I think I'm hitting a small bug when I use -fvset- with -margins-. I assume this is similar to "Slight Bug in Margins after Fixed Effects Estimation" (https://www.statalist.org/forums/for...s-estimation)? Example:
Code:
sysuse auto, clear
est clear
/* make an outcome */
gen y = (price>5000)
/* generate a categorical variable x1 */
gen x_str_temp = substr(make,1,1)
encode x_str_temp, gen(x1)
bys x1: drop if _N<5 /* drop some obs to avoid having few obs per cell */
ta x1, m
keep y foreign x1
qui {
* no problem here
probit y foreign i.x1
noi eststo fvsetNO: margins, dydx(*) post atmeans
* now foreign "not estimable" because of -fvset-
fvset base 1 x1
probit y foreign i.x1
noi eststo fvsetYES: margins, dydx(*) post atmeans
* -noestimcheck- option can solve this problem, it seems
probit y foreign i.x1
noi eststo fvsetYES_and_noestimcheck: margins, dydx(*) post atmeans noestimcheck
* show results side-by-side
noi estout *, cells( b(fmt(a4) star) se(fmt(a4) ) ) modelwidth(30)
}
exit
