In early October, Minnesota had it's second state-wide recount in 40 years. When the vote difference between two candidates is less than 0.5% of the total votes cast in that race, state law demands a complete hand recount.
I observed the recount in my county as a representative of Citizens for Election Integrity (www.ceimn.org), and the county provided me with a printout of the county-wide results. The race in question was a non-partisan down-ballot race for Judge--and the second-place finisher for the race in the primary earned a ticket slot on Nov. 4.
I'm an Excel wonk and math geek; I couldn't resist extending the results of the Primary recount to the upcoming recount for Senator.
The two judges candidates were competing for second place on the ballot; the judges lumped the first-place finisher, the undervotes, and any finishers after 3rd together, so I couldn't calculate anything based on the entire vote.
Here are the results of two contested elections, and my calculations. "J" indicates "judge"; S indicates "senator"; "Est" is "estimate", "Diff" is "difference".
Initial Primary results:
HiVotes_J LoVotes_J TotVotes_J VoteDiff_J % diff_J
3329 3230 6559 99 1.51%
Here are the errors in the primary found during the recount:
Abs Error_J HiChange_J LoChange_J
16 +1 +5
Abs Error is the count of errors: either an overvote or an undervote counts as 1
HiChange is the net vote change (1 overvote + 1 undervote = net change of zero) for the race with the initially higher vote; LoChange is the net vote change for the race with the initially lower vote.
Here are the currently posted Senate results:
HiVotes_S LoVotes_S TotVotes_S VoteDiff_S % diff_S
1211565 1211359 2422924 206 0.00850%
If votes were pennies and you doled them out to Franken and Coleman one at a time, both of them would have $100 -- AND CHANGE! -- before Franken oops, Coleman!!(and how do I do strikeout?) would be a full penny ahead!
I decided the fairest way to make a prediction would be to assume the error rates would not change from one race to the next.
Determining the expected error for the Senate race is easy:
EstAbsError_S = AbsError_J x TotVotes_S / TotVotes_J
We can expect 5,910 counting errors; with 2.2 votes added for each one vote subtracted.
The interesting part is to figure out who will gain (and how much).
It seems reasonable to treat Franken and the lower-drawing Judge similarly, as both finished the initial election with fewer votes.
If I multiply the two primary Net Changes by the initial Senate vote divided by the initial Judge vote, I get this:
EstHiChange_S = HiChange_J x HiVotes_S / HiVotes_J = 364
EstLoChange_S = LoChange_J x LoVotes_S / LoVotes_J = 1875
In other words, Franken would pick up 1511 votes more than Coleman, and win by 1305 votes.
Of course, the closer race for Senate might divide the errors more equally--and Franken might pick up only 975 votes to Coleman's 900--and still lose after all votes are counted.
FWIW, my county voted 102696 for Coleman; 85287 for Franken, 34066 for Barkley, and 1819 other. It did go for Obama, but backed Kline, and my district kicked out my much-loved state rep, Shelley Madore.
And yes, if you have the time, please contact www.ceimn.org and volunteer to observe the ballot counting!