Dummy variables

[harvardjanitor7]

I'm relatively new at everything; could anybody explain to me what a dummy variable is?

I'm having trouble understanding this question:

Run a regression of ln(profits) on majority owner’s age; a dummy for whether the majority owner
is female; dummies for the majority owner being Black, Asian, or Hispanic; dummies for the majority
owner being a high school dropout or having a high school degree but no 4 year college degree.
How do ln(profits) change if the owner is female compared to male? How about if the owner is 65
compared to 55?

So far, this is what I think I'm supposed to do:

ls(h) profits c ownage dfemale dblack dasian dhisp lehs d hsltcoll

/\ this would be a regression of profits including all the variables asked in the question

and then:

view -> coefficient tests -> wald ->
c(1)=0, c(2)=0, c(3)=0, c(4)=0, c(5)=0, c(6)=0, c(7)=0

And then I'm lost. I'm not even sure if I did that Wald test right.

Any help would be appreciated! :D

[hannibal]

hi,
dummy is variable who take 1 or 0. if you want to restrict your analyse to a specific groupe within your sample you have to use dummyvariable. in your case you have to create,say dum_black, a dummy who take 1 if the person is black and 0 otherweis.
you can for example introduce the dum_black, dum_white ect.. in you regression equation as additional variable to compare between many ethnicgroupe. you can also, as i sayed, restrict you sample just to black (sample if dum_black=1) and run the regression just for this groupe.

[harvardjanitor7]

ok cool

so to run a regression of certain variables, i type:

ls(h) profits c ownage dfemale dblack dasian dhisp lehs hsltcoll

and to test whether or not certain variables are statistically significant, i use dummy variables right?
so i use a wald test and set certain variables to 0 or 1, and then look at the coefficients and probability that eviews gives...right?


i'm having a little trouble actually understanding the ls(h) actually
basically what i put above would be testing "profits" against "ownage defemale dblack dasian dhisp lehs hsltcoll," right?

[bensamen]

Hey,

I think that u're trying to estimate an ancova model. It stands for analysis of covariance and is a general linear model with one continuous (quantitative) dependent variable and one or more qualitative and quantitative independent variables.
1. Qualitative variables are like sex and race (in your case). The difference between sex and race is that sex is a binary (binomial) while race is multinomial.
My point is this. You need to be careful when you wanna convert such variable in dummies. It's simple and easy for sex because you have 2 events : male =1, female = 0.
But for race (multinomial), the dummies must be n-1, with n = number of cases. Thus, the dropped category will be interpreted like the reference case.Unless, there will a multicolinearity problem.
So, you can decide to drop hispanic and estimate the effect of being black and white for example.

2. Ls(h) estimate the equation using Least squared method with white correction of heteroscedasticy. First of all, you need to estimate your model using ls command only. After, you can check if some of OLS assumptions are violated. If so, you can use (h) option to correct heteroscedasticity, normality and so on.

3. You dont have to use wald restriction test because there is no restriction in your model. We use restriction when using the Coob Douglas production function Y = A*Lexp(α)*Kexp(β) for example, we wanna test the increasing return of scale α + β > 1.
In your case, i think that the impact of being a female can be found by the sum up of the intercept and the dfemale coefficient, ceteris paribus.

Best regards,