Calculating Jacobian for Delta Method standard errors with a logit model including quadratic and interaction terms

[Jake Meyer]

Hello,

I'm attempting to calculate the marginal effects and their standard errors after a logit estimation including an interaction and a quadratic term (both using the same continuous variable). However, this is also part of a mediated estimation where the effect of the same variable used in the quadratic and interaction term is being tested indirectly through another control variable. Margins (as far as I know) doesn't do this, so I'm attempting to essentially redo the output of the margins command manually similar to what I've found here:

https://www.statalist.org/forums/for...argins-command

Only altering the Jacobian and average marginal effect to account for this mediated effect. I've figured out how to reproduce the marginal effect of the output for the specification with the quadratic and interaction term - and therefore can calculate the average marginal effect including the mediated effect, but I'm stuck on the Jacobian, leaving me unable to figure out standard errors. If someone one could help me by explaining how the process in the link is carried out when an interaction term is included I'd much appreciate it.

Thanks,
Jake

[Wilson Law]

It has been almost a year, so I am not sure if you have found a solution. If I understand your problem correctly, one way you can accomplish this is to derive the marginal effect and use "predictnl" to compute the result and generate the standard error for you. Logit fortunately has a closed form solution so it should work. I hope that helps.