Blocking and MatchingPosted on October 19, 2007
Helena Kraemer (
bio)
discusses reasons for taking a non-random sample.http://www.4researchers.org/articles/85 (1:58)
If you go out and you simply randomly take a sample, whatever randomly means in that context, then it’s a naturalistic sample, meaning that the characteristics of the sample should be sort of a miniature representation of what’s going on in the population. On the other hand, if you decide beforehand that I’m going to take half women and half men and half of the women are going to be minorities and half of the men, minorities, what you’ve done is you’ve blocked out the selection procedure. And the reason you would do that is because you are particularly interested in that particular case in gender and ethnicity effects on the outcome. Otherwise you just do a representative sample.
The extreme of that is where you decide that you basically are going to enter people in matched pairs, for example. That you are going to select pairs of subjects who are matched in ethnicity, gender, age, location, maybe even clinician and so on and you are going to randomly assign one of the pair to the treatment and the other to the control. The reason for that is you are really trying to control all those variables that you are matching on as closely as possible. There are situations where that increases power and there is situations where that decreases power. So it’s a hard decision to make beforehand but the one thing that you can be sure of is that it makes it harder to do the study. So you have to be sure that it’s going to increase your power, give you advantages before you decide to do it.
To give you a very pertinent example. One of the things we know in the weight loss studies is that the more the person weighs in the beginning, the more weight they are going to lose during the weight loss study. And so if you don’t control in a very serious way on initial weight, you are going to lose a lot of power. And in that case, what you might do is match them, bring them into the study in matched pairs where they’re matched in terms of their BMI at baseline, randomly assigned one to treatment and one to control.