A comparison of approaches for estimating the effect of women's education on the probability of using modern contraceptive methods in Malawi

•Education–contraceptive use relationship follows a non-linear increasing pattern.•Endogeneity of education is accounted for using a flexible two-stage approach.•The assumption of linear or pre-specified effects of continuous covariates is relaxed using penalized regression splines.•If endogeneity is not accounted for then biased estimates are expected.

The aim of this study is to estimate the effect of education on the probability of married Malawian women using modern contraceptive methods by accounting for both observed and unobserved confounders. We conduct a sensitivity analysis and compare the results of naive models with instrumental variable models to account for the potential endogeneity of education. Our findings demonstrate conflicting results between the two modelling approaches. The naive models report smaller education effects on the probability of using modern contraceptive methods compared to instrumental variable models. We also find that by relaxing the functional form assumption on the effect of continuous covariates, using a flexible instrumental variable model, the education's effect follows a positive, nonlinear pattern. This finding is not observed with a classic instrumental variable model.