We focus on two utility programs intended to reduce energy usage and the associated CO2 emissions—a home energy audit and rebates on the purchase of high-efficiency air-source heat pumps. We use a unique panel dataset from participating and non-participating households to estimate the average treatment effect of participating in either program on electricity usage. We fit models with household-by-season, season-by-year, and household-by-year fixed effects to account for all possible confounders that might be influence energy usage. Since the programs are voluntary, we seek to restore near-exogeneity of the program “treatment” by matching participating households with control households. We deploy coarsened exact matching (CEM; Iacus et al., 2011) as our main matching method. We ask whether it is sufficient to match households based on past electricity usage, or if we gain by adding structural characteristics of the home, including heating system type. We find that the two programs reduce electricity usage by 5% on average. The effects are strong in both winter and summer for the energy audit group but appear to be stronger in the winter for the heat pump rebate group. Adding house characteristics to the matching variables does seem to affect results, suggesting that using past usage alone may not be sufficient to identify the effects of program participation.