Margins command error (not found in list of covariates) for continuous variables interaction effects

[Valentina Schneider]

Hello, my name is Valentina and I am a PhD student.

I am trying to conduct a post-hoc analysis on a significant interaction I found. The interaction is between power and closeness and their effect on anger:

reg anger c.power##c.closeness

I then followed Aiken & West (1991) and UCLA"s Stata help (https://stats.oarc.ucla.edu/stata/se...ions-stata/#s3), with following code and without mean-centering the variables:

sum closeness
return list
global closea = round(r(mean) + r(sd), 0.1)
global close = round(r(mean), 0.1)
global closeb = round(r(mean) - r(sd), 0.1)
display $closea
display $closeb
margins, dydx(power) at(closeness=($closea $close $closeb))

This worked well. I then wanted to do exactly the same with mean-centered variables to avoid extrapolation. I thus created mean-centered variables:

summarize closeness, meanonly
gen centered_closeness = closeness - r(mean)
summarize power, meanonly
gen centered_power = power - r(mean)

and then ran the same commands again:

sum centered_closeness
return list
global centered_closea = round(r(mean) + r(sd), 0.1)
global centered_close = round(r(mean), 0.1)
global centered_closeb = round(r(mean) - r(sd), 0.1)
display $centered_closea
display $centered_closeb
margins, dydx(centered_power) at(centered_closeness=($centered_closea $centered_close $centered_closeb))

Now, I received an error message from Stata: centered_power not found in list of covariates or centered_closeness not found in list of covariates.

I tried to fix the issue by reading similar error message posts but couldn"t find a solution, since I have only continuous variables.

Please help! Thank you.

Valentina

[Andrew Musau]

margins refers to the last estimates in memory. As the error message indicates, there is no estimation with the variables centered_power and centered_closeness. You need

reg anger c.centered_power##c.centered_closeness

prior to the margins command.

[Valentina Schneider]

Thank you so much, Andrew!