Letters to the Editor
Categorizing Continuous Variables
Dr David Streiner impressively demonstrates that statistical power is lost if variables measured along a continuous scale are converted into categorical variables (1). He observes that the categorization of a continuous variable is justified only if the data are markedly skewed or if the variable shows a nonlinear relation with another variable.
I suggest that there is one more situation in which continuous variables are better categorized: when the values obtained are “guesstimates.” In India, for example, many patients who belong to lower socioeconomic strata, and some rural patients, do not know their exact age. They provide approximations that are usually a multiple of 5, and thus, a data set might contain a large number of patients aged 30, 35, or 40 years. When using such data, it may make more sense to analyze age as a categorical variable. Expressed otherwise, it might be a good idea to categorize continuous variables when there is reason to believe that the variable cannot be, or has not been, accurately measured.
1. Streiner D. Breaking up is hard to do: the heartbreak of dichotomizing continuous data. Can J Psychiatry 2002;47:262–6.
Chittaranjan Adrade, MD