correlation coefficient

[ulas alkan]

Hi,

I have the following code for a simulation. I need help to conclude it.

set obs 100
gen ut=.
gen vt=.
gen yt=1
generate order=_n
gen zt=1
capture program drop my_model
program define my_model, rclass
replace ut = rnormal(0,1)
replace vt = rnormal(0,1)
replace yt=4+yt[_n-1]+ vt if order!=1
replace zt=2+zt[_n-1]+ ut if order!=1
regress yt zt
return scalar constant = _b[_cons]
return scalar coef = _b[zt]
return scalar t = _b[zt]/_se[zt]
return scalar r2 = e(r2)
exit
end
simulate constant=r(constant) coef=r(coef) t=r(t) r2=r(r2), reps(1000): my_model


How can I record correlation coefficient between y and z for each simulation? Corr gives matrix but I need to record only the correlation coefficient and then take the average of them. Thanks in advance.

Ulas

[Nick Cox]

Given something like

corr yt zt

the resulting single correlation will be returned in r(rho). This is documented in the help for correlate.

However, as you are already returning R-squared from the corresponding regression, this hardly seems necessary. The (absolute value) of the correlation is just its (positive) square root and nothing informative is added by adding the correlation to your other results. (If the correlation is negative, the sign is also preserved as the sign of the slope.)

By the way, I wouldn't usually average correlations; I would work with the FIsher z transform in combining correlations. For more on that, see http://www.stata-journal.com/article...article=pr0041

EDIT: I would advise setting the seed for work like this.

[ulas alkan]

Thank you for your response but I could not fix my code. I did not understand where I should add "corr yt zt" and "r(rho)". Stata help does not answer this question or I could not get it.

[Clyde Schechter]

As Nick points out, you don't even need to use -corr- in this context, because you already have R², which is the square of the correlation you seek, and its sign will always be the same as the sign of the zt coefficient. So after you get your simulate results you can just calculated it as -gen r = sign(coef)*sqrt(r2)-.