I developed logit models for mode choice behavior using a questionnaire-based dataset. However, a reviewer suggested me to go for mixed logit models. I do understand that mixed logit model solves the limitations of the logit model. However, logit models are easier to estimate (using software packages such as SPSS) and interpret, and I have already developed and explained these models in my paper. So, I want to know if there is a way to determine whether or not developing mixed logit models will be any good.
Conditional logit model (this is what you mean by logit, right?) assumes the property of independence of irrelative alternatives (IIA). If your data don't satisfy this assumption, your parameters would be biased. In that case, you can use mixed logit model. The interpretation of parameters is not that much different from the conditional logit model. The mixed logit model adds an error component to the equation, which induces heteroscedasticity and can be correlated among alternatives. Thus, the marginal utility of attributes is allowed to vary among individuals. I don't know for SPSS, but in SAS, the code is very simple.
I hope the following link will help you to a certain extent. Please read the first paragraph.
https://en.wikipedia.org/wiki/Mixed_logit
Getting different results when using the mixed logit or multinomial probit estimator in contrast to the conditional logit estimator is when you have a highly parsimonious model.
This is captured by the distribution parameters of the coefficients of the included variables, leading to the false conclusion that the coefficients are not fixed.
With regards to the Monte Carlo investigation, the outcome clearly indicates that omitted variables could cause a rejection of the true hypothesis of fixed coefficients.
Conditional logit model (this is what you mean by logit, right?) assumes the property of independence of irrelative alternatives (IIA). If your data don't satisfy this assumption, your parameters would be biased. In that case, you can use mixed logit model. The interpretation of parameters is not that much different from the conditional logit model. The mixed logit model adds an error component to the equation, which induces heteroscedasticity and can be correlated among alternatives. Thus, the marginal utility of attributes is allowed to vary among individuals. I don't know for SPSS, but in SAS, the code is very simple.
Guillaume answer is very clear.
You cannot buse SPSS to estimate a mixed logit. However you can do it in R without spending any license money.
By experience I can tell you that you need a big sample to identify random parameters. The drawback of using the mixed formulation is that you cannot have a close formulation to use your model.
Let me know if you have any questions as I have used a lot of these modelling solutions
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS