I am using GAMs in mgcv to look at relative stem densities and weights for a paired dataset (control and treatment) stem data for each sampling location. Should my response variable be, for example, the stem weight ratio of treatment/control or instead, would it be more appropriate to have the treatment stem weight as the response and include the controls stem weight as a co-variate?
What about using the difference between treatment and control as the response variable, since the associated data at each sampling location is the same for both treatment and control? Cheers
Hi John,
It's usually best to model the raw data, which in this case is stem weight with control/treatment as a covariate. It can also be easier to choose what distributional assumption you want to make about your error structure of the raw data. For example the error structure of weight will likely be best described as a log-normal distribution. On the other hand, what is the distribution of the error structure of a derived ratio between two log-normal distributed weights......not quite as clear.
Wish you well,
Dan
Thank you both for your input. I'm currently working through the dataset using control/treatment as a covariate and also exploring a 2nd approach of using the proportion (ie dock/dock+control) and using a beta distribution or logit transformation.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology