Moving up 10 pp in the distribution and its effect

[Neg Kha]

Hi,

Say I have three variables.
One is grades and one is days in school, the last is id. A below:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(days id grades)
 17  1 4
 17  2 2
 17  3 4
 19  4 1
 19  5 5
 20  6 3
 34  7 3
 37  8 3
 37  9 4
 37 10 2
160 11 5
160 12 4
160 13 4
160 14 3
160 15 2
160 16 4
160 17 2
200 18 4
200 19 4
200 20 1
end

I want to show how moving up 10 percentage point in the days distribution correlates with grades. How should I approach this?

Thanks in advance

[George Ford]

summ days
g daysp = days/r(max)
reg grades daysp
margins, at(daysp = (0.1(0.1)1))
marginsplot

[Neg Kha]

summ days
g daysp = days/r(max)
reg grades daysp
margins, at(daysp = (0.1(0.1)1))
marginsplot

Thank you! If I also have a dummy variable for gender, how can I plot the margins by gender?

[Erik Ruzek]

This should get you slopes for the genders in your data at each of the requested daysp.

summ days
g daysp = days/r(max)
reg grades c.daysp##i.gender
margins gender, at(daysp = (0.1(0.1)1))
marginsplot

[Neg Kha]

This should get you slopes for the genders in your data at each of the requested daysp.

summ days
g daysp = days/r(max)
reg grades daysp##i.gender
margins gender, at(daysp = (0.1(0.1)1))
marginsplot

Unfortunately it is not possible to interact gender with daysp in the reg since daysp contains noninteger values. Is there any other way?

[Mike Lacy]

Having non-integer values for a variable in no way precludes specifying that it might interact with some other variable in its effect. Figuring out how and why you came to that incorrect idea could help you now and in the future. In specifying to Stata that daysp is to be treated as a continuous variable, you'd do well to use the c prefix with it in your regression command, i.e. reg grades c.daysp##i.gender. (The c. prefix is Stata's default presumption for a factor variable, but it's good to be in the habit of being explicit.)

[Neg Kha]

Having non-integer values for a variable in no way precludes specifying that it might interact with some other variable in its effect. Figuring out how and why you came to that incorrect idea could help you now and in the future. In specifying to Stata that daysp is to be treated as a continuous variable, you'd do well to use the c prefix with it in your regression command, i.e. reg grades c.daysp##i.gender. (The c. prefix is Stata's default presumption for a factor variable, but it's good to be in the habit of being explicit.)

Thanks, previously I ran the following line of code:

reg grades daysp##i.gender

and I got the error I talked about. I guess it is necessary to include the c so Stata knows it is a continuous variable.
Now your code works, thank you!