The literature on energy efficiency provides numerous examples of apparently profitable technologies that are not universally adopted. Yet according to the standard neoclassical theory of investment, profit-maximizing firms should undertake all investments with a positive net present value. The standard theory also holds that the discount rate for computing the present value of a project should be the return available on other projects in the same risk class, and therefore should not depend on characteristics of the firm. This model as applied to energy-saving investments is tested by examining whether firms'characteristics influence their decision to join the Environmental Protection Agency's voluntary Green Lights program. A discrete choice regression is estimated over a large sample of participating and nonparticipating firms. Missing values in the data matrix are replaced with multiple imputations from a distribution estimated using the expectation—maximization algorithm. The results show that (1) substantial improvements in the power of hypothesis tests can be achieved through maximum-likelihood imputation of missing data, and (2) contrary to the conventional theory, the characteristics of firms do affect their decision to join Green Lights and commit to a program of investments in lighting efficiency.