Best method of analysis for negatively skewed longitudinal environmental data?

Previous Topic Next Topic
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

Best method of analysis for negatively skewed longitudinal environmental data?

I have a dataset composed of a dependent variable (species percent cover) and a range of abiotic variables (salinity, temperature, pH, water movement etc). It is a longitudinal study, in which species percent cover was measured once a month for five months. The abiotic variables were measured using data-loggers every five minutes for the entire duration of the study. I have organised those data as monthly means in order to compare them with the species percent cover values. I have 13 study sites, at each of which I used three settlement collectors (each of the collectors is composed of three settlement plates). The goal is to determine the extent of influence of the abiotic variables on the species growth.

I was initially planning on carrying out partial least square regression, but I have been unable to account for the repeated measures aspect of the study. I have very limited R experience and have been unable to write script to carry out an appropriate analysis with PLS, so have been primarily been using SPSS and JMP.

I am now trying to use Linear Mixed Models. The issue here is that the data are heavily negatively skewed, given the large number of zero percent cover points, especially as every single site I worked at started as zero percent cover. There is also a smaller peak on the positive side where there are a reasonable number of 100% values. In SPSS -> Analyze -> Mixed Models -> Linear I use the settlement collectors as the Subject and Months as the Repeated Measure. Then I use % cover as the Dependent Variable, Site as the Fixed Factor and the abiotic variables of interest as covariates. Fixed effects of abiotic variables and a random effect of the intercept and the subjects. I am not sure if this is appropriate given the non-normality of those data.

Any suggestions of alternative methods or enhancements of the ones I have mentioned would be greatly appreciated.