Neal (1995) examines the wage-cost of switching industries among displaced workers. Neal finds evidence that suggests the wage-cost of workers switching industries is highly correlated with both work experience and tenure in their previous job. Workers who switch industries and have high industry-specific capital are faced with lower wages in general in their newer job, and for workers who remain in their predisplacement industry, firm-specific human capital does not play as large a role in wage determination as does industry-specific human capital.
The largest implication of this paper on my research so far is that industry might play a large role in determination of how long one might stay in the job search following termination. Industries that in general have more competitive job markets might influence individuals to stay in the job search longer than their counterparts, as they would not risk switching to another industry and incurring lost wages. Another insight is that a focus on firm-specific human capital might be too narrow, and industry-specific human capital should be included as well.
With regards to problems of the classic OLS regression, the largest problem might be the cross sectional data that I have (heteroskedasticity), and the possibility of selection bias. To get around this problem, Neal uses a variant of Heckman's (1979) two stage procedure to produce the corrected versions of his regressions and thus control for selection bias. Neal continues by estimating selection equations and is able to effectively control for any possible selection bias. It is possible that I could correct for the cross sectional issue by using a time-series variant of my data, but it'll require some looking in to.
References
I'm a bit confused on your plans for trying to reduce heteroskedacity in your model. You claim that Neal used Heckman in order to reduce selection bias. However, you claim to have cross sectional issues and possibly selection. So while I understand using Heckman to counter selection bias, your solution seems to be adding more cross sectional data which in turn also leads to more heteroskedacity. I'm not sure if it'll make the model less overall, but you should think about it.
ReplyDelete