Learn About
- Funding
- Research Design
- Participants
- Study Management
- Collaboration
- Dissemination
- Career Advancement
Community as Unit of AnalysisPosted on March 19, 2007 Robert Saltz (bio) talks about issues to consider when your sample size is at the community level. |
So the big challenge of doing community level research is that you have to have enough communities. In the old days, you could get by with doing one community and comparing it to a non-treatment community, but now usually the research is designed so you have to have at least a dozens, maybe two dozen, depending on the research requirements. And that makes an enormous difference in the difficulty of getting something to happen at one of say six, eight, ten of the treatment communities.
The reason you have to have more than just a couple, a handful, of communities is that in order to show a change or the effect, the impact, of your intervention at a community level, you need to have enough communities; a sample size is of communities rather than individuals in the community.
So you’re barely getting by in most cases with only having say 12 communities, half of which are gonna get an intervention. You imagine a study in which you’re gonna run people through an experiment and you only had 12 of them. That would seem pretty low and the same is true of doing community level research. The community, in other words, is the unit of analysis.
Changing of a unit analysis from the individual to the community is not simply the case that you have to add all these extra communities but it runs contrary to a lot of our training about sample sizes generally because now, although we have to take on these extra communities, the sample size within any one of them is actually lower than what most people think.
So you get a little bit of a tradeoff there so it doesn’t help you much if you have, in the study I’m working with now with 14 college campuses, it doesn’t help you a lot to get a thousand students in your sample from each campus. If you look at power analyses, you can get by with maybe 50 to a hundred students per campus, if indeed you’re using an experimental design in which seven of the intervention campuses are all getting the same kind of intervention.