The problem I want to solve is, given a free sports bet (i.e. if we lose we get our money back) and given no special knowledge of sports betting on the part of the player, what selection would have the most +EV odds to choose to bet on.
TL;DR part, skip to last paragraph if you want
Instinctively, if we bet our (e.g.) 10 dollar free bet at too low odds, say evens, even with a genuine 50-50 shot, we don't take the free money often enough and our EV is $5....
EV of loss = $0, probability of loss = about 0.5
EV of win = +$10 probability of loss = about 0.5
total = about +$5
With odds of 2/1 (+200 in US odds) and a 30% chance of winning, our EV is $6
EV of loss = $0, probability of loss = 0.7
EV of win = +$20, probability of win = 0.3 EV =$6
even though the bookie has vig in the second example, but not the first, our profit is higher because the free bet adds to our EV more often.
Over a certain point this factor will be less important (we're getting the free money almost all the time anyway) and the fact that "long shots never win", or rather they don't win often enough to justify their odds, will become more important.
J Dowie in "Efficiency and equity of betting markets" (1976), using data from the 1973 UK flat horseracing season gives the following figures showing how bookies' vig increases as odds increase (i.e. long shots are generally poor value).
Odds ...... winnings/wager % (i.e. percent of money returned to bettors as winnings)
Odds on ..... 99.2%
Up to 5/1 ... 90.0%
Up to 10/1 .. 89.4%
Up to 16/1 .. 80.3%
All ... 60.5%
Given that the EV of normal bets drops off at odds above 10/1, we are better off using our free bet at e.g. an 8/1 shot with only a 10% chance (EV $8), than an 11/1 shot with a 6.67% chance (EV $7.33).
So that suggests we are best off using our free bet on a selection with odds in the high single figures.
Hyung Song Shin in his paper
http://www.math.ku.dk/~rolf/teaching/thesis/Shin91.pdf
quotes Dowie and gives formulae to calculate real probabilities given the odds in a sporting event. It would be better to use this than the 1973 data, but I'm forced to admit that the maths is over my head until I see a worked example (my maths is like that, once I've seen it done with real numbers I can reproduce it, but not before).
end of TL;DR
Does anyone have access to a worked example of calculating Shin probabilities from the odds for a real sporting event?
If not then for example this weekend there is an English soccer match with odds of: Middlesbrough 1.4/1, West Ham 2.1/1 and Draw-Tie 2.2/1 Those probabilities add up to 105.17%, can someone show me how Shin would take the overround off to find the probabilities the bookies are working with in this simple 3-runner event?