Skip to content

SUR in lmer

4 messages · Giulia Dotti Sani, Markus Jäntti

#
Hello everyone,

is there a way of running a seemingly unrelated regression under lme4
or other multilevel packages?

Thank you!

Giulia
#
On 03/29/2012 10:58 AM, Giulia Dotti Sani wrote:
Take a look at glm in nlme. You need to specify both a correlation structure (to 
allow the seemingly unrelated units to be correlated through their error terns) 
and a variance function (to allow for different residual variances for the units).

Best,

Markus

  
    
#
Thank you. However, I can't see how I can specify two dependent  variables.
On Thu, Mar 29, 2012 at 11:06 AM, Markus J?ntti <markus.jantti at iki.fi> wrote:
#
On 03/29/2012 05:08 PM, Giulia Dotti Sani wrote:
You don't. You  reshape the data into "long" format, have a factor that 
indicates which unit's observations you are using, interact the explanatory 
variables with that indicator and also use that indicator for the variance 
function.

Markus