enCYCLONEpedia: Shut Up Stat Guy
- Thread starter ChrisMWilliams
- Start date
This isn't directed anywhere but at the general topic at hand, but one of my favorite quotes: "He uses statistics as a drunken man uses lamp posts- for support rather than for illumination." Andrew Lang
The most significant part of the article to me was the way that rankings make some differences look much bigger than they are, such as the difference between #5 and #15 might be much smaller than the difference between #1 and #5 (or even #1 and #2).
It is also interesting to pay attention to how we remember our team's previous seasons. All of us know what round of the tournament we lost in for nearly every year in recent memory, but how many of us know our ranking on any poll from even two years ago (or last year)? I don't.
It is also interesting to pay attention to how we remember our team's previous seasons. All of us know what round of the tournament we lost in for nearly every year in recent memory, but how many of us know our ranking on any poll from even two years ago (or last year)? I don't.
My only issue with stats in college basketball is the small sample size, especially when you focus on only conference games. At this point, only 14 conference games have been played. That means that a statistical outlier will have a major impact on the numbers. (ISU@WVU)
Advanced statistics make a lot more sense in baseball where you have a much larger sample size and tell trends can be established.
Advanced statistics make a lot more sense in baseball where you have a much larger sample size and tell trends can be established.
I'm a big fan of Haaland's statistical breakdowns. Someone that takes the time to do the analysis that he does gets my respect (not like he cares what I think).
My office is littered with armchair sports fans that want to regurgitate some ESPN analyst's opinion like it's gospel. I prefer a casual sports chat based on opinion, but when someone wants to try to blazenly cite opinion as fact, I relish in a good statistical head shot.
The fun part of watching any sport to me lies in the unknown stat, which is the athletes themselves. Vegas's success at setting lines over the years is remarkable to me because of that fact. We really never know when a normal 5 point scorer is going to explode with 25 points off the bench. We don't know how the rest of a team will react if their star player goes to the bench with 2 quick fouls in the game or completely goes out of the game due to an injury.
I know those instances lie outside of the standard deviation......and I guess that's why I watch. Well that and I'm an ISU fan. My whole fandom has pretty much based of hoping for things outside of the standard !
My office is littered with armchair sports fans that want to regurgitate some ESPN analyst's opinion like it's gospel. I prefer a casual sports chat based on opinion, but when someone wants to try to blazenly cite opinion as fact, I relish in a good statistical head shot.
The fun part of watching any sport to me lies in the unknown stat, which is the athletes themselves. Vegas's success at setting lines over the years is remarkable to me because of that fact. We really never know when a normal 5 point scorer is going to explode with 25 points off the bench. We don't know how the rest of a team will react if their star player goes to the bench with 2 quick fouls in the game or completely goes out of the game due to an injury.
I know those instances lie outside of the standard deviation......and I guess that's why I watch. Well that and I'm an ISU fan. My whole fandom has pretty much based of hoping for things outside of the standard !
KHall nails it when he says it's about parameters.
I see the value in KenPom. I see the value in the AP. I see the value in RPI. I see the value in Sagarin.
I think in terms of seeding a tournament according to a team's achievements the RPI or other non margin of victory computer polls like Sagarin ELO make way more sense than KenPom. The value KenPom puts on our ability to beat TCU twice vs our ability to blow them out makes it worthless in seeding a tournament, doesn't mean it's worthless for everything or making predictions though. Don't want to sound egotistical, but he mentions this in the story but in a way that might seem complex to people who don't follow the results of computer polls every week or know ISU's computer rankings off the top of their head.
He says to base it on evidence, it seems to me that Sagarin and the AP poll are both better at predicting what teams will excel in the tournament than KenPom's rankings. They're certainly both much more accurate after the season has ended the past few years.
I see the value in KenPom. I see the value in the AP. I see the value in RPI. I see the value in Sagarin.
I think in terms of seeding a tournament according to a team's achievements the RPI or other non margin of victory computer polls like Sagarin ELO make way more sense than KenPom. The value KenPom puts on our ability to beat TCU twice vs our ability to blow them out makes it worthless in seeding a tournament, doesn't mean it's worthless for everything or making predictions though. Don't want to sound egotistical, but he mentions this in the story but in a way that might seem complex to people who don't follow the results of computer polls every week or know ISU's computer rankings off the top of their head.
He says to base it on evidence, it seems to me that Sagarin and the AP poll are both better at predicting what teams will excel in the tournament than KenPom's rankings. They're certainly both much more accurate after the season has ended the past few years.
Last edited:
Thanks for all the kind words and for reading, guys. I really appreciate it.
