0
$\begingroup$

I have a logit model with my dependent variable being "probability of exiting unemployment" and one of my independent variables is "individual duration of unemployment". Obviously there is a huge problem of endogeneity between these two variables, and I have been struggling to think of a possible instrumental variable to use to address this endogeneity issue. Can anyone suggest any possible instrumental variables that may be used, or is the endogeneity problem just too massive between these two variables to be solved?

Advice would be much appreciated!

$\endgroup$
1
  • $\begingroup$ This post might be broadly useful for potential sources of instruments. Typically, instruments will be "local", so without knowing more about your institutional setting (country, age, industry, etc.), people will be unable to help. $\endgroup$
    – dimitriy
    Jan 25, 2018 at 2:43

1 Answer 1

0
$\begingroup$

I don't see the endogeneity here. probability to exit unemployment in this period should not feed back into the length of unemployment (in the past).

Instrumental vars (IV) are just one way of dealing with endogeneity. lagged regressors is another one. however, in your case I don't see the need for the lags since the length of unempolyment is already in the past, and is separated from the probability in question by today.

what's questionable to me is the measurement of the probability. it's not observable on the individual level, while the length of unemployment is observable for an individual. so, it's not clear to me what is your dependent variable, and how do you obtain values for it.

$\endgroup$
2
  • $\begingroup$ The OP's question seems like a meaningful one to me. Imagine you observe 1000 unemployed people with varying spells of unemployment. Some got fired yesterday, others have been looking for work for years. All else equal, do longer spells lead to a lower probability of exit because skills and habits atrophy with disuse? Or is there some sort of selection going on where longer spells happen to worse workers that no one wants to hire, so duration is merely a proxy for low productivity and has no causal effect on its own? $\endgroup$
    – dimitriy
    Jan 25, 2018 at 2:56
  • $\begingroup$ @DimitriyV.Masterov the question is meaningful, of course. I don't doubt that X can cause Y. What I'm questing is how Y feeds back into X in her case and how does she measure Y. The latter is important because it can explain why she may expect Y->X relation. Also if the probability is persistent, so that today's probability is very much like it was a year ago, then a probability a year ago clearly impact the length of unemployment to date. However, in this case there's a specification problem, in my opinion. $\endgroup$
    – Aksakal
    Jan 25, 2018 at 3:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.