Programming the statistical procedures from SAS

Marginal effect of dummy variable in tobit model

Reply
Occasional Contributor
Posts: 6

Marginal effect of dummy variable in tobit model

own votefavorite

I have a tobit regression model in hand and I want to calculate the marginal effects at means for the dummy variables in the reg equation. There are two dummy variables in the regression equation and three continuous variables. For SAS, this link has some information for the marginal effects for a tobit model : (http://support.sas.com/rnd/app/examples/ets/margeff/). However this tells me how to compute the marginal effects for continuous variables and not for a dummy variable in the case of a tobit regression.

This is what I have done following the SAS link. Note that f and g are dummy variables which takes the value of 1 or 0.

proc qlim data=tobit1;
model a = b c d f g;
endogenous a~censored(lb=0 ub=1);
run;

Here I input the means of the variables. Note that for the marginal effect of dummy variable g, I cannot set the value of f to its mean value as it is a dummy variable and set it at 0.

data tobit2;
input b c d e f g a;
datalines;
0.48 0.34 5.59 5.57 0 1 .
0.48 0.34 5.59 5.57 0 0 .
run;

data tobit;
set tobit1 tobit2;
run;

proc qlim data=tobit noprint;
model a = b c d f g;
endogenous a~censored(lb=0 ub=1);
output out = tob2 marginal;
run;
proc print data = man2;run;

After inputting the extra values and running the tobit regression, I get the marginal values for g 'Meff_g' for the last two added observations. In order to get the marginal effect of g at means I use:
Meff_g(at g =1) - Meff_g(at g=0) i.e. substracting the Meff_g for the last two observations. Is this correct? Could somone help please?

Respected Advisor
Posts: 2,655

Re: Marginal effect of dummy variable in tobit model

After reading the Web example about 3 times, I think you have it down.  My vision missed the periods in the tobit2 dataset, and once I saw what you were doing, it all seems to fit.  Now it is just a matter of doing the subtraction, and looking at the plots (always look at the plots).

Steve Denham

Occasional Contributor
Posts: 6

Re: Marginal effect of dummy variable in tobit model

Ok. Thank you so much. For some reason I get very low marginal effect values (which I probably shouldn't be getting here, not sure why this is happening). So in the above case, I have the marginal effect for f = 0. I need to calculate the marginal effect for f =1 and state that separately, correct? What happens when I have more than 2 variables? For eg if I have three dummy variables male (0= male, 1 = female), employed (0= unemployed, 1 = employed) and college educated (0 = no, 1 = yes), and I want to find the marginal effect of college education on say income, then would I need to get the marginal effects for the groups (male,employed), (male,unemployed), (female,employed), (female,unemployed). Is there a way to get one marginal effect value for all the four groups?

Respected Advisor
Posts: 2,655

Re: Marginal effect of dummy variable in tobit model

You might want to take a look at PROC LIFEREG.  It accommodates a CLASS statement and both LSMEANS and LSMESTIMATE statements, so that a tobit model could be fit, and marginal estimates (the LSMEANS) be obtained for all levels of each variable in the CLASS statement. Look through the documentation for this and see if it could do what you need.

Steve Denham

Occasional Contributor
Posts: 6

Re: Marginal effect of dummy variable in tobit model

Will do. Thanks a ton again.

Ask a Question
Discussion stats
  • 4 replies
  • 475 views
  • 4 likes
  • 2 in conversation