0

I am running a Random Coefficient Mixed Model in R using lme in {lme4}. I had to transform my dependent variable by square-root because of problems of uniqual variance of the errors. However, with this formulation of the DV, the interpretation of my coefficients' predictors becames quite tricky.

  1. Fitting a Non-Linear Mixed Model with nlme of the homonimous package in R would be a solution?
  2. Can someone help setting up the code for my case? I do not really get the meaning and the use of the argument start =.
Cœur
  • 37,241
  • 25
  • 195
  • 267
Caserio
  • 472
  • 1
  • 3
  • 14
  • 1
    There is not much information here to base solid advice on. I would only want to mention that fitting a variance structure with the weights parameter could be an alternative to the tranformation. – Roland Jul 18 '16 at 17:02
  • Could you give more information on how to proceed using the `lme` package? Given that I know that the heteroskedasticity comes from the non linear relation among DV and covariates and that DV_sqrt better describe this relation, how should I set the `weights =` option? In a nutshell which function should I use in `weights = varFunc()`? – Caserio Jul 18 '16 at 17:15
  • 2
    Sorry, I really don't want to go into more details without a lot more information. Possibly you even should use a GLMM. Do you have a non-linear model from the scientific background and know something about the expected error structure? I think your question might be better suited for stats.stackexchange.com. – Roland Jul 18 '16 at 17:24
  • I have tried fitting my model with `glme` using a gamma distribution, but the error structure does not improve. I have been able to identify part of the source of heteroskedasticity by including a specif variable in the random part of the equation, however the non-normal distribution of my DV translate in non-normal errors which cause heteroskedasticity. Transforming the DV with sqrt the qqnorm plot shows that the transformed-DV follow exactly a normal distribution. However, as already said interpretation of coeff.s is quite tricky. – Caserio Jul 18 '16 at 17:31

0 Answers0