ll or ll(0) for this case? the tobit regression

[Eunhye Kang]

Dear experts,

I am trying to use tobit for my model and am a bit confused of which one I should use between "ll" and "ll(0)".
This is the distribution of my dependent variable and it means the amount of hours respondents spent on taking care of children per day.
As you see, many cases belong to "0 hour" which means they didn't have time at all for childcare or they did not allocate their time for childcare.
I am using tobit regression model for my analysis and I wonder which is the correct command in such situation:

1) tobit DV IDV1 IDV2 ..., ll
2) tobit DV IDV1 IDV2 ..., ll(0)

[Fei Wang]

If the minimum of DV is 0, then -ll- and -ll(0)- lead to identical results. BTW, the DV can't be childcare hours per day. I guess it's per month?

[Eunhye Kang]

Thank you for your reply.
It is the daily hours spent on childcare collected by time diary method.

Results do differ from each other between using just "ll" and "ll(0)", and that's why I want to make sure I am using the correct one.
Would you have any ideas?

[Fei Wang]

Maybe you can show us the two results? Or display an example data using -dataex-?

[Eunhye Kang]

The associations independent variables have with the dependent variable (childcare time) come differently when I use "ll" and "ll(0)" for the tobit regressions.
Education matters when I use command with "ll" but this association is not statistically significant when I use "ll(0)", for example.
So I really need to know which one is the correct way of using it in this situation.
If the minimum value of my dependent variable is 0, do I still need to type "ll(0)" for the command or is "ll" the correct one?

[Joao Santos Silva]

Dear Eunhye Kang,

Rather than Tobit, I suggest you just use Poisson regression. After all, your dependent variable is a count (and it is not censored).

Best wishes,

Joao

[Eunhye Kang]

Thank you Joao Santos Silva for your comment. I've heard about using Poisson for the time use data. Not many people conduct that model, though.
I still want to hear your further comments on the reason for Poisson in this case. I'd appreciate it very much if you would.

And if I use tobit, what would be the best option?

[Joao Santos Silva]

Dear Eunhye Kang,

Poisson regression is explicitly designed for cases like this where the dependent variable is a count; more generally, it is suitable to any case where the dependent variable is non-negative. This contrasts with Tobit which is designed for linear models where there is censoring (which is not what you have). The reason few people use Poisson regression is that they are not aware of it, but see here

https://blog.stata.com/2011/08/22/us...tell-a-friend/

Best wishes,

Joao