Eg., Looking at correlation between rate of vocalising and rate of aggression in five groups of apes. I want to control for the group size (which is obviously the same for every individual in the same group). I'm not interested in comparing the different groups as such.
Is this possible?
You have clustered data. The effect of vocalisation could perhaps be to influence the rate of aggression to vary around the basal level for the group. In this case, you need to use the group identifier as a stratum variable.
The current preferred analysis technique for this kind of clustered data is multi-level modeling (also known as hierarchical linear regression). It is designed to handle violations of the assumption of independent observations when dealing with multiple members of the same groups. I am not sure, however, how well it will work with only five groups.
Alternatively, you could do ordinary regression where your "stratum" is captured by a series of dummy variables for group membership.
Thank you both very much for your response. I'll have to do a bit of reading into these recommendations; much appreciated.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression