You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I try to align my results with SAS. SAS offers the following options:
residual - practically useless in most applications, so I skip this.
containment. This is what nlme probably uses. SAS uses it by default when random effects are specified.
betwithin. This is what SAS uses for the random-only part by default. This is very in demand in clinical biostatistics. I think neither R package offers that.
Sattherthwaite and Kenward-Rogers (employs the Sattherwaite in it).
Now I'm trying to reproduce the following SAS models:
model with only random effects covariance structure, residual covariance set to zero.
model for repeated observations - only the residual covariance is set.
model with both random effects and residual covariance is set.
For each I need to set appropriate degrees of freedom. I could use boostrap, but I have to be 100% compliant with SAS. I got positive results with glmmPQL, which matched SAS more/less, being also flexible, even if biased in case of discrete conditional distribution of the response. Also the nlme::gls() and nlme:lme() somehow worked. But I cannot get closely matching results with glmmTMB (lme4 wouldn't allow me to set the residual covariance).
I cannot share the data, so instead I would like to ask about the default algorithm of calculating the degrees of freedom in glmmTMB to obtain both SE and the p-values. Is this just infinity (since I see z)?
I try to align my results with SAS. SAS offers the following options:
Now I'm trying to reproduce the following SAS models:
For each I need to set appropriate degrees of freedom. I could use boostrap, but I have to be 100% compliant with SAS. I got positive results with glmmPQL, which matched SAS more/less, being also flexible, even if biased in case of discrete conditional distribution of the response. Also the nlme::gls() and nlme:lme() somehow worked. But I cannot get closely matching results with glmmTMB (lme4 wouldn't allow me to set the residual covariance).
I cannot share the data, so instead I would like to ask about the default algorithm of calculating the degrees of freedom in glmmTMB to obtain both SE and the p-values. Is this just infinity (since I see z)?
For references:
https://blog.as.uky.edu/sta707/wp-content/uploads/2016/01/sas-ddfm.pdf
The text was updated successfully, but these errors were encountered: