Mixed Proportional Hazard (MPH) Model
Expert TheoryReview
Evaluating dynamic exit rates from unemployment requires separating true policy treatment effects over time from compositional changes caused by individual baseline traits or unobserved participant differences.
Picture this
The model breaks down an individual's probability of leaving unemployment into three distinct multiplying parts: a baseline time clock, a score based on known background factors (like age and past work), and a hidden personal factor that accounts for unobserved drive or luck.
What the evidence says
The MPH framework isolated a 20% to 40% time-varying increase in unemployment escape rates attributable to the policy regime, preventing bias from a small sub-population (1-2%) of unobserved 'immune' individuals with zero exit probability.
- Who was studied
- N = 4,513 unemployed individuals in two Danish counties (2,393 treatment, 2,284 control).
- How
- Parametric hazard estimation with discrete unobserved heterogeneity distributions (two-type mixture model) and duration-varying treatment indicators.
What to do
Estimate duration models using a mixed proportional hazard specification with unobserved heterogeneity controls when analyzing dynamic panel data from social experiments.
From the source
"We shall begin by making an assumption of a mixed proportional hazard (MPH) model, that is, the hazard rates can be written in the following form: \theta_1(t|X,V) = \lambda_1(t) \cdot exp[f_1(X)] \cdot V"
bc5b372e-5470-4162-9864-568cdcec4a00-Experimental Evidence on the Nature of the Danish Employment Miracle.pdf