Association for Public Policy Analysis & Management

Technology subsidy policies like the California Solar Initiative (CSI) redistribute public funds to accelerate the adoption of technologies with perceived social benefits. Previous work has focused on analyzing the accumulation of these benefits as a policy outcome, such as the degree to which societal benefits offset the costs of implementing the rebate program and the degree to which end-users capture the financial benefit of the redistributed funds. In the context of CSI, the largest state-level solar photovoltaic (PV) rebate program in the U.S., these outcome-oriented analyses are tied specifically to the CSI rebate program through two factors: 1) a particular technological, geographic, and temporal context, and 2) the CSI program’s output – accelerated adoption of solar PV technology. However, because context and output are so closely intertwined these outcome measures are not ideal for comparing across different modeling strategies.

In this paper we directly estimate the effect of the CSI rebate program to accelerate the adoption of residential solar PV, while accounting for contextual differences across the program’s implementation. The analysis uses a regression discontinuity (RD) design that leverages exogenous decreases in rebate levels to estimate the casual relationship between rebates and the empirical growth rate of residential solar PV adoption. Estimating this causal impact, “purified” of the indirect effects of context, establishes a benchmark of program effectiveness that can be used in cross-model comparisons, for example between an RD and an agent-based model. Furthermore, we focus on the proximal output of the rebate program rather than on more distal policy outcomes to enable comparison across a broader range of alternative models, for example models that may ultimately aim to estimate impacts in terms of private, social, or environmental benefits. This work thereby paves the way for an apples-to-apples comparison of top-down equation-based modeling to bottom-up individual decision-making based modeling approaches.