interpretation of control variable and i.industry interpretation

[Trang Quyen]

Dear Statalist users,


I have the following situation
My code in stata is xtreg ROA A B C D E F i.INDUSTRY,re robust, when ROA is my dependent variable, A,b,c,d,e is my control variable, CCC is one of my independent variable, when I enter code in stata, it turns the result :
xtreg ROA Age size AssetTurnover Leverage SaleGrowth CCC i.INDUSTRY,re robust

Random-effects GLS regression Number of obs = 1,014
Group variable: Company Number of groups = 205

R-sq: Obs per group:
within = 0.0930 min = 4
between = 0.3451 avg = 4.9
overall = 0.2506 max = 5

Wald chi2(20) = 576.09
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

I want to compare the influence of each industry to ROA but not sure how to interpret that with 1 industry omit, furthermore, regarding to control variables: Age, size, assetturnover, leverage, sale growth, with each model, it turns out different result ( I have four model with 4 different independent variable, but control variables the same)
Can anyone help me at this, I'll be very appreciate.

[Clyde Schechter]

Well, your industry variable is a categorical variable with many levels. The standard way to represent this in any kind of regression is with inidicator (also called "dummy") variables. One level of the categorical variable is omitted, because if you enter an indicator for every level, then those indicators always sum to 1, and that colinearity makes it impossible to do the regression. So one is omtited. In terms of comparing those different industries, I wouldn't do that from the regression output. I would, after the regression, run -margins, industry-. You will get an expected value of ROA for each industry (including the one that was omitted from the regression) that is suitably adjusted for the distribution of all the other variables. Comparing those makes more sense to me.

furthermore, regarding to control variables: Age, size, assetturnover, leverage, sale growth, with each model, it turns out different result ( I have four model with 4 different independent variable, but control variables the same)

Yes, of course the results are different: you are using different independent variables, so everything changes. Why would you expect them to be the same?

[Trang Quyen]

Thank you for your reply
What I mean is, beside identify the influence of independent variable to dependent variable, I also want to know the relations between control variable and dependent variable?
So if I have 4 different model for four different independent variable ( Same dependent ROA and same 5 control variables for four model)
how can I identify the relation between control variable ( size , age... on ROA)
what the commands margin will do and how can I interpret that? I’m so sorry I’ m new stata learner and I tried to learn it by myself but it quite tough for me

[Clyde Schechter]

how can I identify the relation between control variable ( size , age... on ROA)

You can't. There is no such thing as the relationship between these variables and the outcome: there are many different relationships depending on what other variables are in the model. Each model will give you different relationships for these variables.

Within any specific model, the model coefficients give relationships between these variables and the outcome that are valid with exactly that set of variables. But if you change anything else, the coefficients of these variables also can (and usually do) change.