Re: PSPP-BUG: Logistic Regression bugs

[Renan Levine]

Dear Mr. Darrington,

I understand that logistic regression is in development, and am happy to 
help in any way I can. When I teach social statistics, we end with a 
very basic introduction to logistic regression, so a logistic regression 
routine in PSPP will enable me to cover almost the entire course (and 
then some!) with PSPP.

The problem with the error message only concerns dichotomous dependent 
variables, not predictor variables. Missing values on the predictor 
variables do not pose any problems. Cases with missing values on any 
independent variables are dropped just like when completing OLS regressions.

I think unequivocally that what the routine needs to do is to ignore all 
missing values and just focus on the non-missing categories. For 
example, STATA's manual says:  logit fits a maximum-likelihood logit 
model.  depvar=0 indicates a negative outcome; depvar!=0 & depvar!=. 
(typically depvar=1) indicate a positive outcome.

The way I understand that SPSS statement (if its not a typo) is that the 
SPSS routine will generate a predicted value for any observations with a 
missing value on the dependent variable, assuming that none of the  
independent variables contain any missing values for that observation. 
This is one way that some use maximum likelihood techniques to impute 
missing values.

Yours,
Renan



On 13-Nov-12 3:29 PM, John Darrington wrote:
> You are right.
>
> As you may have noticed, the logistic regression feature is still in 
> development.
>
> As of the version you mentioned, the /CATEGORICAL subcommand was not 
> implemented,
> but it has been since, try the most recent HEAD.
>
> I think there may still be a problem with missing values in the categorical
> variables, and I am working on that.  Missing values in the non-categorical
> predictor variables should be ok though.
>
> I'm unsure exactly how it should behave in response to missing values on the
> dependent variable.  The spss documentation says:
>   "For a case with a missing value on the dependent variable, predicted values
>    are calculated if it has non-missing values on all independent variables."
> This statement doesn't make sense to me.  Why would a predicted value be
> calculated? from what?  - I'm still thinking about that ...  Have you any
> idea?
>
> So thanks for your feedback.  I appreciate you taking the time to test and
> report these things.
>
> If you can do some similar tests with a very recent version (
> 3cd65292e3cc6bd6532214dcc8c8ddc65bdc2972 or later, I would appreciate it).
>
> Particularly, more tests with missing values and with weighted values would
> be great.
>
> Regards,
>
> John
>
> On Tue, Nov 13, 2012 at 01:08:19PM -0500, Renan Levine wrote:
>       Dear PSPP users and programmers,
>       
>       Thank you for your hard work in developing new capabilities for
>       PSPP. I'm using the most recent version of PSPP, psppire.exe
>       0.7.9-gaef7f5
>       
>       I recently encountered some problems while running logistic regressions:
>       
>       1) There appears to be a bug in the logistic regression routine
>       that causes it to recognize missing values in the dependent
>       variable as a value category. So, even when a variable is coded
>       0, 1 and [system] missing (common in public opinion data), PSPP
>       gives an error message: "Dependent variable's values are not
>       dichotomous."
>       
>       I've run logit analyses on three different .por and one .sav
>       datasets, tried to see if user-missing is treated differently
>       than system-missing, and if declaring missing values works any
>       differently than a recode statement. The only way I manage to run
>       a logistic regression is if I recode the dependent variable to be
>       two integers with no missing values.
>       
>       2) Less critically, I'm not sure the syntax /CATEGORICAL=var is
>       working correctly. When I include that line, letting the computer
>       know that an independent variable is dichotomous, I get an error
>       message: .3-13: error: Syntax error at 'categorical'. HOWEVER,
>       just including the variable on the initial line with the other
>       independent variables seems to work (I can't be certain because I
>       did not cross-reference my results with another statistics
>       program).
>       
>       Yours,
>       Renan
>       

--
Renan Levine
Department of Political Science
University of Toronto - Scarborough
address@hidden
http://individual.utoronto.ca/renan
(416) 208-2651