Proper tests to examine parameter/treatment effects?

Question

I am running simulation models on 100 plots randomly selected from a real study area. I am testing the effect of different modeled treatments that are applied uniformly across the study area. The data are not normally distributed (even if log transformed).

(a) What statistical test is most appropriate to determine whether a single treatment (e.g., doubling the number of organisms in a plot) has an effect on a dependent variable (e.g., the number of potential habitat patches occupied). The plots are independent but the dependent variable is measured before and after a parameter change. Are these paired samples or not?

(b) What statistical test is most appropriate to determine whether multiple treatments (e.g., changing the number of organisms in a plot, increasing the connectivity potential, etc.) has an effect on a dependent variable (e.g., the number of potential habitat patches occupied).

Thanks!

Answer 1

I think you want to use a moderated regression analysis here to answer both questions at once.

Why not use separate tests?

You want to know if your independent variable have an influence on your dependent variable. If you conduct one analysis for each potential predictor you might find significant effects of independent variables that explain similar variance in your dependent variable.

Why use a moderated regression analysis?

Using a moderated regression analysis you will be able to identify independent variables as valid predictors for your dataset while being able to evaluate (1) how large specific effects are (2) how much variance can be explained by your predictors and (3) check how the variance is explained.

Regarding your second question, you can calculate all possible interactions between your independent variables. For instance, treatment A might be stronger or even only successful when combined with treatment B.

In other words, a moderated regression analysis will enable you to investigate direct effects (e.g. "treatment A does the job all alone"), indirect effects (e.g. "treatment A is only successful when combined with treatment B") as well as moderated effects (e.g. "treatment A and treatment B both help when applied alone but lead to much better results when combined").

At last you could think about using a stepAIC algorithm to automatically identify the best fitting model for your available variables.