|
Abstract:
|
Dichotomization is the transformation of a continuous outcome to a binary outcome. This approach is not harmless from the viewpoint of statistical estimation and hypothesis testing. We show that this leads to a loss of information, which can be large. For normally distributed data, this loss in terms of Fisher's information is at least 1--2/pi. In other words, 100 continuous observations are statistically equivalent to 157 dichotomized observations. The amount of information lost depends greatly on the a priori choice of cut points, with the optimal cut point depending upon the unknown parameters. The loss of information leads to a loss of power. Only in certain cases, when estimating a value of the cumulative distribution function and when a used model is very different from the true model, the use of dichotomized outcomes can be considered as a reasonable approach.
|
