Dear all,
I have one more methodological question on the multinomial logit model in relation to selection bias.
Now, the Durbin McFadden correction is applied when the outcome equation is non independent from a choice equation.
In my case, I have a large cross section dataset of observations involving thousands of individuals, and I perform a multinomial logit regression using as independent variables some individual characteristics against the dependent (choice) variable represented by the heating system (i.e., gas, electric, wood, carbon, other) that each individual chooses.
When I subsequently regress with standard OLS the energy consumed by each individual (outcome equation) against some independent variables, I apply the Durbin McFadden correction, since energy consumed is reasonably affected by the type of heating system chosen.
Now, my question is: when I run the multinomial logit model at the beginning, I have some individuals which do not utilize any type of heating system (therefore, in correspondence of those individuals, I have missing observations with regard to the heating system chosen). Pardon my ignorance, but shouldn't I have to apply some form of correction also here, before performing the multinomial logit? Or I should just run the multinomial logit simply ignoring (deleting) the observations related to these individuals?
I am asking this, because for what I have seen so far, it seems to me that when using multinomial logistic regressions the correction is always (and only) applied when you regress an outcome equation, not when you use the multinomial logit in the choice equation.
If someone could provide me some sugegstions in this regard I would be very glad.
Thank you.
Kodi
I have one more methodological question on the multinomial logit model in relation to selection bias.
Now, the Durbin McFadden correction is applied when the outcome equation is non independent from a choice equation.
In my case, I have a large cross section dataset of observations involving thousands of individuals, and I perform a multinomial logit regression using as independent variables some individual characteristics against the dependent (choice) variable represented by the heating system (i.e., gas, electric, wood, carbon, other) that each individual chooses.
When I subsequently regress with standard OLS the energy consumed by each individual (outcome equation) against some independent variables, I apply the Durbin McFadden correction, since energy consumed is reasonably affected by the type of heating system chosen.
Now, my question is: when I run the multinomial logit model at the beginning, I have some individuals which do not utilize any type of heating system (therefore, in correspondence of those individuals, I have missing observations with regard to the heating system chosen). Pardon my ignorance, but shouldn't I have to apply some form of correction also here, before performing the multinomial logit? Or I should just run the multinomial logit simply ignoring (deleting) the observations related to these individuals?
I am asking this, because for what I have seen so far, it seems to me that when using multinomial logistic regressions the correction is always (and only) applied when you regress an outcome equation, not when you use the multinomial logit in the choice equation.
If someone could provide me some sugegstions in this regard I would be very glad.
Thank you.
Kodi
Comment