Cross-over with sequence effect
Looking at the relationship between the number of bees present and the amount of vegetation at a location.
Data
Bees
- 3 years of data
- 11 different sites (unbalanced across years)
- 5 survey periods per year
- measure total number of bees
- 5-6 different trap types (currently aggregating over trap types)
- currently looking at bee abundance
Vegetation
In each site, there are
- 10 quadrats
- measure coverage in each quadrat
- currently averaging across the 10 quadrats
Question
What is the relationship between coverage and bee abundance?
Model
Poisson regression with fixed effects for
- average coverage
- survey period
- year
- coverage*period
- coverage*year
- period*year
- coverageperiodyear
random effects for
- site
- site*year
- siteyearperiod (observation specific)
Other models
Also considered normal regression models based on log and sqrt transforms.
Negative-binomials failed to converge.
Results
Inidcation of overdispersion in a model without the observation-specific random effect.
Zero estimate for variance for year*site interaction in the model with the observation-specific random effect.
Question
How to deal with overdispersion?
Advice
Pull all coverage interactions.
Model
Fixed effects for
- coverage
- use year-period combination
Random effects for
- site
- site*year
Possibly add an observation-specific random effect in a Poisson model to deal with overdispersion.