Hi, I'm Dr. Cameron Richardson, research scientist for the Clearinghouse for Military Family Readiness. My background is in developmental psychology. Before we begin, I'd like to thank my colleagues for their help in the development of this presentation. Let's begin.
In our previous podcast on the use of control groups, we talked about the fact that comparable groups allow researchers to make stronger conclusions about program effects, and we can make stronger conclusions because alternative explanations that exist as a result of group incomparability have been eliminated. Today we'll be talking about how one might generate comparable groups, because despite what you may think, comparability across groups doesn't just happen by accident. So let's dive in.
Our goal is to create comparable groups because they allow for strong conclusions about program effects. Well, we have a hurdle that we must overcome when trying to maximize our chance of generating comparability across groups, and that is that we need to make sure that all the variability in your participant pool is equally represented across your intervention and control groups. One solution to the problem of acquitting across groups is to use a random process for participant allocation to groups. I can't stress the following point enough: Randomization to groups does not try to get rid of the natural variability in your participant pool, but rather seeks to equate across your groups on this variability.
Let's look at an example to see if we can make sense of what randomization actually does for us. Let's say a program developer decides to market a special program of couples therapy for individual struggling with alcohol use, claiming that this program will reduce problem drinking. So in your implementation community you have a number of individuals that vary along a number of dimensions that you think relate to alcohol use, for instance, gender and impulsivity. For simplicity's sake let's take two levels of impulsivity: On the right, you have the ones who took a dare to paint their bodies yellow, the impulsive group, and on the left stand those who decided not to take the same dare, or the not impulsive group. Not that we don't have to know who is who; that is, participants don't have to be measured on all dimensions of interest, and in this case would be indistinguishable from each other.
As randomization will work whether variability within the sample is to find or not; but for our purposes let's say that we do know who is who and that everyone has chosen a ball with a number on it, from one to twelve in this case. We throw all the balls into a bin, shake the bin up real well before each draw, and when we pick one out blindly. There's no way to determine who will be picked, or you could say selection is random. What this means is that on every draw of a ball from the bin, each person left to be chosen has an equal chance of being selected. And that's the kicker: A random process is one of our most powerful means of generating group comparability, because it represents a means of selecting groups without bias. And it turns out the many software programs have powerful algorithms that generate random numbers making a lottery style selection process obsolete for most purposes.
An important caveat to the randomization phenomenon is that with larger samples, you are more likely to get comparable groups, just as it is more likely that with more and more random flips of a fairly weighted coin you expect to come out even on heads and tails. So with this caveat in mind you should take this small groups example with a grain of salt, used solely for simplicity's sake in the manipulation of this Powerpoint file. But to play this hypothetical out, with everyone having an equal chance of being selected at any round you'd expect three impulsive and three non impulsive individuals in each group as a result of the random process. That, is you wouldn't expect all six impulsive individuals to have ended up in one group.
So the main point? Well, it's that randomization serves to create comparability across groups with larger samples more likely to create this comparability which in turn allows for stronger conclusions about program effects because, as we said before, alternative explanations for these observed effects that derive from the lack of group comparability are eliminated; and if you're still not convinced that comparable control groups are necessary, then check out the control group's presentation or speak with us because I've run out of time. I want to thank you for stopping by.