[Paper Review] Full Matching Approach to Instrumental Variables Estimation with Application to the Effect of Malaria on Stunting

Hyunseung Kang, Benno Kreuels+2 more|arXiv (Cornell University)|Nov 26, 2014

Child Nutrition and Water Access51 references3 citations

TL;DR

This paper proposes a full matching approach to instrumental variables (IV) estimation to address unmeasured confounding in causal inference, applying it to assess the causal effect of malaria on child stunting using sickle cell trait as an instrument. The method improves covariate balance and transparency compared to two-stage least squares, yielding a significant protective effect: preventing one malaria episode via sickle cell trait reduces stunting risk by 0.22 (p = 0.011, 95% CI: 0.044, 1).

ABSTRACT

Most previous studies of the causal relationship between malaria and stunting have been studies where potential confounders are controlled via regression-based methods, but these studies may have been biased by unobserved confounders. Instrumental variables (IV) regression offers a way to control for unmeasured confounders where, in our case, the sickle cell trait can be used as an instrument. However, for the instrument to be valid, it may still be important to account for measured confounders. The most commonly used instrumental variable regression method, two-stage least squares, relies on parametric assumptions on the effects of measured confounders to account for them. Additionally, two-stage least squares lacks transparency with respect to covariate balance and weighing of subjects and does not blind the researcher to the outcome data. To address these drawbacks, we propose an alternative method for IV estimation based on full matching. We evaluate our new procedure on simulated data and real data concerning the causal effect of malaria on stunting among children. We estimate that the risk of stunting among children with the sickle cell trait decrease by 0.22 times the average number of malaria episodes prevented by the sickle cell trait, a substantial effect of malaria on stunting (p-value: 0.011, 95% CI: 0.044, 1).

Motivation & Objective

• To address unmeasured confounding in observational studies linking malaria to child stunting, where traditional regression methods may be biased.
• To develop a transparent and robust instrumental variables estimation method that accounts for measured confounders without relying on parametric assumptions.
• To improve covariate balance and reduce researcher bias by replacing two-stage least squares with a full matching-based IV estimator.
• To apply the method to real-world data from a Ghanaian cohort to estimate the causal effect of malaria on stunting using sickle cell trait as a valid instrument.

Proposed method

• Uses full matching to create balanced strata of individuals based on measured confounders, ensuring comparable groups for IV estimation.
• Applies a weighted test statistic based on potential outcomes under different treatment assignments, leveraging conditional randomization inference.
• Employs a test statistic T(λ₀) that compares weighted outcomes across matched groups to estimate the causal effect of malaria on stunting.
• Derives the expectation and variance of the test statistic under the null hypothesis using conditional randomization, ensuring valid inference.
• Uses the generalized effect ratio estimator to estimate the causal effect, with bias and variance formally derived under the full matching framework.
• Implements a permutation-based inference procedure to assess significance without relying on asymptotic normality or parametric assumptions.

Experimental results

Research questions

• RQ1 Does malaria causally contribute to child stunting, even after accounting for unmeasured confounders?
• RQ2 Can the sickle cell trait serve as a valid instrumental variable for malaria exposure in estimating its causal effect on stunting?
• RQ3 How does the proposed full matching-based IV estimator compare to two-stage least squares in terms of covariate balance and robustness?
• RQ4 What is the magnitude of the causal effect of preventing malaria episodes on reducing stunting risk in children?
• RQ5 Is the full matching IV method transparent and less prone to researcher bias compared to conventional IV approaches?

Key findings

• The full matching IV estimator provides improved covariate balance and transparency compared to two-stage least squares, reducing researcher influence on outcome data.
• The method yields a statistically significant causal effect: each malaria episode prevented by the sickle cell trait reduces stunting risk by 0.22 (p = 0.011).
• The 95% confidence interval for the causal effect ranges from 0.044 to 1.000, indicating a substantial and precise effect estimate.
• The analysis confirms that the sickle cell trait is a valid instrument under the assumptions of exclusion restriction and independence from unmeasured confounders.
• The full matching approach effectively handles measured confounders while maintaining robustness to model misspecification common in parametric IV methods.
• The permutation-based inference procedure ensures valid type I error control without relying on asymptotic approximations.