Appropriate model when Y is an integer

[Shreya Jain]

I have constructed Y = Wife years of schooling -Husband's years of schooling , so Y is education gap/ difference. It ranges from -18 to 15.
Now I regress faher's education and other covariates on edu_diff variable. My variable of interest is Father's education so I see a negative coeff which is significant as well. How do I interpret it?
Also if this is the best model that we can use or we should use some other model?

[Carlo Lazzaro]

Shreya:
as per FAQ, please use CODE delimiters to share what you tyoed and what Stata gave you back. Thanks.
That said:
1) -poisson- works when y takes on integer values;
2) without further details from your side, I'd say that, other things being equal, one year more in father's education reduces Y of Z.

[Shreya Jain]

Thank you for your reply Sir Carlo Lazzaro. I tried poisson, negative binomial also, but the stata error, which says Y must be greater than or equal to zero (r(459), pops up. So please let me know if there is something else one can try. My Y ranges from -18 (minus 18) to 15 (positive 15)

[Carlo Lazzaro]

Shreya:
I misread that you have negative values too.
Let's try with -regress-, then.

[Shreya Jain]

So, well I am looking at the effect of father's years of schooling (fyos) on education difference of the daughter and son in law (edu_diff).
edu_diff= daughter years of schooling- son in law years of schooling
edu_diff ranges from (-18 to 15)

negative values -18 till -1 implies the marriage type is hypergamy
0 means homogamy
1-15 means hypogamy

given this information I wanted to understand how do I interpret the fyos's coeff which is negative and significant.

so I have already created the indicator variable and applied logit (hypo=1, non hypo=0) and mlogit (homo=0, hypo=1 and hyper=2). But wanted to see when we take edu_diff then what happens do my results change and it's easier to interpret a linear model.

[Carlo Lazzaro]

Shreya:
-fyos- coefficient tells you that, for those in the reference category (that is, 0.level) of -poor- categorical variable, each year of difference in -fyos- reduces the regressand by -.1742296.
You can get it easier by adding the -allbase- option to your OLS.

[Shreya Jain]

Thank you Sir for your reply. I just have a follow-up question: can we interpret in terms of the marriage type? Like each year of difference in fyos for non-poor (poor=0) reduces the education difference thus leading to homo or hypergamy. Whereas for poor we see an increase in edu_diff means higher hypogamy?
Also, Yos is years of schooling of daughter, so is it advisable to control for it in the regression? since my Y is constructed using that variable?

[Carlo Lazzaro]

Shreya:
1) as the interaction does not reach statistical significance, I would re-run the model without it and see how the coefficients change;
2) your statements should be tested in the light of 1);
3) no, -yos- is part of the y.

[Shreya Jain]

Thank you so much Sir Carlo Lazzaro

[Carlo Lazzaro]

Shreya:
Carlo is enough. Thank you.