A psychologist is interested in attitudes towards the disabled. She believes that contact with someone who is disabled might have an effect on peoples’ attitudes. To test her hypothesis, she measured attitudes toward the disabled both before and after contact with an individual in a wheelchair. What type of statistical test should she use to determine if contact with a disabled person changes peoples’ attitudes toward the disabled?
Pearson’s correlation (r)
Correlation is used to describe to what degree pairs of data are related. In this case, correlation could be used to describe the relationship between before and after attitudes, but not to test whether the means are different.
Regression analysis can be used to derive an equation from which we can predict scores on one variable based on scores on one or more other variables. In this example we have measures of the attitudes before and after an intervention for each individual. Thus, we have a pair of scores for each subject, and we are interested in the difference between the means of these scores.
t-test for independent samples
A t test for independent means is used to compare means derived from unrelated samples. We do not have two independent groups; instead our data are measured before and after contact with a disabled person, indicating that our data are paired.
t-test for paired data
A paired samples t-test is used to compare data which consists of scores that are paired or matched, such as before/after. In this case we have two measures that are related because the same people were measured concerning their attitudes both before and after exposure to a disabled person. Because the measures are paired, a paired samples t- test is appropriate in this situation.
Need some help? Ask the Expert!
The Expert says…
The null hypothesis is that the mean attitude before the intervention is the same as the mean attitude after the intervention.
There are two observations from each person – a before measure and an after measure. Thus, the data are pairs of observations from each person. The t-test for paired data is designed just for this situation.
Of course, you need to check that assumptions of the test have been met. In particular, you assume that the distribution of difference scores (after scores minus before scores) are normally distributed. The before and after scores themselves do not need to be normally distributed. It is helpful to include a measure of precision, such as a confidence interval, when describing the findings.