Re: st: [Why does Stata drop an additional category?]

[Richard Williams <Richard.A.Williams.5@nd.edu>]

At 08:26 AM 2/17/2004 -0500, Marcello Pagano wrote:
> The first four of the following commands all end in _Ieduc5_1 dropped
> due to collinearity; the last gives the coefficient for this category
> but the model statistics seems strange: no constant and no comparability
> with the output in SPSS for the same data (the syntax and the results
> are attached).
>
> xi: logistic unemployed  i.religion i.educ5 i.cath*e5c
> xi: logistic unemployed  i.religion i.educ5 i.cath*e5c,coef
> xi: logit    unemployed  i.religion i.educ5 i.cath*e5c, or
> xi: logit    unemployed  i.religion i.educ5 i.cath*e5c
> xi: logit    unemployed  i.religion i.educ5 i.cath*e5c,nocons or
>
> I would be most grateful if anyone could explain to me why in the first
> four commands the first category of education (_Ieduc5_1) is dropped (I
> used both Stata 7 and Stata 8); and how to obtain an output similar to
> that in SPSS. The results for models 2-4 are not presented because they
> are very similar to results from model 1.
With regards to educ5, something has to be dropped; if you have a 5 
category variable only 4 dummies can be created from it and included in the 
model.  When you specify nonconstant you make it possible for all 5 
categories of educ5 to be entered, but of course you lose the constant then.

With regards to SPSS versus Stata:  Stata is reporting the odds ratios by 
default.  If you add the parameter -,coef- to the end of your logistic 
commands, you'll get the coefficient estimates instead. Or, use the -logit- 
instead of the -logistic- command, as -logit' reports coefficient estimates 
by default. It looks like you are doing that, so...

Other seeming differences may reflect differences in how the dummy variable 
coding is handled, i.e. Stata and SPSS may be using different reference 
categories.  -xi- uses the smallest/first value as the reference 
category.  SPSS Logistic, I believe, uses the highest/last category as the 
reference.  So, you'll get algebraically equivalent results, but the 
coefficients will look different because different reference categories are 
used.

If you look at the SPSS documentation for Logistic, you'll see how you can 
change the reference category.  Or, in Stata, you can change the reference 
category with a command like

char varname[omit] #

e.g. char educ5[omit] 5

In short, I am guessing your results are equivalent, but SPSS and Stata are 
reporting the results in different but equivalent ways.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/