search

PSLmodels / tax-microdata-benchmarking

A project to develop a benchmarked general-purpose dataset for tax reform impact analysis.
https://pslmodels.github.io/tax-microdata-benchmarking/
2 stars 6 forks source link

Testing weight-deviation penalty when we reweight #114

Open donboyd5 opened 2 months ago

donboyd5 commented 2 months ago

@nikhilwoodruff, FYI @martinholmer

Reminder that we'll want to start imposing a weight-deviation penalty in the reweighting algorithm. My last successful run of make data, using current master, has this for s006 and s006_original -- largest weight before reweighting was 9675 and after reweighting largest was 70664, which is dramatic and I think we want to prevent it. I know we've got tighter bounds about to be implemented on weight ratios, which will probably solve the problem so we can discuss impact of that first. Simply opening this issue to make sure we put some effort into analyzing and addressing this as we think best.

image

martinholmer commented 1 month ago

@donboyd5, Given the results of PR #148, there seems to be less reason to experiment with a positive value of the REWEIGHT_DEVIATION_PENALTY. But, after PR #148 is merged, it would be interesting to do this experiment: use the master branch and make one change (changing the PENALTY value from 0.0 to 0.2) and use the method illustrated in issue #146 to see how the 2023 tax expenditure estimates change when imposing the penalty.