Date:         Wed, 4 Jun 2003 16:15:49 -0300
Reply-To:     Alexandre Cechin <grothaal@hotmail.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Alexandre Cechin <grothaal@hotmail.com>
Subject:      Re: spss DUMMIE variables
Content-Type: text/plain; charset=iso-8859-1; format=flowed

Thanks Bin Fan, I will try to reduce the amount of independent variables as
you say, as i allready tryed to run a model with more than 40 variables and
it didnt work (the memory thing) .

But SPSS do accept this quantity od independent variables if you put only
one variable (original with the 40 different values) and ask him to aply a
contrast into it and set it as categorical into the dialog box. This way he
makes "virtual Dummies" only for the procedure, but it enter the variable as
a block, and not one at a time.
The problem is that the procedure inputs all categories that way, and not
only the significant ones.

My real question at this time is:  there a big problem in letting the non
significant ones into the model, is there a big change into final results?


>From: "Bin Fan" <fanb@childpsych.columbia.edu>
>To: <grothaal@hotmail.com>
>Subject: Re: spss DUMMIE variables
>Date: Wed, 04 Jun 2003 14:43:13 -0400
>
>
>Alexandre Cechin:
>
>As my understanding is:
>You have a variable which with 40 different categories, and you created 40
>New Dummy variables (0 or 1), and you want run
>all the 40 dummy variables in Logistic Regression at same time.
>
>If that true, my answer is: YOU CAN NOT DO THAT IN SPSS. (Because there is
>the limitation for the RAM on your computer).
>
>I did this before, the maximon of the independent variables is <=20 (if you
>have thousands cases).
>
>YOU CAN TRY:
>(1) Decrease your independent variables number: you can run linear
>correlation between your dependent variable (1) and 40 independent
>       variables, and eliminate these not highly significant variables, and
>keep less than 20 variables in Logistic Regression;
>
>(2) Run this one in SAS (Logistic Regression): Convert your SPSS data file
>into SAS data file, and run same Logistic procedure in SAS.
>
>(3) Run this one in SUDAAN (Logistic Regression): Convert your SPSS data
>file into SUDAAN or SAS data file, and run same Logistic
>       Regression in SUDAAN or SAS Callable SUDAAN:  Logistic  Dependent
>Variable = 1 Independent Variable (with 40 categories),
>       SUDAAN will get the same result as SPSS or SAS (40 Variables).
>
>Do hope this will works for you.
>
>Best wishes.
>
>
>Bin Fan
>Columbia University
>
>***********************************************************************************************************************************
>
> >>> Alexandre Cechin <grothaal@hotmail.com> 06/04/03 01:11PM >>>
>Hi,
>
>Im using the logistic regression model in SPSS, and i have one variable
>that
>is categorical and have about 40 diferent categories.
>
>When apllying contrast, it run the model well, but all 40 categories are
>tested as a block into the model.
>This results in some coeficients that are significant at the .05 level and
>others that are not.
>
>Is there any recomendation on what to do with these non significant
>coeficients? I already tried to input these variables with manual
>'dummiefication' but my SPSS does not support a large amount of variables
>and cases (1million, that is already a subset from the universe of cases)
>in
>this procedure (memory overflow).
>
>Is it safe to let these non-significant categories into the model without
>much 'damage' to the results?
>
>
>Thanks all,
>
>_________________________________________________________________
>MSN Messenger: converse com os seus amigos online.
>http://messenger.msn.com.br
>
>

_________________________________________________________________
MSN Messenger: converse com os seus amigos online.
http://messenger.msn.com.br