Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: ordered logistic integration problems
From
Richard Williams <[email protected]>
To
[email protected], "[email protected]" <[email protected]>
Subject
RE: st: ordered logistic integration problems
Date
Thu, 21 Mar 2013 12:58:38 -0500
At 09:28 AM 3/21/2013, Bontempo, Daniel E wrote:
Thanks Richard. I follow about the difficulty of thresholds between
sparse categories, or even when some categories are not at all
levels of the IV's.
I do lack insight into why "ologit" quickly picked thresholds and
gave results, while gllamm and gologit2 seemed unable to pick
thresholds. I am going to avoid thresholds and use the glm with
link(logit) family(binomial) as suggested in another reply, but it
would be great to have more insight into why ologit had no apparent
problem and gologit2 failed - even when I used the parallel
assumption, and they were both estimating the same model.
If anyone else is interested, -gologit2- is available from SSC. To
estimate the same model, I assume you did something like
ologit y x1 x2 x3
gologit2 y x1 x2 x3, pl
If ologit ran fine and gologit2 did not, it may just be because
ologit is a better written program! Or, you might try adding the
-difficult- option to gologit2.
If you didn't use the -pl- option with gologit2, then you probably
were not estimating the same model.
If you can provide some syntax and output or a replicable problem I
can try to see if I can figure anything out. But make sure you really
were trying to estimate the exact same model. A generalized ordered
logit model potentially estimates far more parameters than an ordered
logit model does, which can be difficult to do if you have thin
counts in a category and/or many variables.
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Stas Kolenikov
Sent: Wednesday, March 20, 2013 8:26 PM
To: [email protected]
Subject: Re: st: ordered logistic integration problems
I second Richard. The message probably comes from the difficulty of
identifying the threshold parameters for the categories with the
fewest observations, especially if they interact in some odd ways
with the random effects and/or variance parameters. For as much as you
(understandably) hate to run this as a linear model, this may be a
better option, as the prior work did. Or, at the other extreme,
create a dummy "less than 100%", which will only have 10% non-trivial values.
-- Stas Kolenikov, PhD, PStat (SSC)
-- Senior Survey Statistician, Abt SRBI
-- Opinions stated in this email are mine only, and do not reflect
the position of my employer
-- http://stas.kolenikov.name
On Wed, Mar 20, 2013 at 6:20 PM, Richard Williams
<[email protected]> wrote:
> Occasionally adding the -difficult- option will work miracles.
>
> My guess, that you are spreading the data too thin. If I follow you,
> the DV has 12 values, and 90% of the cases are a 1, which means the
> other 11 values average less than 1% of the cases. With gologit2 you
> are estimating 11 sets of coefficients. I am not surprised you have to
> collapse to only 3 categories.
>
> But why are you using an ordinal model in the first place? Why not a
> model specifically designed for proportions? See, for example,
>
> http://www.stata.com/support/faqs/statistics/logit-transformation/
>
> http://www.ats.ucla.edu/stat/stata/faq/proportion.htm
>
>
> At 06:04 PM 3/20/2013, Bontempo, Daniel E wrote:
>>
>> Can anyone explain the kind of data conditions that cause gllamm or
>> glogit2 to spit out:
>>
>> flat or discontinuous region encountered numerical derivatives are
>> approximate nearby values are missing could not calculate numerical
>> derivatives missing values encountered r(430);
>>
>>
>> I have a colleague with proportion data that only has about 12
>> discrete values between 0 and 1 with about 90% 1's. Skew -3.27,
Kurtosis>15.
>>
>> We want to model for 3 groups (between) and 3 occasions (within).
>> Prior work published in 2000, had similar proportions and used HML
>> (Gaussian) and got interpretable results. After looking at the
>> distributions, I suggested ologit might be more appropriate than regress.
>>
>> I was already concerned about these proportion DVs because my
>> colleague has calculated proportion correct of however many scorable
>> events there were, and the number of events differs a lot from
>> subject to subject. Some have 2 some have 10. BUT - my question for
>> the moment is technical difficulty with numerical derivatives.
>>
>> Since there is occasion nested within person, I was interested in
>> gllamm with the ologit link, as well as robust ologit with
>> "cluster(subject)". I also tried glogit2 because I was unsure the
>> parallel regression assumption was met.
>>
>> I easily get ologit to run. However both gllamm and glogit2 make
>> similar complaints about missing or discontinuous numerical
>> derivatives and do not complete. I tried the log-log link in glogit2
>> since the values rise slowly from 0 and suddenly go to 1. I kept
rounding to get fewer levels.
>>
>> I have to collapse to only 3 levels to get glogit2 to run. gllamm
>> keeps telling me to use trace and check initial model, but when I do
>> I see reasonable fixed effect values.
>>
>> Is ologit able to use an estimation method that avoids these
>> integration issues?
>>
>> I am trying to get the disaggregated data so multilevel logistic
>> regressions can be done, but it is not clear disaggregated data will
>> be available.
>>
>> Any pointers, advice, suggestions, references ... all appreciated.
>>
>>
>> *
>> * 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/
>
>
> -------------------------------------------
> Richard Williams, Notre Dame Dept of Sociology
> OFFICE: (574)631-6668, (574)631-6463
> HOME: (574)289-5227
> EMAIL: [email protected]
> WWW: http://www.nd.edu/~rwilliam
>
>
> *
> * 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/
*
* 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/
*
* 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/
-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME: (574)289-5227
EMAIL: [email protected]
WWW: http://www.nd.edu/~rwilliam
*
* 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/