I think that is a fair point worth noting. I like to pare it down to conference games solely to eliminate fluff from the non-con schedule. That could be done selectively too, it's just a whole lot easier and it avoids in questioning of why some non-con games were excluded and others weren't. It is kind of just a fact of life though as far as sample sizes usually being too small, statistically speaking.
I don't know which is better at predicting teams that excel in the tournament. You could probably go back in history and try and deduce that. The thing there is that so much comes into play with how a game results. Matchups, shooting performance, did the point guard's girlfriend cheat on him while he was at the tourney site? All of those variables are so impossible to predict and account for and then re-apply to your predictions that it is the main reason I don't even try to predict games based off of a computer model.
I'd have just as likely of a chance of colonizing on the moon by stealing my daughters' Radio Flyer wagon and plastic play house, you know? I'll let the Vegas guys try and do that, I mean, I am on limited resources here!
Thanks for the comments.
I think that is a fair point worth noting. I like to pare it down to conference games solely to eliminate fluff from the non-con schedule. That could be done selectively too, it's just a whole lot easier and it avoids in questioning of why some non-con games were excluded and others weren't. It is kind of just a fact of life though as far as sample sizes usually being too small, statistically speaking.
I don't know which is better at predicting teams that excel in the tournament. You could probably go back in history and try and deduce that. The thing there is that so much comes into play with how a game results. Matchups, shooting performance, did the point guard's girlfriend cheat on him while he was at the tourney site? All of those variables are so impossible to predict and account for and then re-apply to your predictions that it is the main reason I don't even try to predict games based off of a computer model.
I'd have just as likely of a chance of colonizing on the moon by stealing my daughters' Radio Flyer wagon and plastic play house, you know? I'll let the Vegas guys try and do that, I mean, I am on limited resources here!
Thanks for the comments.
Perhaps intangibles explaion why we are better than rated. At the last five minutes, Kane and Georges can go one on one time after time. Then Melvin can shoot the three ball or go inside. Monte handles the ball well enough that he too can get off a shot. Hopgued can hit free throws. Matt can play solid.The defense pressure can be ratcheted some for us, but not as much as other teams, so, it is our offensive capability that can win close games.
Our team has a dependability in the last five minutes. This means wins.
Getting a W against a good team is the only stat for that, and why the committee uses RPI instead of a rating system like KenPom that rewards teams who may get close in every game but lose big games more often than ISU has.
Some people don't like facts getting in the way of their arguments. Stat Guy is the opposite of that :twitcy:
"Weight your own perception based off all of the facts and information you can possibly gather along with your own evaluation from watching as many games as possible. "
Shut up sound logic guy!
Shut up sound logic guy!
Good article. Here is the question you asked, which I would like to see a stats guru attempt to answer: "Are Pomeroy’s predicted results any better than Vegas or Sagarin or Joe Blow Sports Writer?" In other words, "which ranking system is the best at predicting the actual winner of each game?" I think the method is pretty straight-forward, but would require a ton of number crunching. You look at every game, and who is predicted to win that game based on the ratings as of that date. Do this for an entire season, and measure the rate of correct predictions. Home/away differences should be a wash with a larger sample size. Is anyone up for the challenge?
It depends on how you frame the question. Take shooting percentage for example. You could look at the percentage in each of 14 conference games (14 data points). But another way to look at it is the accumulation of hundreds of individual shots, either made or missed, each with a given probability of going in. Now you've got hundreds (or even thousands) of data points. And whichever way you do it, you need to express the uncertainty in the value (assuming you are evaluating them as "sample" stats rather than "population" stats).
In my opinion it's not so much that people manipulate statistics as that they don't fully define their frame of reference before throwing out results.
Generally, in terms of prediction, ratings that rely on point differentials and ignore W/L (Sagarin predictor, KenPom etc.) consistently outperform ratings that only utilize W/L and ignore point differentials (Sagarin ELO, RPI etc.). Almost all rating systems can be thought of as being one of these two extremes or a hybrid somewhere in the middle. It's also my understanding that finding ways to incorporate W/Ls to improve a predictor over using pure points tends to be very difficult. Also, I think Vegas lines tend to outperform all of them so adding a human factor can help, likely related to injuries, game turnaround time, possibly style of play and matchups.
That all said, I think the rankings that utilize only W/Ls are still useful. And I generally feel that they should be used to select and seed the tournament (along with human input). They're kind of like the overall standings for the season and represent a team's resume. I think you should win your way into the tournament, not score more points to get into the tournament. Even if the other is a better predictor. It'd be like if the NFL or MLB playoffs were determined by points or runs scored rather than the standings.
