I am wanting to determine if I can use one disease as a proxy of another. My data are on sheep from various farms. I have first created a logistic regression model to test if there is an association between my proxy and my disease of interest (my data are positive negative values for each sheep). I then pooled my data at the farm level to determine if there was an association between my proxy and my disease of interest at the farm level (my data are now prevalence values for each farm).

I would like to determine at which level (animal or farm) my proxy best predicts my disease of interest, how would I do this? I can calculate McFadden's pseudo R2 for the logistic regression model and regular R2 for the linear model, but I am concerned that it is not appropriate to use R2 values to compare the goodness of fit of one model relative to another when the datsets are different. I have an odds ratio which tell me about the strength of the logistic relationship and a slope that tells me about the strength of the linear relationship, but both are on different scales.