Analyzing Correlated Pairs Posted on January 15, 2008 Howard Abikoff (bio) explains how small data sets can still provide powerful evidence.

You don’t need a large N when you do that kind of work. In essence what you’ve generated are correlated pairs. Each child has a classmate that he’s linked or yoked with over time, and that classmate’s behavior serves as a standard or a reference point for typical behavior in that classroom. And when you have a hundred children in a study each of whom has a yoked classmate, what you have are correlated data. And the information that you can get from that is very informative, and the power is quite strong.

There are a variety of ways to analyze those data. A very simple way of course is to look at how well your measure differentiated your typical child from your study patient at baseline. If your measure is valid in terms of discriminative validity, you would hope that those groups, the patients and their paired classmates, are going to be different at baseline. In fact, if they’re not, you probably have to go back to the lab and rethink the measure you have.

In terms of the classroom methods, observation methods that we developed, the measure was very sensitive to differences between these youngsters. So you now have an index of the degree of difference between the study population and the typical children prior to treatment. And then one question is, what happens at the end of treatment? Does that difference remain, or in fact have the children now, who are in the study population, has their behavior changed so that they’ve moved closer to their classmates?

You can then do a very simple comparison of their scores at post to see whether or not the differences remain or in fact are they no longer present. And certainly in some of the studies that we’ve done, using medication as the treatment we were evaluating, we were able to show quite clearly that at least for some aspects of functioning, patients who were very different than their classmates prior to treatment were indistinguishable from them statistically at the end of treatment on some of the behaviors that were of interest.

And that’s a very powerful way of understanding what’s happening to these children in terms of their movement from the clinical range into the normative range.

« Back to Article