We report impacts of a randomized housing quality improvement intervention among Indian migrant workers. Despite modest improvements in conditions, respondents experienced a decline in satisfaction and a large increase in psychological distress as a result of treatment. In contrast, residents who faced the same treatment-induced variation in living conditions as the original sample, but who arrived after treatment had already been initiated, had increased satisfaction. Impacts on turnover echo these patterns. We interpret this as evidence of reference dependence: residents who were primed to expect larger-than-realized improvements in living conditions suffered utility losses, while exposed but unprimed residents experienced gains.