An Explanation of KenPom and the Algorithm We Used to Predict the Results of the Games
Okay, what’s KenPom? KenPom.com is a site founded by a guy named Ken Pomeroy (clever name for the site!) that takes regular old statistics, and uses formulas to generate these statistics into predictive measures. These predictive measures also help you to compare how one team might stack against any of the other 350 NCAA Division I Men’s basketball teams.
Pomeroy explains his ratings further here. A lot of this math stuff is based off of the stuff put up on his site, so it’s worth checking out.
I’m going to use some advanced statistics to see what’s going on here. I’ll utilize some formulas to illustrate just what I’m talking about as I go along- let’s say in this example we’ll take Wednesday’s upcoming matchup with Georgetown taking on DePaul on the road, and use those as our two teams to better differentiate between what I’m talking about. Now let’s get started.
For each team, I’m going to find the projected offensive rating for each team. To do this for each team, I have to find their respective AdjO, (short for Adjusted Offensive Efficiency, which is that team’s points scored per 100 possessions, adjusted for opponent) and add the opponent’s AdjD (short for Adjusted Defensive Efficiency, which is points allowed per 100 possessions, also adjusted for opponent). From there, I’ll subtract the overall national average Adjusted Offensive Efficiency, which is all 351 D1 NCAA teams’ AdjO averaged out.
To use the Georgetown-DePaul example, here’s what this looks like as a formula.
Gtown projected offensive rating: Gtown AdjO + DePaul AdjD – Natl Avg. AdjO
DePaul projected offensive rating: DePaul AdjO + Gtown AdjD – Natl Avg. AdjO
Next, we have to find the projected tempo of this game as each game has a different tempo based upon which teams play in it. This means that the projected points scored could be more or less based on if the teams combined for a more uptempo and downtempo game. To find the projected tempo, we take the two teams’ AdjT (which is short for adjusted tempo, which is the possessions per 40 minutes for that team, adjusted for opponent). We take the two teams’ AdjT, add them together, then we subtract the national average AdjT (again taken from the average AdjT of all 351 Division 1 teams). Here’s what that looks like using the Georgetown-Depaul example for the formula.
Projected Tempo of Georgetown-DePaul: Gtown AdjT + DePaul AdjT – Natl Avg. AdjT
Now here’s the fun part where we put it all together! To predict the projected points for each team, we take their respective projected offensive ratings, multiply that by the projected tempo for the game, and finally divide that number by 100. Fun stuff! Here’s what that looks like using Georgetown-DePaul as our two teams.
Georgetown’s projected points: (Georgetown’s projected offensive rating * projected tempo)/100
DePaul’s projected points: (DePaul’s projected offensive rating * projected tempo)/100
Of course, this doesn’t account for home-court advantage. Not that DePaul has such a thing! Jokes aside, we should still calculate it with home court advantage. We simply add 1.4 points to DePaul’s number (since they have “homecourt advantage”) and subtract 1.4 from Georgetown’s (since they’re the road team). If the game’s a neutral site, we don’t need to calculate anything further than the projected points. Here’s what the revised scores as a formula looks like using Georgetown-DePaul as our two teams.
Georgetown’s projected points: ([Georgetown’s projected offensive rating * projected tempo]/100) -1.4
DePaul’s projected points: ([DePaul’s projected offensive rating * projected tempo]/100) +1.4
Before we get back to the post, a few notes on the numbers. When we look for the National AdjT or AdjO, there might be some variation day to day because of the different teams that are playing that day, even when there’s not a Big East game. Also, even when there isn’t any Big East games being played, there can also be variation with any of the different advanced statistics we used, as things can shift as metrics like AdjO and AdjD are balanced against the national field for sake of comparison. However, the time when I ran the simulations (after all games had ended on Sunday night) and when this post is up shouldn’t fundamentally change too much, especially since there won’t be any Big East games until Tuesday night.
Okay, now to resume back where we were, click on the button below.
[…] third place, we would actually place in fifth because of how the tie-breaker rules work, something I explain more in depth here. That’s not to mention if Seton Hall also gets into third, then we’re probably at sixth and […]
[…] Tournament kicking off today in Madison Square Garden, it’s a great time to look back at how the end of season predictions we generated using an algorithm for the last nine games of the Big East conference slate fared, and then to again […]