Specification of the dependent variable

[eric mount]

Hey

I have a dataset where I want to check whether the model with a logistic dependent variable (log(y)) fits the model better than the dependent variable in levels (y). We find a lot of opportunities to test specification of the independent variables, but none to test for the specification of the dependent variable.
Are there any easy ways to test this in STATA?

[Carlo Lazzaro]

Eric:
you should probably provide the list with more details about your problem.
Do you mean you're performing a -logistic- regresssion? Or what else? Thanks.

[eric mount]

Instead of using the dependent variable (hours) in levels, I log transform hours and reestimate the model with log hours as the dependent variable (while keeping the independent variable in levels). I want to find the best model, but I understand it isn´t as easy as just comparing the r-sq.

Hours is not binary btw. Thanks for the answer!

[Rich Goldstein]

"better" is here undefined; in general, I would not treat R-squared as defining "better" by itself; however, if that is one stat you want to look at, I suggest you search for a download "brsq" which originally appeared in the STB - you can download the original paper if you want also (for free from the Stata web site)

[Carlo Lazzaro]

Eric:
you did not mention -hours- in your post #1, btw.

[Marcos Almeida]

Hello Eric,

Welcome to the Stata Forum.

Apart from the excellent comments in #2 and #4, I wish to remark that due to the confusing term in #1 ("a logistic dependent variable"), only after reading attentively your message did it became clear to me that you meant a logarithmically transformed dependent variable, for the word "logistic" would prompt us to think about a binary variable under a - logistic - regression.

This being the case, if I guessed correctly, you may be dealing with linear regression, having time - or the log of time - as the dependent variable.

To end, please prefer to write "Stata" as recommended in the FAQ.

Best,

Marcos