Storing marginal effects in a loop of regressions

[Marco Kuehne]

I would like to run several regression via a foreach-loop, calculate marginal effects, store and compare them. So far I managed to calculate the loop. I am able to store normal estimates in a table. I can access one marginal effect at a time, since its in a matrix object. That's where I got stuck. Related posts are:

• https://www.statalist.org/forums/for...sion-in-a-loop
• https://www.statalist.org/forums/for...s-from-margins

Here is a MWE:

*************************************************
* Load Panel Data
*************************************************
webuse nlswork, clear
xtset idcode

*************************************************
* Loop for different outcomes
local outcomes "ln_wage hours wks_work"
foreach o of local outcomes {
quietly xtreg `o' age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure c.tenure#c.tenure, fe
estimates store reg_`o'
}
estimates table reg_ln_wage reg_hours reg_wks_work
*************************************************

*************************************************
* Loop regressions and save marginal effects (difficult due to matrix object)
local outcomes "ln_wage hours wks_work"
foreach o of local outcomes {
quietly xtreg `o' age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure c.tenure#c.tenure, fe
margins, dydx(age)
matrix B`o' = r(b)
}
* to be continued
*************************************************

Thank you

[Phil Bromiley]

If you look at the documentation on matrices, it will show you how to read one element out of a matrix.

If you want to manipulate the output afterwards, sometimes it helps to create a counter in a local macro and host variables that are all missing, and then write each parameter into a single observation in the host variable using the counter.

There may also be a way to do this by directly converting the matrix to variables. This is documented in the PDF documentation.