Social protection in Kenya: designing a proxy means testing tool to achieve universal health coverage

Health subsidization for the poor in Kenya is one of the social protection priority programs for achieving universal health care. This study aimed to develop a proxy means testing (PMT) tool that can be used by the Ministry of Health in Kenya for targeting the poor in health insurance subsidization and to determine the ability of households in the informal sector to pay for social health insurance. 

To develop the PMT, least absolute shrinkage and selection operator (lasso) regression, which uses a machine learning algorithm, was used for both variable selection and prediction. The fitted models demonstrated good generalization since the R2 values were consistent between the train and test datasets. Easily observable household and individual characteristics were used to predict income for the informal households. 

For targeting purposes, Findings show that targeting the bottom 25th percentile would lead to the exclusion of only 2.7% from the targeting program with a leakage of 29.7% (inclusion error). Comparison with the previous PMT shows a good improvement. With a focus on the 40th percentile, the new PMT had inclusion errors of 28.7% and exclusion errors of 25.9%.