Outliers Assessed as Dependent Adverse Effects

Abstract

In a well-designed treatment trial the only difference between a treatment group and control group is the treatment. This is of course theoretically so. In practice many differences do exist, and raise the risk of biases.

Graphs like data plots and regression lines are convenient for visualizing outliers in therapeutic data patterns. Outlier data are considered as dependent adverse effects of the predictor data on the outcome data.

They are, however, arbitrary, and, with large data files, both data pattern and outlier recognition require a more sophisticated approach. Also, the number of outliers, generally, tends to rise with the sample size. BIRCH is the abbreviation of “balanced iterative reducing and clustering using hierarchies”, and is available in SPSS’s module Classify, under “two-step cluster analysis”.

The current chapter, using a simulated and a real data example, examines whether BIRCH clustering is able to detect previously unrecognized outlier data. Step by step analyses were performed for the convenience of investigators.