|
Hi Bhupinder,
Thanks for the response.
I think I understand somewhat. In you're first sentence you say that if you
think there's is a random effect it should be specified. Can't the argument
be made that there is always an subject specific random effect? If that's
the case I guess my question is when is meaningful to specify it?
I not quite sure what you meant by "if the estimates of the parameters are
big enough". Would you please clarify?
Thanks,
Andy
On Mon, Aug 2, 2010 at 4:48 PM, Bhupinder Farmaha
<bhupi80singh@yahoo.co.in>wrote:
> Hi Andy
>
> Based on my knowledge, if you have random components then specific it
> otherwise they will get pooled into error terms. It might make your model
> more complex but make sense if the estimates of the parameters are big
> enough. Otherwise you can exclude that from the random statement.
>
> Does it make sense ?
> Bhupinder
>
> -----Original Message-----
> From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of
> Andrew
> Agrimson
> Sent: Monday, August 02, 2010 4:41 PM
> To: SAS-L@LISTSERV.UGA.EDU
> Subject: Random Effects question
>
> Hello all,
>
> I have a question regarding random effects.
>
> Originally I was building a logistic regression model with multiple
> observations per subject. To account for this I included a random intercept
> term to induce correlation within subjects. My main focus though was the
> fixed effects and the inclusion of the random intercept term was only to
> account for the within subject correlation.
>
> I have recently decided to role up the observations to the subject level
> and fit a binomial model instead. As I was preparing to do this I began to
> wonder if including a random intercept in the binomial would still be
> appropriate, i.e., most similiar to the random intercept logistic model. I
> guess my thoughts are that a random intercept will induce the needed
> correlation within subject, but it's also a bit more than that. It's also
> going to estimate the unmeasured effects. It seems that this approach would
> be similar to a frailty model in survival analysis. Is this appropriate
> given that the main focus is on the fixed effects? Does anybody have any
> thoughts on this?
>
> Thanks Andy
>
>
|
