search

washingtonpost / elex-live-model

a model to generate estimates of the number of outstanding votes on an election night based on the current results of the race
48 stars 5 forks source link

replace point pred with bootstrap mean #102

Closed lennybronner closed 1 month ago

lennybronner commented 1 month ago

Description

We are replacing the unit level point prediction with the bootstrap sample mean.

This makes sure that the point prediction is always centered around the average of error_B_1, which forces our interval to be the error_B_2's (fake test error) deviation from that. This would make issues around weirdly shaped intervals easier to debug on election night.

Jira Ticket

https://arcpublishing.atlassian.net/browse/ELEX-4464

Test Steps

elexmodel 2017-11-07_VA_G --estimands=margin --office_id=G --geographic_unit_type=county --pi_method bootstrap --features baseline_normalized_margin --percent_reporting 10 --aggregates postal_code --aggregates county_classification