Answers for "Tobit Regression not working"
https://developer.ibm.com/answers/questions/217917/tobit-regression-not-working.html
The latest answers for the question "Tobit Regression not working"Answer by kathrin.s
https://developer.ibm.com/answers/answers/217921/view.html
After your last post I tried again, and the tobit works just as it should (no idea why it didn t yesterday). Results are the same as in R.<br>
So the problem with the missing values and not needed variables in the dataset is solved!<br>
The problem was just the scale level I think. With scales the function works and as you told me, it is correct to use only scales (I create dummies for all kategories of a variable und leave one out of the regression specification to refer to).<br>
<br>
Thanks for your help!!!<br>
KathrinThu, 14 Apr 2011 17:18:22 GMTkathrin.sAnswer by SystemAdmin
https://developer.ibm.com/answers/answers/217920/view.html
The Tobit dialog and syntax provides for listwise deletion of cases with missing values, so I don't see why you would have to do that differently. And variables not referred to in the regression specification should make no difference.<br>
<br>
As for the measurement level issue, categorical variables are converted automatically into R factors, so they enter the equation differently from scale variables. That is equivalent to creating a set of dummy variables corresponding to the factor variable. But some R packages do not handle the resulting singular regression matrices. It is okay to create the dummies, including only enough to avoid singularity, and treat them all as scale.<br>
<br>
If you can post a dataset that illustrates the problem, we can investigate further. And what version of Statistics are you using?<br>
<br>
HTH,<br>
Jon PeckWed, 13 Apr 2011 16:52:08 GMTSystemAdminAnswer by kathrin.s
https://developer.ibm.com/answers/answers/217919/view.html
In the meantime I found out what the reason for the "singular data" error was. In my SPSS dataset there are several cases with missing values. If I delete all cases with a missing value in any of my regression variables and also delete all other variables in the dataset I do not use in my regression, it works. The problem was just that R seems not to be able manage missing malues.<br>
But there is another problem in using the tobit regression now.<br>
In my data I have only categorical independent variables. So I have to work with dummys (0/1). But the tobit function only works if I define all my dummys as scale measured. So my question is, does the system know (and if so, how does it know?) which dummys belong together? E.g. the variable employment status (not employed, parttime, fulltime), here parttime and fulltime should be compared with unemployed people (my reference category). I think if I add the dummys parttime and fulltime to the model, the system only compares parttime-workers to all others and fulltime-workers to all others. Is that right? Someone calculated an identical model for me in R as I did in SPSS and there are different results.<br>
Are there special requirements using dummys in the tobit regression in spss?<br>
<br>
Regards,<br>
KathrinWed, 13 Apr 2011 10:42:40 GMTkathrin.sAnswer by SystemAdmin
https://developer.ibm.com/answers/answers/217918/view.html
It appears that the procedure thinks that your data are singular. It could also be that the distributions considering the zero and nonzero values separately are degenerate. Can you post a dataset that shows the problem?<br>
<br>
Regards,<br>
Jon PeckFri, 08 Apr 2011 16:46:20 GMTSystemAdmin