Immune Subpopulation in Unobserved Heterogeneity

Picture this

In a group of people trying to exit a building, a tiny subgroup is completely unable to move. Assuming everyone moves at varying speeds distorts speed calculations unless modelers explicitly separate the immobile subgroup from the rest.

What the evidence says

Model estimations consistently identified a distinct subpopulation of 1% to 2% 'immune' individuals exhibiting a zero exit rate from unemployment, whose separation was required to accurately identify policy impacts among the remaining population.

Who was studied

N = 4,513 unemployed insured workers across Storstrøm and Southern Jutland counties, Denmark.

How

Maximum likelihood estimation of a two-type discrete mixture model for unobserved heterogeneity within a mixed proportional hazard framework.

What to do

Specify discrete mixture distributions allowing for zero-probability exit fractions when estimating labor duration models containing structural non-responders.

From the source

"the maximization of the log likelihood invariably ends up preferring a specification with two unobserved 'types', one small group (1-2% of the inflow into unemployment) of 'immune'persons with a zero exit rate from unemployment, and the other a large group with a positive exit rate from unemployment."

bc5b372e-5470-4162-9864-568cdcec4a00-Experimental Evidence on the Nature of the Danish Employment Miracle.pdf

Tags

• unobserved heterogeneity
• immune subpopulation
• duration models