MATCHED SAMPLES

Suppose you are looking for improved performance in a sample of children between ages 6 and 10 following an intervention aimed at teaching them how to solve a mathematics problem. You only have access to a relatively small number of children, say less than 60, and you know that at least one third will not want to participate in your study, leaving you with 40.

In order to control for improvements that may have happened anyway, without the intervention, you could randomly assign 20 children to the intervention and 20 to a control group that does not receive the intervention - RANDOMISATION. But with small samples, this might mean you accidentally create different samples, such as most of the nine and ten-year-olds ending up in the intervention group and most of the six to eight-year-olds in the control group. Obviously the better performance of the intervention group might be because the children in that group were older, an example of a CONFOUNDING VARIABLE.

Provided you can recruit equal numbers of children of the different ages, you can then take pairs of age-matched children and randomly assign them to the two groups. This will ensure you have matched groups with respect to the age-mix in each group, controlling for any possible age-effects on performance in the two groups.

For a longer discussion see:

Robson, C. (1995) Real World Research. Oxford, Blackwell, p. 93

To return to previous page, use "Back" button on your browser

Back to Which Test Home Page