Finding Sub-samples Based on OLS Results

[Hossein Hosseini]

Dear STATALIST,

I have an OLS regression of household's pollution on household's income and other characteristics (cross-sectional data). I want to find two sub-samples for which the coefficients of income are significantly different. How can I do this in STATA?

PS: Doing the regression with dummies for income brackets doesn't help because I want the income OLS coefficients to be different between groups not the income levels.

Thanks,

[Attaullah Shah]

If you can further clarify this

the coefficients of income are significantly different.

. Also, an example data set will help

[Hossein Hosseini]

My equation is :Pollution=a+b Income + c (Other Characteristics) +residual
By income coefficients I mean two sub-samples that their coefficient b are significantly different than each other. How can I find these two groups?

[Attaullah Shah]

Still your question is not clear. You want to estimate many regressions and then collect their coefficients, make two groups based on the coefficient values. The question is , what is your grouping variable in the first place that you shall use as a if qualifier for running different regressions. I would like to remind you to post a sample of your data using dataex (from SSC) if you want more helpful replies.

[Hossein Hosseini]

There aren't many regressions. There is just one regression. And there is no grouping variable. I want STATA to find groups in my data for which the effect of income on pollution are different than each other. I will post data ASAP, but suppose data is simply a cross-section of pollution levels of 120000 households with their income levels and other characteristics.

[Attaullah Shah]

I think you are looking for some threshold of income from where the effects start changing for pollution. If so , then type

findit threshold

in the Stata command window and you will find many helpful resources.

[Hossein Hosseini]

Regarding

I think you are looking for some threshold of income from where the effects start changing for pollution.

: No, I am not looking for an income threshold. As I mentioned in my initial post this could be done by income bracket dummies and interaction of income levels with those dummies to see how b changes between income brackets.

I want the two groups to be different just in terms of their b coefficients. This is the only condition that I want to impose initially. Then, after I find these two groups, I am going to investigate the income levels and other characteristics of households in each group. Ideally, I want to have two groups which b coefficients are different but have the similar income distribution: both groups to have sufficient number of both low-income and high-income people.
Hossein