Calculate economic significance for censored (Tobit) regression with multiple imputation

[Valdemar Meng]

Hi,

I'm trying to calculate economic significance from a censored (Tobit) regression with multiple imputation according to this definition:

"Economic significance is indicated by the predicted change in the dependent variable from a movement from the tenth to the ninetieth percentile in the independent variable as a percentage of the mean of the dependent variable".



So far I have the following code (with y being my dependent variable, and x being my independent variable(s)):

mi register y x
mi impute mvn y x, add(10)
mi estimate, cmdok post: tobit y x, ll(0) ul(1) vce(robust)



I have tried using the following code (which works without mi estimate) immediately after the above:

sum y, meanonly
local y=r(mean)
margins, expression(predict(ystar(0,1))/`y_m') at((p10) x) at((p90) x) contrast(atcontrast(ar._at) marginswithin) atmeans



But I get the error:

e(sample) does not identify the estimation sample



Any help would be very much appreciated! :-)


Best regards,
Valdemar

[Valdemar Meng]

I did some more digging, and found this promising thread:

https://www.statalist.org/forums/for...design-mimrgns


Unfortunately, I still can't get it to work. I adapted the code proposed by daniel klein as shown below:

mi register y x
mi impute mvn y x, add(10)
mi estimate, cmdok post: tobit y x, ll(0) ul(1) vce(robust)

capture program drop mimargins
program mimargins, eclass properties(mi)
svy: tobit y x, ll(0) ul(1) vce(robust)
sum y, meanonly
local y_m=r(mean)
margins, expression(predict(ystar(0,1))/`y_m') at((p10) x) at((p90) x) contrast(atcontrast(ar._at) marginswithin) atmeans post
end

mi estimate: mimargins, expression(predict(ystar(0,1))/`y_m') at((p10) x) at((p90) x) contrast(atcontrast(ar._at) marginswithin) atmeans

When running the code, I get the error "varlist specification required" after mi estimate: mimargins ... .


Once again, any help would be very much appreciated.