Methodology question

[Yahya Ghazali]

Dear All

My query is regarding methodology. I have a sample of firms from 2003 until 2014, which include state-owned firms and non-state owned firms.

I have the following regression equation and I want to examine the impact of the reform on state owned firms (DSTATE):

reg COVERAGE DSTATE REFORM DSTATE*REFORM

The reform started in 2005 onwards. REFORM is the dummy variable which takes value of 1 for year >=2005, and 0 otherwise. DSTATE is a dummy variable if the firm is state-owned, and 0 otherwise.

My query is regarding the variable DSTATE, does this variable shows the association between COVERAGE and DSTATE in the pre-reform period only, or does it account for the whole period i.e 2003 until 2014?

Many thanks

[Clyde Schechter]

reg COVERAGE DSTATE REFORM DSTATE*REFORM

will get you no results, only a syntax error.

What you want is:

reg coverage i.dstate##i.reform

The coefficient you get for dstate, with the definitions of dstate and reform you are using, will estimate only the association between coverage and dstate in the pre-reform period. By using an interaction model, you are stipulating in your model that there is no such thing as the association of dstate with coverage over the entire period, so nothing in your model will, in fact, estimate that non-existent parameter. By using an interaction model you are stating that there are two different associations of dstate with coverage, one for the pre-reform period and the other for the post-reform period. Those parameters are estimated, respectively, by the dstate coefficient and the sum of the dstate and dstate#reform coefficient.\

Actually, the easier way to get those estimates it to follow the regression with the -margins- command:

margins reform, dydx(dstate)

[Yahya Ghazali]

Dear Clyde

Thank you for this.

I have another query regarding my next methodology. The sample starts from 2003 until 2014 and includes state-owned firms and non-state owned firms. The reform started in 2005 onwards. REFORM is the dummy variable which takes value of 1 for year >=2005, and 0 otherwise. DSTATE is a dummy variable if the firm is state-owned, and 0 otherwise.

The dependent variable is share price, PRICE, and I predict that earnings, EARN, will be significant for state-owned firms relative to non-state owned firms in the post-reform period. This is more like a difference in difference approach and I have used the following model:

reg PRICE i.DSTATE##i.REFORM##c.EARN

My query is regarding the interpretation of the different variables which i will get from the above regression. Also, for my prediction, which variable I should be focusing on?

Many thanks

[Marcos Almeida]

Your query, if I understood right, is basically related to the core-kowledge concerning regression analysis, particularly the use (and interpretation) of interaction terms. Surely an introductory book on the matter (plus the Stata Manual) will be very helpful. That being said, using - margins - and - marginsplot - can give you a fine way to interpret the results. By typing - help margins - in Stata's command window, excellent information will be shared as well.

[Clyde Schechter]

The dependent variable is share price, PRICE, and I predict that earnings, EARN, will be significant for state-owned firms relative to non-state owned firms in the post-reform period.

If this is a complete and correct statement of the research hypothesis, I would suggest a different approach. This hypothesis says nothing at all about what happens in the pre-reform period. So you can greatly simplify your model by leaving out i.REFORM and restricting the estimation sample to the post-reform era. That way the model just becomes

reg PRICE i.DSTATE#c.EARN if REFORM

which leaves you in the more familiar territory of a dichotomous by continuous interaction, which is much easier to interpret and understand.

Not only would this simplify the model, it improves it by omitting extraneous information that would contribute noise, but not signal to your test of the hypothesis.

[Yahya Ghazali]

Dear Clyde

Many thanks for this. Yes, that makes it simple.

Best regards