This paper demonstrates a formal statistical test that can be used to help researchers make decisions about alternative statistical model specifications. This test is commonly used by researchers who would like to test whether adding new variables to a model improves the model fit. However, we demonstrate that this formal test can also be employed when alternative representations of variables or constructs are considered for inclusion in a regression. An empirical example is provided using information from a widely cited US Department of Education report. Substantively, we find evidence that an alternative representation of an important policy variable would have been a better fit to the data than the index that was used in the analysis conducted in the report.