.'
- [Decomposition Analysis text] The text refers to 'Table 7' and 'Table 8' when describing the decomposition summary and detailed results, but the tables are numbered Table 4 and Table 5 in the manuscript; please correct the cross-references.
- [Table 4] The table reports 'Pr(Morbidity!=0|G=Male)' and 'Pr(Morbidity!=0|G=Female)' but the text says the decomposition is run 'grouping by Women.' Clarify the direction of the decomposition so that the reported difference is unambiguous.
- [Empirical Evidence, Maharashtra results] The sentence 'Widowed Men were more likely to report morbidity as compared to Widowed Men' is tautological; it presumably should compare widowed men to widowed women or to other male groups, and should be corrected.
- [Decomposition interpretation] After Table 5, the discussion of 'positive' and 'negative' contributions is confusing because the raw gap is negative; for example, age groups 35-39 and 40-44 have negative coefficients in Table 5 yet are described as having the highest positive contribution. Please define the sign convention explicitly with reference to the sign of the raw difference.
- [References] Some references are incomplete or inconsistent, such as 'Bango, M., & Ghosh, S. (2003)' while the text cites 2023, and the Oxfam entry lacks a clear report title; please verify and standardize all references.
Circularity Check
No significant circularity: the gender-gap decomposition is a direct application of Fairlie (2005) to independently operationalized NFHS-5 variables, with no self-citations and no quantity defined in terms of the claim it supports.
full rationale
The paper is a purely empirical application of standard statistical methods (gender-stratified logistic regression and Fairlie decomposition of the gender gap in self-reported morbidity) to public NFHS-5 microdata. The outcome (binary self-reported morbidity) and the covariates (age, residence, education, religion, caste, household size, wealth, insurance, marital status) are operationalized independently of the target results; there is no equation in which an output quantity is defined in terms of the claim it is meant to support. The explained component of the Fairlie decomposition equals the difference in mean predicted probabilities by construction, but the paper presents this only as the method's standard accounting identity applied to freshly fitted coefficients, not as an independent prediction. There are no self-citations by the present authors and no imported uniqueness theorems or ansatze; the only method citation is to Fairlie (2005), an external standard reference, and the model choice (logit for a binary outcome) is justified by the data structure rather than by an assertion of uniqueness. Concerns raised by the skeptical review - that the decomposition's explained share is tiny and even negative for India (-6.4 percent), that the single largest contribution (Age 50-54) is an artifact of men being sampled to age 54 while women are sampled only to age 49, and that the unweighted pooled sample ignores NFHS-5's complex design - are substantive correctness and interpretation risks, not circularity, and are outside the scope of this pass. The central claims about the gender gap in morbidity and its covariate decomposition are self-contained computations on public data, so the derivation chain contains no circular step.
Assumptions & free parameters
assumptions (3)
- domain assumption NFHS-5 data are analyzed without survey weights and still treated as nationally representative.
- domain assumption Self-reported morbidity, defined as a binary indicator from nine conditions, is a valid and comparable measure of health across gender and socio-economic groups.
- domain assumption The Fairlie decomposition implemented in Stata is correctly specified and identified.
Cite this review
Pith. "Pith review of Unmasking inequility: socio-economic determinants and gender disparities in Maharashtra and India's health outcomes -- Insights from NFHS-5." pith.science (2026). https://pith.science/paper/PIDYHJYG
@misc{pith2026250608206,
author = {Pith},
title = {Pith review of: Unmasking inequility: socio-economic determinants and gender disparities in Maharashtra and India's health outcomes -- Insights from NFHS-5},
year = {2026},
howpublished = {\url{https://pith.science/paper/PIDYHJYG}},
note = {Machine review of arXiv:2506.08206}
}
read the original abstract
This research examines the persistent challenge of health inequalities in India, departing from the conventional focus on aggregate improvements in mortality rates. While India has achieved progress in overall health indicators since independence, the distribution of health outcomes remains uneven, a fact starkly highlighted by the COVID-19 pandemic. This study investigates the socio-economic determinants of health disparities using the National Family and Health Survey (NFHS)-5 data from 2019-20, focusing on both national and state-level analyses, specifically for Maharashtra. Employing a health economics framework, the analysis delves into individual-level data, population shares, self-reported morbidity prevalence, and treatment patterns across diverse socio-economic groups. Regression analyses, stratified by gender, are conducted to quantify the impact of socio-economic factors on reported morbidity. Furthermore, a Fairlie decomposition, an extension of the Oaxaca decomposition, is utilised to dissect the gender gap in morbidity, assessing the extent to which observed differences are attributable to explanatory variables. The findings reveal a significant burden of self-reported morbidity, with approximately one in nine individuals in India and one in eight in Maharashtra reporting morbidity. Notably, women exhibit nearly double the morbidity rate compared to men. The decomposition analysis identifies key drivers of gender disparities. In India, marital status exacerbates these differences, while insurance coverage, caste, urban residence, and wealth mitigate them. In Maharashtra, urban residence and marital status widen the gap, whereas religion, caste, and insurance coverage narrow it. This research underscores the importance of targeted policy interventions to address the complex interplay of socio-economic factors driving health inequalities in India.
Reference graph
Works this paper leans on
-
[1]
Regression Analysis: Decomposition of the measures is performed using regression analysis of the morbidity variable which is binary ( 1 for if the individual has reported morbidity and 0 if the individual does not have any morbid condition or the same has not been reported.) against the explanatory variables to find their coefficients then calculate the m...
-
[2]
The independent variables are the same that have been considered for regression an alysis
Decomposition Analysis: We have performed decomposition for two groups based on gender, Group 1 representing Men and Group 2 representing Women. The independent variables are the same that have been considered for regression an alysis. The analysis was performed using the “Fairlie” code in STATA/MP 13.0. For non-linear models, they suggest the extension o...
work page 2005
-
[3]
mean predicted probability of outcome
− 1 𝑁2 ∑ 𝐹(𝛽1𝑋𝑖 2) 𝑁2 𝑖=1 ] + 𝑁1 𝑖=1 [ 1 𝑁2 ∑ 𝐹(𝛽1𝑋𝑖 2) 𝑁2 𝑖=1 − 1 𝑁2 ∑ 𝐹(𝛽2𝑋𝑖 2) 𝑁2 𝑖=1 ] Where N is the sample size, with two groups 1 and 2 and Y represents the difference in "mean predicted probability of outcome" between the two groups. The decomposition method provides data on ho w much and in which direction ( widening the differences or narrowing ...
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.