A large literature in labor economics seeks to decompose observed gender wage gaps (GWGs) into different sources, including portions explained by cross-gender differences in education, occupation, and experience. This paper provides new methods for decomposing GWGs and applies them to data from the Panel Study of Income Dynamics (PSID). We build upon a recent technique that develops a “foundation model” of career transitions estimated using a large dataset of resumes, creating low-dimensional representations of worker histories. First, we develop a method for fine-tuning the foundation model to predict wages while ensuring that the representations do not omit features of history whose exclusion would bias decompositions. Second, when predicting wages in the PSID, we show that this method better predicts wages relative to regression-based approaches that use hand-crafted summary statistics of worker history. Third, we study GWG decompositions in the PSID, showing that relative to prior approaches, our method for including worker history explains a larger share of the GWG. Fourth, we introduce a novel decomposition of the change in GWGs between two periods, one early in careers and one later, for workers in the workforce in both periods. Fifth, we apply this decomposition, decomposing changes in GWGs in the PSID over 12-year intervals into two sources. The first source is predictable changes in GWGs associated with gender gaps in initial characteristics; we estimate that this source closes the GWG over time. The second source is differences in worker transitions; we estimate that this increases the GWG over time. We show that when analyzing changes in the GWG over 12-year intervals that begin early in workers’ careers, the second effect dominates, and differences in worker transitions lead the GWG to widen over time; in contrast, for intervals that begin later in worker careers, the first effect dominates, and the GWG narrows over time.
