bayes_vote_transfer
Use bayesian linear regression to find how many voters for each candidate in first round of election, voted for candidate X in second round? Based on real data from each poll station on Ukrainian presidential election-2019.
What are the assumptions of a model:
- all standart assumptions for bayesian linear regression
- there are different levels of support for each candidates in second round from different group of voters in first round (each group - voters who voted for some candidate) (uniforms as priors)
- turnout modelled for each such group as different parameters, too (student as priors)
- final share of voters for candidate Z from first tour, who voted for candidate X in second, is a multiple of level of support for candidate X and turnout for this group of voters: total_x * turnout
- finally, on each poll station, sum of all total_x[i] * turnout[i] * votes_from_first_round[i] == number of votes for X in second round, where i - index of each different group of voters, and votes_from_first_round[i] - rezult for group i on this station.
Result for Zelensky
Result for Poroshenko
Also, see here in Ukrainian: