spss ordinal regression

[Knuvty]

Unexpected singularities in the Fisher Information matrix are encountered. There may be a quasi-complete separation in the data. Some parameter estimates will tend to infinity.
How to tackle the proplem

[Rich Ulrich]

(a) increase the sample size, by a lot;
(b) decrease the number of predictors;
(c) decrease the number of categories in the ordinal outcome;
(d) use a linear model (Least Squares Regression).

You probably can't do (a).
You should do (b) in any case, if you are including more than a few.

The possibilities for (c) and (d) depend on information that you have
not mentioned, such as the N, the k (predictors), and the success of
prediction that you would reasonably expect ... which we could possibly
judge if you mentioned details such as "predicting what from what?"

--
Rich Ulrich

> Date: Tue, 14 Jun 2016 00:32:58 -0700

> From: [hidden email]
> Subject: spss ordinal regression
> To: [hidden email]
>
> Unexpected singularities in the Fisher Information matrix are encountered.
> There may be a quasi-complete separation in the data. Some parameter
> estimates will tend to infinity.
> How to tackle the proplem
>
>

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Jon Peck]

Other possibilities -

- collapse some dependent variable categories

- Use NOMREG

- Use the STATS FIRTHLOG extension command, which fits a penalized logistic model

On Tue, Jun 14, 2016 at 8:38 AM, Rich Ulrich <[hidden email]> wrote:

(a) increase the sample size, by a lot;
(b) decrease the number of predictors;
(c) decrease the number of categories in the ordinal outcome;
(d) use a linear model (Least Squares Regression).

You probably can't do (a).
You should do (b) in any case, if you are including more than a few.

The possibilities for (c) and (d) depend on information that you have
not mentioned, such as the N, the k (predictors), and the success of
prediction that you would reasonably expect ... which we could possibly
judge if you mentioned details such as "predicting what from what?"

--
Rich Ulrich

> Date: Tue, 14 Jun 2016 00:32:58 -0700

> From: [hidden email]
> Subject: spss ordinal regression
> To: [hidden email]
>
> Unexpected singularities in the Fisher Information matrix are encountered.
> There may be a quasi-complete separation in the data. Some parameter
> estimates will tend to infinity.
> How to tackle the proplem
>
>

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

--

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Art Kendall]

For some reason I do not see the OP on Nabble.

Are you sure you need ordinal regression?

Try CATREG in SPSS.  see whether there is a meaningful difference between MODEL fits with with ORDINAL specification and INTERVAL specifications.  (You might also try a NOMINAL specification.)
CATREG does a test between models with different assumptions about measurement level.