Is a new evaluation always necessary

Abstract Although generalization of evaluation findings from one instance to another is likely to be a matter of degree, the standard rules of statistical inference argue against virtually all generalization. Program providers also frequently reject the notion that evaluated findings of similar programs might apply to their own. An approach to forecasting generalization was applied to the evaluations of home energy audits in two Illinois samples. Generalizability of findings for audit impacts across samples was identified. It is proposed that standard evaluation procedure include assessments of generalizability of findings across the range of contexts, subsamples, and program implementation variations already existing within a study. Such findings will facilitate efforts to forecast generalizability by program providers and will advance theoretical understanding of the processes underlying program effects.