Regression-Discontinuity design

Explanations > Social Research > Design > Regression-Discontinuity design

Description | Example | Discussion | See also

Description

The regression-discontinuity design uses a cut-off selection method, for example where subjects are selected based on scoring above or below a certain value on a previous test.

In design notation, this is:

C	O	X	O
C	O		O

The key aspect about this design is that the control group is made up of those who fall at the other side of the cut-off score. The sample is thus cut in two, with one group as control and the rest as the treatment group.

Example

A training module is designed to increase the visual-spatial ability of people with lower 'IQ'. A sample is selected and tested for IQ, with those scoring below 100 being allocated to the cut-off group.

A test for visual-spatial ability was then administered before the lower-IQ group was given the training. A post-test score showed that their ability had increased in this area.

Visual-spatial test, average score	Pre-test score	Post-test score
Those with IQ below 100 (the treatment group)	25	32
Those with IQ above 100 (the control group)	46	47

Discussion

This is a relatively unusual design that is very useful in specific situations.

In practice, if the cut-off selects only a small number, then the control group may also be constrained. In the same way, more subjects may be sought for the control group if the treatment group takes a significant majority.

The name 'regression discontinuity' arises because the assignment by score causes a discontinuity across this score boundary between pre-test and post-test and between treatment and control groups, as in the diagram below.