[Juret]

If scoring for the Home and Away team on a significant level are negatively correlated, what adjustment would you look to make to your model's predicted spread? Deriving win probabilities from a bivariate discrete distribution, the scoring for Home and Away are assumed to be independent. However, in my case, they are clearly not. Thoughts?


Thanks

[Justin7]

What sport are you analyzing? In most sports, I though scoring of two teams was positively correlated.

[mathdotcom]

It can go either way. In football id guess theyre positively correlated -- you can increase your expected offense per play by choosing risky plays but at the cost of more turnovers.

I wouldnt be surprised if theyre -corr in soccer where they seem to be able to close the door relatively easily with the lead.

[Juret]

I am analyzing a minor Nordic sport called bandy, think field hockey but on ice. The Pearson correlation is -0.236 (p=0.000).

I read once that in NHL, scoring is slightly negatively correlated. I haven't tested that though.

[That Foreign Guy]

What sport are you analyzing? In most sports, I though scoring of two teams was positively correlated.

Only games where tempo affects totals (basketball is the most extreme example I think). Every time I score I am giving you a chance to score so the more I score the more chances you have.

Games with a smaller tempo effect scoring is more negatively correlated. At the other extreme, take snooker which is played best of 15. If I win a frame you can't win it.

[princecharles]

I've heard it described as the, 'Holding Serve' principle.

[That Foreign Guy]

I thought about tennis as an example, but sports with win by 2 rules actually introduce some positive correlation when players are more evenly matched. Small effect to be sure, but I started over thinking my reply.