RE: st: Convergence not acheived

["Momplaisir, Florence" <Florence.Momplaisir@tuhs.temple.edu>]

Thanks David and Joseph for your response. I don't have direct access to the data so it took me time to figure things out. As you mentioned, the number of cells is not the only problem. The main issue is that most women only had one pregnancy in the dataset, so overall, it looks like the dataset is more cross-sectional than longitudinal. I ended up using a regression.

Thanks for your help!

Florence

________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of David Hoaglin [dchoaglin@gmail.com]
Sent: Saturday, September 21, 2013 8:47 AM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Convergence not acheived

Hi, Florence.

The cell sizes are not the whole story.  You did not mention the
number of cells (with nonzero sizes); all 6 of the predictors in that
-xtlogit- command are categorical.  The messages about collinearity
suggest that the distribution of the observations over all the
possible cells is such that some of the predictors are perfectly
correlated.  For example, if two predictors are dichotomous and the
2x2 table for those predictors has nonzero numbers of observations
only in the diagonal cells (or only in the off-diagonal cells), it is
not possible to distinguish their contributions.

Some diagnostic work with the predictors should shed light on the
problem.  The total number of cells (including cells with no
observations) may be too large for you to list all of them and study
their pattern, but you should be able to look at the 15 two-variable
crosstabs and even (if necessary) some of the three-variable
crosstabs.

It may be difficult to collapse categories further, but the data seem
to be telling you that you need to do that (or perhaps even remove one
or more predictors from the model).  The relations between the
predictors may suggest a way of combining two (or more) of them into a
composite (a sort of index).

David Hoaglin

On Fri, Sep 20, 2013 at 6:04 PM, Momplaisir, Florence
<Florence.Momplaisir@tuhs.temple.edu> wrote:
> Hello all,
>
> I'm using a mixed effect model to assess factors associated with getting an elective c-section among HIV + women. I used these commands (listed below) but I'm getting an error message saying that convergence is not acheived. My cell sizes vary between 3 and 111 and it is difficult for me to collapse things further (wouldn't make clinical sense). What should I do in that case?
> xtset Stateno
> .   xi: xtlogit elective_c i.adequate_pre1 i.delivery_agegrp1 i.r
>> ace_cat1 i.illdrug i.C_RPT i.gestage1 if viralsup==2, or pa c(e
>> xc)
> i.adequate_pre1   _Iadequate__0-2     (naturally coded; _Iadequat
>> e__0 omitted)
> i.delivery_ag~1   _Idelivery__0-2     (naturally coded; _Ideliver
>> y__0 omitted)
> i.race_cat1       _Irace_cat1_0-2     (naturally coded; _Irace_ca
>> t1_0 omitted)
> i.illdrug         _Iilldrug_0-1       (naturally coded; _Iilldrug
>> _0 omitted)
> i.C_RPT           _IC_RPT_1-2         (_IC_RPT_1 for C_RPT==False
>>  omitted)
> i.gestage1        _Igestage1_0-1      (naturally coded; _Igestage
>> 1_0 omitted)
> note: _Iilldrug_1 dropped because of collinearity
> note: _Igestage1_1 dropped because of collinearity
> estimates diverging (correlation > 1)
> r(430);
>
> Florence Momplaisir, MD MSHP

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


________________________________

This electronic message is intended to be for the use of the named recipient, and may contain information that is confidential or privileged. This communication may contain protected health information (PHI) that is legally protected from inappropriate disclosure by the Privacy Standards of the Health Insurance Portability and Accountability Act (HIPAA) and relevant Pennsylvania Laws. You can direct questions concerning PHI or HIPAA to the Corporate Compliance and Privacy Officer at (215) 707-5605. If you are not the intended recipient, please note that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this message in error, you should notify the sender immediately by telephone or by return e-mail and delete and destroy all copies of this message.


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/