Problems with Interval Regression

[Antonio Arruebo]

To whom may offer me help,

I would like to run "intreg" since my dependent variable is an interval from 0 to 90. However, the depvar is expressed in point data, not in interval format. See the screenshot:


My intention is to create 9 interval categories:
- 1 (0 , 0)
- 2 (3 , 5.9)
- 3 (6 , 14.9)
- 4 (15 - 29.9)
- 5 (30 - 44.9)
- 6 (45 - 59.9)
- 7 (60 - 74.9)
- 8 (75 - 89.9)
- 9 (90 - 90)

or find an alternative way of running "intreg".

I would highly appreciate some help in this regard. Thanks in advance.

Antonio

[John Mullahy]

It seems to me there are two separate issues here. First is that your dependent variable is doubly bounded. Second is that you wish to coarsen it using intervals.

I'm not sure about your rationales for the second (since it is intentionally discarding information). Ignoring the second consideration and focusing only on the first you might consider estimating your model using -fracreg- after rescaling your LHS variable to a 0-1 interval. E.g.

gen rnewWTP=newWTP/90
fracreg probit rnewWTP x1 x2...

[John Mullahy]

Following up on #2, I think I misread the question. Evidently you know the intervals in advance in which case you would not be discarding information. If that's correct then using -intreg- may be fine.

You will need to code two new dependent variables. This is from the help file.

    depvar1 and depvar2 should have the following form:

             Type of data                  depvar1  depvar2
             ----------------------------------------------
             point data          a = [a,a]    a        a
             interval data           [a,b]    a        b
             left-censored data   (-inf,b]    .        b
             right-censored data   [a,+inf)   a        .
             missing                          .        .

In your case it appears that for the 0 and 90 outcomes you would code depvar1=depvar2=0 and depvar1=depvar2=90. For the other outcomes you would use the a,b coding.

[Antonio Arruebo]

Thank you very much Mr.Mullahy! Really appreciate your help. This sets me in the right direction.

Kind regards,
Antonio.

[George Ford]

maybe tobit or truncreg would do. intreg is just tobit, I think. can use ll and ul rather than create new variables.