Simwinga Lukundo Willy2026-06-202026-6-1210.47191/jefms/v9-i6-16https://pubs.cidrz.org/handle/123456789/12949<jats:p>The standard methodological approach used in discrimination research has traditionally been an Ordinary Least Squares (OLS) based decomposition. In the last decade, however, machine learning models have become popular as alternative tools for analysing discrimination, due to their ability to provide robust estimates in high-dimensional settings, even when the functional form of the model is unknown. This paper adds to the Zambian labour economics literature by examining the gender pay gap (GPG) between paid employees and the self-employed (2020-2022). By estimating and comparing GPG estimates obtained from the classic simple OLS decomposition method to those obtained using the doubly robust Double/Debiased Machine Learning (DDML) approach. The findings of the study suggest that the estimates of the GPG derived from the DDML are nearly 10% for paid employees and 34% for the self-employed. This evidence confirms that discrimination seems to be more persistent among the self-employed than salaried employees. This paper generally discusses the advantages and disadvantages of old and new methods to study pay inequality. It provides an in-depth analysis of the GPG in Zambia and makes important new contributions to the literature. Complex econometric methods such as DDML are more appropriate in explaining the nature of wage gaps.</jats:p>https://doi.org/10.47191/jefms/v9-i6-16