The Collins twins’ cautionary tale for Florida

[InstiGATOR1]

From Ken Pomeroy:

Here are the five most dominant seven-game stretches in conference play since 2000, ranked by aggregate margin of victory:
05 Louisville 214 (Games 2-8)
11 Belmont 210 (Games 2-8)
00 Stanford 207 (Games 9-15)
13 Florida 198 (Games 1-7)
04 St. Joe's 183 (Games 3-9)


While Florida’s run has been impressive and unusual, it’s not unprecedented. Even with this run, they will probably lose again in regular-season play, maybe even twice. When that happens, there will be some chatter about what’s wrong with Florida. But taking a longer view, they should still be expected to lose a game. (My system has the Gators as better than 50/50 to go unbeaten in SEC play, but it’s working with an unrealistically high rating.) So there may not be any reason to freak out when they do.

Here is the link for the rest:

http://kenpom.com/blog/index.php/web..._2000_stanford

[Go2gtr]

Very rational.

[iam4uf]

You can already see that the UF players are getting sloppy. They feel like any obstacle can be overcome, but end up creating some of those obstacles. This has been a fun stretch, but a couple of losses might refocus the players (except Boynton who seems to be zeroed in on the Final Four.)

[rserina]

Oh, we will freak out.

[themistocles]

Interesting, but, he makes some rather gross errors in this piece, see below. And, of course, every game is a unique phenomena, as is every season, every conference and out-of-conference experience, every conference tournament and every NCAA tournament. So, saying: "I'm not saying the Gators are going to flame out or lose in the round of 32, but..." is, of course, the correct thing to say, because saying anything other than that would be idiotic.

Personally, I don't buy much into the Streak concept as so many do, but curiously, he was adding up 8 games for Stanford (didn't figure it for the others), and only 7 for the Gators. Seems a bit unfair when you take the total sum to give them an extra game, and not only that, but a game they won by 43 points (111-68).

If you look at mean victory margins, then, both Stanford and UF averaged 28 difference in that, however, if you use a more stable estimate, that is not as influenced by things like Stanford's 51 and 43 point outlying victories, you note that UF has average 31 point victories, with the closes being 17 points and the greatest 39; while Stanford averaged 26 points with the low being 10, and 2 others at a mere 15, and the highs being 51, 43 and 37 (nothing else anywhere near those). Those three games substantially increased their scoring margin.

As an experienced statistician, I always eliminate outliers (like the SC 39) and the Stanford 51. Those substantially alter the results which is the reason the Gator median is 5 points higher, and if he had given the Gators 8 games, they would have totaled 28 during 8 games at their current rate.

[GatorLurker]

He has been saying our rating is unrealistically high for a long time. Either his system is broke or we are WAY better than he thinks that we should be. I think his system is fine. These guys have been playing lights out.

[GatorLurker]

Originally Posted by rserina

Oh, we will freak out.

LOL.

I avoid NBN for a couple of days after a loss. Too much "drama".

[InstiGATOR1]

Originally Posted by themistocles

Interesting, but, he makes some rather gross errors in this piece, see below. And, of course, every game is a unique phenomena, as is every season, every conference and out-of-conference experience, every conference tournament and every NCAA tournament. So, saying: "I'm not saying the Gators are going to flame out or lose in the round of 32, but..." is, of course, the correct thing to say, because saying that would be idiotic.

Personally, I don't buy much into the Streak concept as so many do, but curiously, he was adding up 8 games for Stanford (didn't figure it for the others), and only 7 for the Gators. Seems a bit unfair when you take the total sum to give them an extra game, and not only that, but a game they won by 43 points (111-68).

If you look at mean victory margins, then, both Stanford and UF averaged 28 difference in that, however, if you use a more stable estimate, that is not as influenced by things like Stanford's 51 and 43 point outlying victories, you note that UF has average 31 point victories, with the closes being 17 points and the greatest 39; while Stanford averaged 26 points with the low being 10, and 2 others at a mere 15, and the highs being 51, 43 and 37 (nothing else anywhere near those). Those three games substantially increased their scoring margin.

As an experienced statistician, I always eliminate outliers (like the SC 39) and the Stanford 51. Those substantially alter the results which is the reason the Gator median is 5 points higher, and if he had given the Gators 8 games, they would have totaled 28 during 8 games at their current rate.

I would NOT say Pomeroy made any error gross or not.

I did notice their 51 point outlier particularly and that two games account for almost half the Stanford total. I am not sure that SC is as much an outlier for UF as the biggest margin after all UF also had 31 and 33 point margins during this run of games. So I would not remove the largest margins with a sample size this small. I would just note them.

[themistocles]

Originally Posted by InstiGATOR1

I would NOT say Pomeroy made any error gross or not.

I did notice their 51 point outlier particularly and that two games account for almost half the Stanford total. I am not sure that SC is as much an outlier for UF as the biggest margin after all UF also had 31 and 33 point margins during this run of games. So I would not remove the largest margins with a sample size this small. I would just note them.

The gross error is comparing total score differences across 8 games rather than 7.

Any way you look at it, adding 8 and comparing with 7 has to give an unwarranted advantage to the 8 (Stanford).

Also, people are always afraid of eliminating cases, for some strange reason. I learned as a professional gambler, that if you include the extreme cases in your statistical analyses, you end up overestimating you expected, or average value. In gambling, where the average value estimate determines the profit margin and strongly influences one's decisions regarding where to put one's money, and, because the outliers are always big payoffs as compared with say median payoffs (50% above, 50% below - a legitimate way to estimate the center of a distribution of values) this is a very important thing to pay attention to.

In statistics classes, most of which are horribly taught by people with at best limited empirical experience, the mean is overly emphasized and students are rarely made aware of "better" options for many situations, like a trimmed mean, Winsorized mean or other trim - the median is a 50% trimmed mean.

The thing about Stanford is that their streaks scores were all over the place, while the Gators have exhibited enormous consistency Thus, it is not particularly surprising that they lost some games and managed to lose early in the NCAA tourney

Take, for example, ranges of Stanford and Florida's scores and their opponents scores:
Low - High
67 - 111 - Stanford
64 - 83 - Florida

Opponents
50-68 - Stanford
36-52 - Florida

Stanford's range was 42 points, the Gators, 19
Stanford's opponents range was 18, the Gators, 16

So clearly, the Gators are more consistent.

Also, the Gators are a far superior defensive team, at least from judging by scores in these respective streaks.

While Stanford averaged 85 points per game, and their opponents, 57, the Gators averaged 75 points per game and their opponents 46.4. Taking the medians for those shows the effects of those two extreme cases on Stanford's scores (101 and 111) - Median - Stanford, 80; opponents - 57 (same); Florida 75 (same), opponents - 47 (same).

This reflects on of the huge problems with the modern love affair with the arithmetic mean (modern, from about 1930 to today).

I speak to this somewhat, albeit to many other things on my badly needing updating "Statistical and Other Myths" web page: http://www.freewebs.com/tedstats

[HALLGATOR]

It doesn't take any great predictive powers to say a team will probably lose or they will probably lose again. Odds are in your favor that that's exactly what will happen. After all how many teams have ever went undefeated through an entire season.

[SavageGator]

I'll be surprised if we do not lose one or two games this regular season.