I'm analysing a set of data that's recorded in percentages, for which the data can only be within 0-100%. I've used linear mixed model to analyse this data, however, some of the estimated marginal means (and confidence intervals) were either above 100% or less than 0%. I was told that I could consider transforming the data, but I'm not sure how to do that. Can anyone suggest how I could obtain estimated marginal means that would fit within an appropriate range of my data?
Either use a beta-regression model (R package "betareg", what does not allow random effects, or the package "glmmTMB" with family=list(family="beta",link="logit"), what allows adding random factors), or a logit- or arcsine-transformation (http://strata.uga.edu/8370/rtips/proportions.html) of the response assuming a normal distribution of the errors on the logit-scale (-> lmer).
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS