Correcting contraceptive failure rates for sample composition and sample selection bias

In this paper we address two questions about life table estimates of the proportion of women experiencing a contraceptive failure within one year of initiation of use: Is the observed proportion affected by the socioeconomic or demographic composition of the sample? Is there evidence, even when socioeconomic or demographic influences are controlled, that the propensity to choose a particular method is correlated with the woman's underlying propensity to fail? We found that the answer to the first question is yes; hence observed method‐specific failure rates are affected by sample composition bias. We present new failure rates that correct the bias. We could, however, find no evidence of sample selection bias.