# predicting change in probabilities

I need some help in try to estimate some probabilities. I am trying to find the amount of home field advantage given when it comes to the pre game probability that the home team will when the game.

In other words if the bookmaker gives teamA a .25 chance of beating teamB when teamB is the home team what would the probablity of teamA beating teamB if teamA was the home team or if the game was on a neutral field?

I have a few seasons worth of odds data from which I can compute the probability of the outcomes going in the game. For every possible matchup in the league I took the average of the away probabilities and the average of the home probabilities and filtered out the duplicate games. This is an example of the data.

GameId   AwayWP     HomeWP
1      0.403941   0.687718
2      0.369619   0.686870
3      0.386640   0.515152
4      0.448006   0.671919
5      0.394960   0.605040
6      0.277475   0.541345


Obviously a linear model wont work well. I thinking I will need to use logarithms some way but don't know how.

• it might be better to work with scaling of odds (a shift in log-odds). but note that betting odds contain a margin that don't reflect actual probability estimates. – Glen_b May 29 '15 at 2:29
• assuming the margin is applied equally to both sides it is easy to strip it out – user2333196 May 29 '15 at 2:58
• I don't believe that assumption is correct; shorter odds are typically "trimmed" to a lesser extent than longer odds. – Glen_b May 29 '15 at 3:18
• i don't disagree – user2333196 May 29 '15 at 3:29
• Can you explain your data more. If you are averaging games what is GameId X?? – adam Jun 9 '15 at 15:13