Avoid categorization of a continuous variable

February 15, 2018

Selected literature

• Harrell FE. Regression Modeling Strategies: With Applications, to Linear Models, Logistic and Ordinal Regression, and Survival Analysis, 2nd ed. Heidelberg: Springer.
• Greenland. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology. 1995;6(4):356-65.
• Bennette C, Vickers A. Against quantiles: categorization of continuous variables in epidemiologic research, and its discontents. BMC Med Res Methodol. 2012;12:21.
• Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med. 2006;25(1):127-41.
• Schmidt CO, Ittermann T, Schulz A, Grabe HJ, Baumeister SE. Linear, nonlinear or categorical: how to treat complex associations? Splines and nonparametricapproaches. Int J Public Health. 2013;58(1):161-5.
• Schmidt CO, Ittermann T, Schulz A, Grabe HJ, Baumeister SE. Linear, nonlinear or categorical: how to treat complex associations in regression analyses? Polynomial transformations and fractional polynomials. Int J Public Health. 2013;58(1):157-60.

Conclusions

Avoid categorization of a continuous variable.

• No "biological" reason for steps in effects due to categorization
• Choice of number of categories and boundaries is subjective (arbitrary).
• Loss of statistical information
• Loss of statistical power
• Increasing number of parameters to estimate
• Test of trend over categories is not equal to test of non-linear effects on continuous scale (e.g. in a regression model)
• Measurement error is not reduced by categorization.
• Especially, dichotomizing does not allow modeling non-linear effects.