Hi everyone. I have some questions about lmer() and I'm wondering if you might be able to help me out. I?ve run a mixed-model in lmer() including multiple random effects and a fixed effect. I?ve noticed that if I run a comparison simple regression model with no random effects but the same fixed effect, I get precisely the same unstandardised beta coefficient for my fixed effect in the simple regression as I get if I run the mixed model. Am I correct in thinking, therefore, that the beta coefficients generated in the mixed model in lmer() do not control for the random effects in the model? Would it make a difference in this respect if participants in my dataset were nested in a random effect? Also, how does one summarise an overall R square value for the impact of a set of fixed effects in a mixed model with lmer()? On another note, I was wondering if there were any more recent suggestions about how to handle random effects in REML-based models using lmer() that are fenced at zero. Is it possible that alternative optimizers might assist in this respect? One issue I find when comparing lmer() with Bayesian estimators is that the difficulties in this respect often appear to arise with very small effects (i.e., those that approach near-zero variance estimates). Looking forward to hearing your thoughts. Kind regards, Duncan
Fixed effects in lmer()
2 messages · Duncan Jackson, Thierry Onkelinx
Dear Duncan, If the random effects only explain the noise of the linear regression, then you could get similar fixed effect estimates. The summary of both models would be useful. IMHO R? has only a clear definition under very special conditions: a linear regression with Gaussian distribution and without random effects. Unfortunately, as people start learning statistics with this kind of model, they assume that other models have the same properties. You could do an LRT between models with and without the set of fixed effects. Are you referring to non-negative random effects? The lmer random effects assume a zero mean normal distribution. Which implies the possibility of negative numbers. Best regards, ir. Thierry Onkelinx Statisticus / Statistician Vlaamse Overheid / Government of Flanders INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND FOREST Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance thierry.onkelinx at inbo.be Havenlaan 88 bus 73, 1000 Brussel www.inbo.be /////////////////////////////////////////////////////////////////////////////////////////// To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data. ~ John Tukey /////////////////////////////////////////////////////////////////////////////////////////// <https://www.inbo.be> Op vr 18 jun. 2021 om 08:24 schreef Duncan Jackson <duncanjackson at gmail.com
:
Hi everyone.
I have some questions about lmer() and I'm wondering if you might be able
to help me out.
I?ve run a mixed-model in lmer() including multiple random effects and a
fixed effect. I?ve noticed that if I run a comparison simple regression
model with no random effects but the same fixed effect, I get precisely the
same unstandardised beta coefficient for my fixed effect in the simple
regression as I get if I run the mixed model.
Am I correct in thinking, therefore, that the beta coefficients generated
in the mixed model in lmer() do not control for the random effects in the
model? Would it make a difference in this respect if participants in my
dataset were nested in a random effect? Also, how does one summarise
an overall R square value for the impact of a set of fixed effects in a
mixed model with lmer()?
On another note, I was wondering if there were any more recent suggestions
about how to handle random effects in REML-based models using lmer() that
are fenced at zero. Is it possible that alternative optimizers might
assist in this respect? One issue I find when comparing lmer() with
Bayesian estimators is that the difficulties in this respect often appear
to arise with very small effects (i.e., those that approach near-zero
variance estimates).
Looking forward to hearing your thoughts.
Kind regards,
Duncan
[[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models