Attrition Patterns in Your Study Design Posted on March 1, 2006 John B. Reid (bio) suggests ways to address attrition patterns in school-based studies.

Unfortunately, if you look at the mobility, if you talk to some demographers and people like that, what you find is that 1st graders from the beginning of the year, you have a lot of them won't be around at the end of the year, and a lot of them at the end of the year weren't there at the beginning of the year. We've had schools that have over 100% turnover in a year.

Now think about that for your little design, prevention study. By the time I do my end of the intervention assessment, half of them are gone that went through the intervention, and they're replaced by people who came in and didn't have the intervention. You really, really want to think that through in terms of how you're going to handle that.

You know if I was to design a universal intervention and do this thing again, I'd do it differently. What I did was schools as the unit of randomization. What I would do in a place like Eugene or any place is I'd go and I'd look at, I'd be talking to the school district about their migration patterns. These people are not going randomly. You're not taking a person in the poorest school district and moving them to the rich school district. That's not what's going to be happening. There are some understandable patterns. So if you have a rigorous and a systematic implementation or intervention, you could have it plus your schools. So the same snake oil got given every week, so if a kid were to move it's not going to be as nice but at least he will get the same stuff.

But that's the kind of thing where it's more than just finding out how many poor people are there or what the ethnic diversity is. You need to understand what some of these processes are that may not be theoretically really neat, but they're going to raise hell with your design. You really want to give your intervention a fair chance of succeeding. I have a hunch that a lot of intervention studies that show a minimal or kind of crappy effect sizes and outcomes, probably part of the reason is that the intervention is... just logistical problems.

« Back to Article