Inatennismatchwithplayers a and b, let ab betheprobabilitythat a defeats b. The BradleyTerrymodel is log(ab~ba) = a b, with anidentifiabilityconstraint,suchas c = 0 for player

Inatennismatchwithplayers a and

b, let Πab betheprobabilitythat a defeats

b. The Bradley–Terrymodel is log(Πab~Πba) = βa − βb, with anidentifiabilityconstraint,suchas βc = 0 for player

c. For T players,thismodels T(T −

1)~2 probabilities by (T − 1) parameters. Fitthemodeltothe Tennis data fileatthebook’s website,whichshowsresultsforDjokovic,Federer,Murray,Nadal,andWawrinkabetween January2014andJanuary2018.(Eachlineofthedatafileshowsthenumbersofwinsand losses formatchesbetweentheplayerwithindicator+1andtheplayerwithindicator −1. Using an R modelstatementintheformat glm(won/(won+lost) ∼ -1 +Djokovic+ ⋯ + Wawrinka,

...) suppresses theinterceptterm.)EstimatetheprobabilitythatFedererbeatsMurrayand compare itwiththesampleproportion.