Also, fairly sensibly, known as “restriction of range”. Covers what can happen when the dataset you have may not cover the full range of the values to which you might want your findings to generalise. As often the case if using problem self-rating data from student volunteers and wanting to generalise findings to clients seeking help: the range of scores from the student volunteers may be lower than would be seen in a help-seeking dataset.
Details #
Here’s a linear correlation across 500 pairs of scores (say on two different measures of problems).
That shows a nice linear regression of the second variable score on the first with slope .69. Here is are the data for those scoring below -1.5, i.e. below that vertical line in the above plot.
The restricted range of the V1 scores has markedly reduced the observed slope. This is classical range restriction. We should be wary of it when generalising between datasets to the situations where we want to apply our findings.
Try also #
Chapters #
Not covered in the OMbook.
Online resources #
None forseeable.
Dates #
First created 31.x.24.
