Hello everyone,

I'm working with National Longitudinal Survey of Youth (NLSY79 and NLSY97), which include two datasets, with the same variables, about students surveyed in 1979 and 1997.

I created as output a categorical variable, schooling_status, which assumes values:
- 1 corresponding to "college participants"
- 2 corresponding to "high school drop-outs"
- 3 corresponding to "high school graduates"
- 4 corresponding to "year college graduates".

I've run on both datasest firstly a multinomial probit and, secondly, an ordered probit including a set of regressors as controls and one variable indicating the family income. Consequently, to interpret the effect of income on the 4th outcome (year college graduates) I computed the marginal effects.

For the multinomial probit I had: NSLY79 --> ME_income = .0004999; NLSY97 --> ME_income = 0.0005007.
For the ordered probit I had: NSLY79 --> ME_income = 0.0000599; NLSY97 --> ME_income = 0.0001765.

As you may notice, while for the multinomial logit the marginal impact of the var income does not vary throughout times (only a 0.16% variation), with the oredered probit I have a huge variation in the marginal effect of income between 1979 and 1997... In the 2nd period te impact is 195% higher!

Can someone kindly explain me the reason of this big difference between the two models?

Thanks a lot!