ordinal logistic regression diagnostic

[amatoallah ouchen]

Good day Stata-listers,
I'm currently  running an ordinal logistic regression, all the
predictors are categorical, In order for my analysis to be valid, my
model has to satisfy the assumptions of ordinal  logistic regression,
the problem is that all what I could find out   is material related to
the diagnostic of the logistic regression  and more rarely some hints
related to ologit but that don't seem to work for my case (for example
when I use the fitstat command I get the message: variable _cons not
found
r(111))
In addition I never came up with a similar model ( ordinal logistic
regression with only categorical predictors).
So it is statistically correct to run a such model (or do we need at
least one quantitative predictor?).

Any help would be highly appreciated,

Thanks a lot in advance,
Ama
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

[Martin Weiss-5]

<>

" and more rarely some hints
related to ologit but that don't seem to work for my case (for example
when I use the fitstat command I get the message: variable _cons not
found
r(111))"


-set trace on- and see where it gets stuck. In this example, Stata completes
the task w/o encountering any problems:


*************
webuse fullauto, clear
ologit rep77 foreign length mpg
fitstat
*************



HTH
Martin


-----Ursprüngliche Nachricht-----
Von: [hidden email]
[mailto:[hidden email]] Im Auftrag von amatoallah
ouchen
Gesendet: Montag, 26. April 2010 16:24
An: [hidden email]
Betreff: st: ordinal logistic regression diagnostic

Good day Stata-listers,
I'm currently  running an ordinal logistic regression, all the
predictors are categorical, In order for my analysis to be valid, my
model has to satisfy the assumptions of ordinal  logistic regression,
the problem is that all what I could find out   is material related to
the diagnostic of the logistic regression  and more rarely some hints
related to ologit but that don't seem to work for my case (for example
when I use the fitstat command I get the message: variable _cons not
found
r(111))
In addition I never came up with a similar model ( ordinal logistic
regression with only categorical predictors).
So it is statistically correct to run a such model (or do we need at
least one quantitative predictor?).

Any help would be highly appreciated,

Thanks a lot in advance,
Ama
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


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

[amatoallah ouchen]

Thanks a lot Martin for your prompt answer,
But in the example you gave  below, you have  used 2 quantitative
variables (length and mpg) , the problem I mentioned occur when only
using  qualitative variables,
If I reuse  your example

webuse fullauto, clear
ologit  rep77 rep78 model
-fitstat-
variable _cons not
found
r(111))

I think that my issue is more clear now,
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

[Maarten buis]

--- On Mon, 26/4/10, amatoallah ouchen wrote:
> I'm currently  running an ordinal logistic regression,
> all the predictors are categorical, In order for my
> analysis to be valid, my model has to satisfy the
> assumptions of ordinal logistic regression, the problem
> is that all what I could find out is material related to
> the diagnostic of the logistic regression and more
> rarely some hints related to ologit but that don't seem
> to work for my cas

There is a lot of material available from this websit by
Rich Williams: <http://www.nd.edu/~rwilliam/gologit2/>

You can also consider the -brant- command, which is part
of -spost9- (see: -findit spost-). Rich and I did ran
quite a bit of simulation to assess the performance of
that test, see the thread starting with:
<http://www.stata.com/statalist/archive/2008-04/msg00660.html>
I believe the end conclusion was that each category of
your dependent variable needs to contain "enough"
observations for the test to perform well.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


     

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

[Martin Weiss-5]

<>

Nope, runs without any problems for me:

*************
webuse fullauto, clear
ologit  rep77 rep78 model
fitstat
*************

Output:


Measures of Fit for ologit of rep77

Log-Lik Intercept Only:        -89.895   Log-Lik Full Model:
-59.677
D(60):                         119.354   LR(2):
60.437
                                         Prob > LR:
0.000
McFadden's R2:                   0.336   McFadden's Adj R2:
0.269
ML (Cox-Snell) R2:               0.600   Cragg-Uhler(Nagelkerke) R2:
0.642
McKelvey & Zavoina's R2:         0.653                              
Variance of y*:                  9.470   Variance of error:
3.290
Count R2:                        0.636   Adj Count R2:
0.385
AIC:                             1.990   AIC*n:
131.354
BIC:                          -132.026   BIC':
-52.057
BIC used by Stata:             144.492   AIC used by Stata:
131.354



Which version of -fistat- do you use? Mine is

. which fitstat
c:\ado\plus\f\fitstat.ado
*! version 1.8.1 2007-09-18


HTH
Martin


-----Ursprüngliche Nachricht-----
Von: [hidden email]
[mailto:[hidden email]] Im Auftrag von amatoallah
ouchen
Gesendet: Montag, 26. April 2010 16:49
An: [hidden email]
Betreff: Re: st: AW: ordinal logistic regression diagnostic

Thanks a lot Martin for your prompt answer,
But in the example you gave  below, you have  used 2 quantitative
variables (length and mpg) , the problem I mentioned occur when only
using  qualitative variables,
If I reuse  your example

webuse fullauto, clear
ologit  rep77 rep78 model
-fitstat-
variable _cons not
found
r(111))

I think that my issue is more clear now,
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


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

[amatoallah ouchen]

Greetings,
Thank you very much Martin and Maarten,
Martin: you are right my fitstat version was an old one,
Maarten: thank you very much for the materials,

Regards,

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

[Richard Williams-4]

At 09:52 AM 4/26/2010, Martin Weiss wrote:
>Which version of -fistat- do you use? Mine is
>
>. which fitstat
>c:\ado\plus\f\fitstat.ado
>*! version 1.8.1 2007-09-18

fitstat often causes confusion because there are multiple versions of
it floating around, and (unlike most programs) the version at SSC is
NOT the most current.  Most people will want to do

findit spost9

which contains the most up to date version of fitstat.  It also
contains the brant command, which seems to be about as good a way as
any for testing the assumptions of the ordered logit/ proportional
odds model.  Other options include the omodel and gologit2 commands,
both available from SSC.  gologit2 offers a possible way of dealing
with violations of assumptions, although other options are out there,
e.g. use a different kind of ordinal model or, if the violations are
widespread enough, just use mlogit.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: [phone], [phone]
HOME:   [phone]
EMAIL:  [hidden email]
WWW:    http://www.nd.edu/~rwilliam

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

[amatoallah ouchen]

Thank you very much Richard, spots9 finally allowed me to get the
results obtained previously by Martin,

Regards,
Ama
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/