I have abundance data that was collected from video quadrats along a transect line at 3 treatment levels. I also have percent kelp cover for this dataset. My model is as follows:
Abundance ~ Treatment + KelpCover + 1|Transect
The data is overdispersed, and looks like this:
I have tried fitting negative binomial, gaussian, and log distributions with raw data, standardized data, and scaled data. I continue to get cone shaped residual vs fitted plots and terrible predicted vs actual plots (negative binomial used for this example). The p values and estimates for all the models are similar, just the model fit seems to be bad by visual inspection.
I am at a loss of what to try next. Can anyone suggest why this data doesn't seem to want to be modelled? And possible solutions?