We analyze the impact of product labels indicative of ethical sourcing on consumer demand for coffee beans. The paper complements an existing field experiment by re-analyzing the empirical setting under weak assumptions with a more granular level of study which is subject to selection bias. In contrast to standard assumptions on consumer behavior in discrete choice models, this paper takes a fully nonparametric difference-in-differences approach to flexibly capture heterogeneous treatment effects and substitution patterns using spillovers. We also propose plausible structural bounds for partial identification as treatments can switch off in a multi-period setting. We derive semiparametric estimators based on the moment conditions to avoid the negative weighting problems associated with standard fixed effects models under heterogeneity. However by including certain covariates, we show that average effects may be estimated by a simple linear model on specific subsets of the data that achieve balanced propensity scores. Overall we find similar results as existing work, and in particular, the average consumer is sensitive to prices and unwilling to pay a price premium for ethically sourced coffees.