[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Negative variance for multiple imputed multilevel model

From   Maarten buis <>
To   stata list <>
Subject   st: RE: Negative variance for multiple imputed multilevel model
Date   Mon, 5 May 2008 08:29:40 +0100 (BST)

> --- Asrul Akmal Shafie <> wrote:
> I am Asrul Akmal from Malaysia. I would like to ask your input here
> regarding the use of multiple imputations in STATA for my research in
> HIV. I am interested in applying multilevel model on my
> longitudinal cost data.
> Using 
> xi: mim: xtmixed lncsumoprice tindex2 age i.cd4cat i.vlcat
> i.c2clinstage i.cregcat4 i.cinitregcat4 || patientid: tindex2,
> variance
> results in negative variance for my model result. I also have tried
> to explore running xtmixed without mim using one of the five imputed
> dataset. The result gives positive variance for the same model
> specification. I would like to have your opinion for this
> peculiarity.

--- Maarten buis []  
> Parameter estimates provided by -mim- are means of the parameter
> estimate in each imputed dataset. So my gues would be that, the
> variance is (very) negative in at least one of these datasets. To see
> if that is the case I would do the following:
> xi: mim, noisily : xtmixed lncsumoprice tindex2 ...
> Where ... means the rest of the command as you specified it in your
> message. So, I have added the -noisily- option to -mim-, which will
> display output from execution of the command within each of the
> imputed
> datasets. You are looking for the dataset in which the variance is
> negative. Once you have found that dataset you will probably want to
> compare the imputed and the observed observations and see if there is
> anything strange going on.

--- Asrul Akmal Shafie wrote me privately:
> I have run through your suggestion. Unfortunately I cannot find any
> negative variances in xtmixed of any of the imputed dataset (log
> attached). Could there be any other reason for this negative

The log shows that -mim- did not result in negative variances but in
negative ln(variances), so in order to see the variance all you need to
do is e^parameter, and you will see that these numbers are positive.

-- Maarten

Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

Yahoo! For Good. Give and get cool things for free, reduce waste and help our planet. Plus find hidden Yahoo! treasure
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index