Skip to content

lme4

3 messages · Azra Ramezankhani, Phillip Alday, Ben Bolker

#

  
  
#
This is generally not advisable because the sampling distribution of the
variances is highly skewed. This means that the standard error isn't a
particularly informative statistic.?

This has been discussed in various amounts of detail in various places,
but a good starting point (with examples and links) is this question on
CrossValidated: https://stats.stackexchange.com/q/161225/26743

For a concrete recommendation, I would use either profile or bootstrap
confidence intervals instead of standard errors. You can compute these
with confint(model, method="profile") or confint(model, method="boot").
For more information, see:
http://search.r-project.org/R/library/lme4/html/confint.merMod.html

Best,

Phillip
On 24/6/20 2:30 pm, Azra Ramezankhani wrote:
#
?? If you really, really want these standard errors, I believe you can 
now get the from the merDeriv package (on CRAN). See ?merDeriv::vcov.lmerMod


I'd be very careful with this though!? It gives a standard
On 6/24/20 11:00 AM, Phillip Alday wrote: