There is a strong and growing interest in applying formal methods for causal inference with observational data in social epidemiology. A number of challenges in defining, identifying, and estimating counterfactual-based causal effects have been especially problematic in social epidemiology, particularly for commonly used exposures such as race, education, occupation, or socioeconomic position. The purpose of this article is to revisit these challenges in light of the conceptual and analytic advancements in causal inference over the last two decades. We focus on a central assumption for causal inference known as the stable unit treatment value assumption, which can be divided into two component assumptions: counterfactual consistency and the absence of interference. We give simple hypothetical examples to illustrate how and why these assumptions are often violated in research on the social determinants of health (e.g., education, race/ethnicity, socioeconomic position) and provide strategies that can be used to sidestep these assumptions. In particular, we note that a recently proposed mediation analysis strategy can be used to explore questions about health disparities in a more formal causal inference framework. We emphasize that a central obstacle to estimating causal effects variables such as race, education (e.g., high school versus no high school), or occupation is the need to identify an intervention (possibly hypothetical) that will lead to changes in the exposure of interest.