Heteroskedasticity in bird point counts
Bird (Dabbling Ducks) count data on 30 wetlands for 2 years for about 10 weeks.
Starting with a mixed effect Poisson regression model. Then moved to a mixed effect linear model using square root of the count as the response and week as a weight (due to heteroskedasticity by week).
- Week (numeric or categorical)
- Age of wetland
- WC ?
- site x year
Heteroskedasticity appears to exist with more variability in early weeks compared to later.
- At least for model building, use linear model with square root of count
- Treat week as categorical
- Consider repeated measures structure on random effects
- Try the analysis in SAS so that you have easier control of random effects and error variances
blog comments powered by Disqus