Imagine one wants to estimate the determinants of job changes in a given year. The outcome is '1' when a job change takes place and '0' otherwise. Imagine one wants to (abstracting from all kinds of endogeneity) estimate the effect of the (actual) wage on the job change. My intuition says that, while controlling for the actual wage before the job change is ok, controlling for the wage of the new job is not, since this conditions on the result. (It should be ok however to condition on the average wage of all other jobs or on the individual wages of all potential other jobs), Is this correct?
I think it is more appropriate to conditionon the individual wages of all potential jobs as it may be consistent with the initial obejective of estimating the the effect of the (actual) wage on the job change, since its is a binary response based model.
Thanks Mustapha and Pandelis. Meanwhile and thanks to a discussion with a colleague, my intuition got a bit farther. The problem with conditioning on the wage in the new job is that you don't observe it for those that do not change jobs. Thus, as the two of you suggest conditioning on some sort of average of alternative jobs might be fine (but which), or you might estimate the model only for those that got an alternative job offer and where you observe the wage for the alternative job. (Such data however will probably be rare..)
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Metals
- Alloys
- Steel