Conversion from rankings to point spread

**Pokerjoe** · 09-26-10, 04:54 PM

There must be a relationship between the ranking difference and the chance of winning, right? Top ranked team has x chance of losing to 10th ranked team, and <x chance of losing to 20th ranked team and so on, right? If it has any accuracy at all, as you claim it does, this will be true.

So develop a formula for translating that ranking disparity into a win chance.

Translate the win chance into a pointspread through moneyline equivalencies.

However, mere ranking is problematic, in that the rankings will distribute themselves normally, so that the difference in ability between 40th and 50th will be less than the difference in ability between 1st and 10th, or 90th and 100th.

So stop using rankings, and start using ratings.

Good luck.

**Spektre** · 09-26-10, 05:23 PM

Actually, I guess ratings is the term I should have used. It is indeed a rating that I can use to rank teams.

The distribution of the rating definately does not exhibit the compression you mention.

One problem with the RATINGS is that the dynamic range changes from week to week, getting larger as the season progresses. I guess I can try normalizing the ratings to a set scale and see if there is a good correlation between the normalized ratings and chance of winning.

Thanks,

Spektre

**Pokerjoe** · 09-26-10, 06:43 PM

Ranking means you have teams in an order, best to worst.
Ratings means the teams have a number (by which you can rank, if you want to) which reflects relative strength.
If you have ratings which are already related to points, then I don't understand the question. You rate A 7 pts better than B, you compare that to the pointspread. Sagarin is a well-known example.
If you have ratings which aren't related to points, relate them to points, or win chance, and then win chance to points. But they must come from something, these ratings, so I don't see what the problem could be.
If they are someone else's ratings (there are public ratings which are not obviously (though still in fact) related to win chance) and you don't have their basic algorithm, then you'll have to do some math, but they still must be related to win chance/point spread (and the difference between win chance and pointspread is just math, they are just different expressions of the same thing).

**Spektre** · 09-26-10, 09:44 PM

Originally posted by Pokerjoe

Ranking means you have teams in an order, best to worst.
Ratings means the teams have a number (by which you can rank, if you want to) which reflects relative strength.
If you have ratings which are already related to points, then I don't understand the question. You rate A 7 pts better than B, you compare that to the pointspread. Sagarin is a well-known example.
If you have ratings which aren't related to points, relate them to points, or win chance, and then win chance to points. But they must come from something, these ratings, so I don't see what the problem could be.
If they are someone else's ratings (there are public ratings which are not obviously (though still in fact) related to win chance) and you don't have their basic algorithm, then you'll have to do some math, but they still must be related to win chance/point spread (and the difference between win chance and pointspread is just math, they are just different expressions of the same thing).

This is a rating, not an ordered list. Each team has an associated floating point number which should correlate to its relative strength. It is not related to points.

It is MY rating system, and I will have to backtest it vs. win frequency to see how well correlated the rating spread is to win percentage. I had a link somewhere that related win percent to spread. I'll look for it, but if I cannot find it, do you have one handy?

Thanks again,

Spektre

**Pokerjoe** · 09-27-10, 11:14 PM

Spektre, Justin7's book "Conquering Risk" has a chart for that, but also, here under "tools," I think Ganchrow has a calculator that will do it.