Test Bias

1
Test Bias

• Reference Anastasi, A. (1988). Psychological
  testing, 6th edition. New York Macmillan
  Publishing Co.

2
Looking at population differences in the
effectiveness of a test

• Since the 1960s there has been a lot of research
  on the possible ethnic or cultural differences in
  the predictive meaning of test scores
• Bias is often (mistakenly) thought to be present
  when there are differences in mean scores between
  minority and majority groups. This might lead to
  differences in the proportions of individuals who
  are classified in a certain way

3

• However, these different means are only an
  indication of bias if you can assume that the 2
  groups should have the same distributions
• In identifying test bias, you need to look at
  validity. Is the test equally valid for both
  groups? In this case we are talking about
  predictive validity.

4

• A prediction line is used to make individual
  predictions. Two parameters describe each line
  the y-intercept (a) and the slope (b)
• A test that is NOT biased should have the same
  regression line for minority and majority groups
• In this picture, both groups have same slope and
  same interceptsame regression line. There is no
  bias.

5
Intercept Bias

• Both groups have the same slope, but different
  intercepts.
• This is called intercept bias
• You would need 2 regression equations
• The criterion score (Y) is different depending
  upon group measurement
• Using a combined regression line (red), the Y
  score would be overpredicted for the blue group,
  and underpredicted for the green group

6
Slope Bias

• Here the groups have different slopes but the
  same intercept.
• This is an example of slope bias
• Slope bias is most closely related to the
  validity coefficient
• Predictability in the 2nd group (red) is very poor

7
Slope and Intercept Bias

• These groups have different slopes and different
  intercepts
• Both types of bias are present