Skip to main content

Table 6 Comparison of benchmark and optimal calibration: IA survival probabilities

From: Evaluating search and matching models using experimental data

Parameters

British Columbia

New Brunswick

 

Benchmark

Optimal

Benchmark

Optimal

Annual discount factor (β)

0.82

0.82

0.82

0.82

Elasticity of search costs (z)

1.85

1.93

1.85

3.36

Re-calibrated search cost (c a )

0.185

0.176

0.079

0.033

Simulated versus actual control group (p-values)

    

12

0.033

0.033

0.942

0.942

36

0.171

0.171

0.984

0.984

53

0.298

0.298

0.332

0.332

All months

0.364

0.364

0.576

0.576

Simulated versus actual program group (p-values)

    

12

0.110

0.010

0.001

0.202

36

0.675

0.825

0.001

0.244

53

0.195

0.426

0.001

0.394

All months

0.333

0.994

0.001

0.998

Simulated versus actual impact (p-values)

    

12

0.881

0.551

0.001

0.359

36

0.501

0.261

0.001

0.424

53

0.883

0.841

0.001

0.929

  1. Note: p-values for single months are based on t-tests that the actual fraction still on IA minus the model prediction is equal to zero. For the all months test, we use a log-rank test that the actual (pooled) exits from IA are equal to the predicted (pooled) exits for the model and the data. The optimal parameter values are those with the maximum p-value for the log-rank test on the program group. The statistical tests treat the model predictions as constants