[furtiv3]

Looking at Sagarin's power rankings, for example, how would go calculate the implied win probability of two teams listed if they were to play each other?

Is it just a matter of adding the two power ranking totals together and dividing each individual ranking by that total?

[James Marques]

Sport would help. Honestly, though, it's not all that difficult I don't think. Let's say you're working with NBA lines. Go back and look up the point spreads for the last 5 years or so, and then figure out what percent of the time the favorite won. If you plot the data in Excel, it should follow a logistic regression (low spreads --> around 50% win probability, big spreads --> 100% win probability). Once you calculate that, you can take the 2 Sagarin numbers and subtract them (and home field/court advantage, of course). That number will be the predicted point spread. Now, plug that number into the regression you just made, and you have an approximate win %.

Teamrankings.com should have exactly what you need. (At least, it did for my NFL and NCAA models).

[furtiv3]

Thanks for the quick reply. I'm specifically looking for straight estimated win probability if they were to meet heads-up. Example, if Duke vs. Arizona in NCAA basketball were to meet based on the season ending power rankings from the different sources.

[James Marques]

Gotcha. Well, from what I pulled off Team Rankings, since 2003 the data for point spread vs favorite win % follows the equation:

FavWin% = 0.15672 *ln(S)+0.47179 -- where S is the point spread (positive) for the favorite team

The R-squared value for this fit line is pretty good too at 0.9481. Might be good for a model.

So let's say Arizona is -3.5 against Duke if they were to play one another in the tournament. As a 3.5 point favorite, Arizona would have a win probability of 0.6681, or 66.8%. (This sounded high to me, but I looked it up and 3.5 pt favorites have won 64% of games favored since 2003, 1114 - 631 overall)

Is this what you were looking for?