Regress to Reveal

Abstract

Assumptions for the validity of standard regression tests are often not met. The information contained in the test statistics, however, can often be transformed to give easily grasped, intuitively clear descriptive statistics whose usefulness is in no way dependent on those assumptions. Thus, valid marginal explanatory values (mexvals)can replace invalid t-statistics; valid loss limits can replace invalid standard errors. Regression can be usefully and honestly used to reveal relations in data instead of deceptively claiming to test hypotheses. Regress to reveal, not to test!