We should note some possible concerns with this model. First, there could be different sets of technologies available at different times, which would complicate the selection process. As it happens, all of the technologies considered here were in use by 1960, which is when our dataset on technology choice begins.3 Second, the profitability being compared in equation (1) is in principle the expected profitability over the plant’s lifetime, so expectations about future R and X values at the plant enter the equation. There is a high degree of persistence in cross-state regulatory differences, so current regulatory values should capture most of the available information about future regulation.4