How Well Do Average Efficiencies
Predict Game Outcomes?

Shortly after the impressive victory by UK over a strong West Virginia basketball team, less than 24 hours after Kentucky had laid an egg in losing to a strong Iowa team, “Fiddling,” a regular poster on the UK Basketball Internet Message Boards asked, “Sometimes there are games which go against your numbers. Last night was early in the season (fewer numbers to compare), but I'm sure there are sometimes similar results later in the season. Have you done any research into why these occur? ie injuries, matchups, etc?”

Great question, one that I have given considerable thought to over the years.

I start from the premise that no team plays at the same level every game. Some games the team will play better than its average performance and others not as good as its average. The numbers I report are averages, and in round terms, 50% of the time the performance will be higher and 50% of the time lower than average. The distribution of the actual variable results around an average is measured by the standard deviation and variance. For Kentucky , I have tracked these parameters for offensive and defensive efficiency since the beginning of my work.

By comparing the average efficiencies of two opponents I can identify the average spread between them, and that average spread forms the fundamental basis for my game margin predictions.

Some nights both teams will perform better than average [for them] and the spread will be about as predicted. Other nights, both teams will perform poorer than average, and the spread will be about as predicted. The Iowa game is an example of the latter and the WVU game is an example of the former.

Some games, however, one team will perform better than its average while the other team plays poorer than its average. If the better team has the better performance, a blowout victory, or an embarrassing loss occurs. As in, I thought we would lose, but I never imagined losing to X by 25 points. If the weaker team has the better than average performance, the stage is set for an upset, in which the inferior team defeats a clearly better team.

I have found that "upsets" like this occur in about 20 to 25 percent of all games, whether regular season, or post season.

The likelihood that any particular team will be subject to an upset is measured by the Net Game Efficiency. That is why I have concluded, and firmly believe, teams that establish high Net Game Efficiencies will compete for national honors; they are simply less likely to be upset victims when the lesser team gets hot, and they are not.

Hope this help you understand the method to my madness.

The early season paucity of real data does decrease the reliability of a specific single game analysis, but not as much as you might expect. Last season, I predicted 34 games, missing 3 outcomes [not margins] and the first "miss" was the Louisville game [not real early season] and so far this season I have picked the winner in all 4 games.

With respect to margin, my data can be confusing because I track these predictions by home team - visiting team, rather than Kentucky v Opponent. Last year, my average prediction, home - visitor, was 75-67, margin 8 points, and the actual average results were 71-62, average margin 9 points. My current model tends to under predict scores, but tends to ID game margins fairly well. I continue to refine my model and my data base grows. Through 4 games this year, the predictions average 69-62, margin 7 points, and actual results have been 71-65, margin 6 points. Perhaps I have overcompensated for the under predictions last season, but I need more data.

I know you thought you asked a simple question that deserved a simple answer. However, I want to make one more point about the variable performances we see, and fans, of which I am one.

Fans tend to place entirely too much significance on the extreme performances, good or bad, than they deserve during the regular season. Fans should focus upon, and understand their team's average performance level if the fan wants to have a realistic view of his/her team's real competitiveness. I am as guilty of this as any fan. We saw this at work 24 hours ago after the Iowa game, in which UK 's offensive performance was dismal. What was lost was that either Iowa 's offensive performance was equally dismal or UK 's defensive performance was really quite solid. Last night, we saw the opposite, UK 's offensive efforts was phenomenal, and today the fan chatter is shifted to the positive again. However, what we are missing from last night's game is that WVU's offense was also quite strong against the highly praised UK defense or UK 's defense was quite pathetic.

I have not directly addressed the questions of match ups and injuries. I do not consider individual match ups in my analysis, and I do not consider the individual make up of the teams. My theory believes that a team is a team, and the average numbers will reflect that team's composition and how the coach uses that composition in games. That is why I spend little time worrying about whether Mr. Morris will return, or when, or who starts, or how Tubby substitutes. Certainly, these factors probably do affect the average numbers, but the average numbers are my focus. If those average numbers are not satisfactory to a coach, he will change what he does to improve them.

I have only found one demonstrable case in which a player injury changed the average numbers. When Kentucky lost Derrick Anderson in 97, the averages for the team dropped noticeably, and established a new average for the team. That drop was the difference, in my opinion, between winning in 97 and not. One other situation made a material change in the average numbers for Kentucky , in 2002-03, following the embarrassing loss to Louisville . Tubby obviously was not satisfied with the averages for his team to that point, and he made changes that impacted the numbers in a material way. Other than these two specific case, I have not noticed any injuries or match up issues that affect the numbers.

I admit that my view of variable performance probably masks some of these impacts, but this is my approach.

I really appreciate your question, and am honored to share my views with you and others.

The averages tell the real story for a long term perspective. Upset do occur, and will occur. It is this fact of sport that provides the underdog eternal hope and the powerful constant doses of humility.


Copyright 2004 Richard Cheeks
All Rights Reserved