when fitting a nlme on nested repeated measurements data , models can be unstable if sample sizes across groups are highly unbalanced. In my dataset I have 4 unbalanced groups in terms of size.
Should I consider a random sampling of my biggest group of individuals?
No; use all the data you have; put simply the standard error depends on the square root of the number in each category, bigger n more precise.
unbalanced in Mixed models usually concerns the number of repeated occasions within the individual, but even then balance is not required.
No; use all the data you have; put simply the standard error depends on the square root of the number in each category, bigger n more precise.
unbalanced in Mixed models usually concerns the number of repeated occasions within the individual, but even then balance is not required.
Thank you for your answer Professor Kelvyn Jones . Much appreciated!
I am still facing some problems regarding normality and heteroscedasticity.
Is it recommended, as for the linear models, to use log or sqrt transformation to meet the normality assumption? Or un-normality means that the nonlinear function is not the best to be used?
I did not find any article addressing these issues for nonlinear models. any recommendations?
The assumption of normaility etc is not about teh response but about the residuals Can you give more details on which nlme model you are using eg logit with binomial random does not assume normality and has inbuilt heterogeneity.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size