Failure of nonlinear models of drinking history variables to predict neuropsychological performance in alcoholics

To test whether the relationship between drinking parameters and cognitive functioning can be visualized as a curvilinear surface (rather than as a line, which more traditional regression equations presuppose), the authors tested 84 recently detoxified and 72 longer-term abstinent alcoholic men with the Halstead-Reitan battery. Quadratic statistical models did not predict test performance in the first group; a few predictions were found in the second. When the multiple correlation was adjusted for number of cases and variables, the "nonlinear" results were statistically no more significant than those of simpler linear models. It is concluded that drinking history does not readily explain cognitive findings among sober alcoholics and that polynomial models can produce inflated correlation coefficients.