Computing marginal effects for each equation after using the multivariate tobit model

[Eutteum Lee]

Dear Statalist users,

I have estimated a bivariate tobit model where I have two dependent variables, using the mvtobit command (multivariate tobit).
Now, I would like to compute marginal effects for each equation.
Accordingly, I have two tasks to do:

a. prediction corrected for bottom coding at 0 (since both of the dependent variables are bottom-censored at 0)
b. prediction for each dependent variable

I know the following commands:
i. "margins, dydx(x) predict(ystar(0,.)))": prediction corrected for bottom coding at 0
ii. "predict(pmarg1)" and "predict(pmarg2)": prediction for equation 1 and equation 2, respectively.

Command i. is not appropriate because I have two equations (hence two dependent variables).
Moreover, command ii. does not work; it seems to me that the commands (i.e., predict(pmarg1), predict(pmarg2)) are packaged in Stata.

Is there any way I could compute marginal effects for each dependent variable, corrected for bottom coding at 0?
I would very much appreciate your help.

[John Mullahy]

Eutteum: Hopefully Stephen Jenkins might weigh in further with his expertise. However, if it is the case that the post-estimation options in -mvtobit- do not support these -margins- commands, then you might consider estimating each equation separately by univariate Tobit and then using -margins- after each. For instance, see

help tobit_postestimation##margins

I do not think that the the parameters of interest to you involve the correlation parameter(s) from bivariate/multivariate Tobit, so univariate Tobit should provide consistent—even if not fully efficient—estimates of those parameters.

[Eutteum Lee]

Thank you so much for your reply.
Then, I was wondering if bivariate tobit, instead of multivariate tobit, might have commands to compute marginal effects for each equation.
I know bitobit and mvtobit are the same as long as there are two equations, but I was just curious whether I might have marginal effects calculated for each equation.

Thank you again so much for your advice.

[John Mullahy]

Eutteum: Alas, I have not used -bitobit- so cannot offer a useful response. Hopefully others will share their wisdom.

[Eutteum Lee]

Dear Mr. Mullahy,

Thank you again so much for your reply!

[Fredrick Nuani]

Hello everyone,

I wish to ask for the steps to follow or commands to use when analysing categorical data with three dependent variables using a multivariate Tobit model. Thank you.