I am running mixed models analyses because I have participants nested in groups. Across the different analyses, I have very different degrees of freedom in the lme output with nlme package (and also different when I export them through sjplot). Sometimes it's higher for the group level variable than for the level-one covariates, sometimes it's the reverse. How reliable is it?
Thank you
Can you please provide more information on between what models the df change and what you exactly mean by : "export them through sjplot"?
Some R code and model details might be very helpful.
The df in general might change between packages and analyses because there seems to be no gold standard how to obtain df for multilevel models, instead there are different approaches to eatimate dfs that might produce differing results.
Can you please provide more information on between what models the df change and what you exactly mean by : "export them through sjplot"?
Some R code and model details might be very helpful.
The df in general might change between packages and analyses because there seems to be no gold standard how to obtain df for multilevel models, instead there are different approaches to eatimate dfs that might produce differing results.
I'd recommend using lmer and lmerTest instead. While the authors of lmer advise against trying to calculate dfs at all, some simulations suggested that the p-value estimates lmerTest are more accurate than the estimates you get from any of the other df-less alternatives (such as bootstrapping, just pretending the t-value is a z-value, etc.); see Luke 2017.Article Evaluating significance in linear mixed-effects models in R
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R