Statistics tell the whole story. Independently verified?

[daj2576]

There are three types of lies: Lies, damn lies, and statistics.

My brother is a statistician for an Archeology firm, and even he will tell you Mark Twain's comment is true.

There are many sites and companies that do this, (lately impart to the gambling community), but it is no more than glorified guess work.

And no offense to whatever it is you do. Statistics is one of the most fascinating branches of mathematics, and it is certainly interesting to run trough numbers and see what they say.

But I most instances statistics are more biased than any half wit on ESPN or anywhere else.

So succesful gamblers who win large amounts of money by using tested statistics about past events to predict future events are to be discounted because your relative likes quotes by Mark Twain? I think I understand.

 

[alaVOL]

Statistics are facts and history repeats itself. hmmmmm

 

[daj2576]

I am totally beating my head against a table here.
1) statistics are evidence, not proof. That is why when I talk about these number's predictive abilities, I say that they are correct X% of the time. If you notice, that number X is always less than 100. That means, summarily, that these predictions are not always correct, but they are correct the majority of the time.
2) Bringing up an exception, either an exception to this sort of analysis, or an exception to the general predictive ability of statistics is also not proof that ALL predictions fail. In essence, you are saying that statistical predictions fail because you know of times when they have. Does anyone see the irony of attempting to use poor statistics to disprove the use of good, conservative, tested statistical analysis?
3) The Mark Twain quote is amusing, but it must be taken in context. Do you realize that quote, is from the late 1800's? Here is another funny quote from the same time period: "Everything that can be invented has been invented." or try this one on from 1949 "Computers in the future may weigh no more than 1.5 tons."

 

[TreeGreen]

So succesful gamblers who win large amounts of money by using tested statistics about past events to predict future events are to be discounted because your relative likes quotes by Mark Twain? I think I understand.

I think a lot of people who like to use this sentiment don't really understand how statistics work. They see someone say that they're 95% confident that something will pass, and then the handful of times where that other 5% occurs they point and go "SEE! You're statistics are nonsense and obviously don't work!" Which is baloney. Models like this only tell us what we should anticipate to happen not what is going to happen with 100% certainty. Stj has sort of touched on it. If the record you're predicting doesn't occur it likely is not because this model is inherently wrong, but because something we thought unlikely occurred instead. Such as the team just flat out under performing in spite of its talent. I will admit that there are bad statistics out there, but no one is trying to hide anything here. We can all see the model and decide if its something we can get behind.

*Edit - of course that assumes you are knowledgeable enough about statistics and predictive models to evaluate them in the first place.

 

[Vol4life25]

I am totally beating my head against a table here.
1) statistics are evidence, not proof. That is why when I talk about these number's predictive abilities, I say that they are correct X% of the time. If you notice, that number X is always less than 100. That means, summarily, that these predictions are not always correct, but they are correct the majority of the time.
2) Bringing up an exception, either an exception to this sort of analysis, or an exception to the general predictive ability of statistics is also not proof that ALL predictions fail. In essence, you are saying that statistical predictions fail because you know of times when they have. Does anyone see the irony of attempting to use poor statistics to disprove the use of good, conservative, tested statistical analysis?
3) The Mark Twain quote is amusing, but it must be taken in context. Do you realize that quote, is from the late 1800's? Here is another funny quote from the same time period: "Everything that can be invented has been invented." or try this one on from 1949 "Computers in the future may weigh no more than 1.5 tons."

What's your picks for the 13 season then?


Come on bro, put your money where your mouth is.

 

[daj2576]

I think a lot of people who like to use this sentiment don't really understand how statistics work. They see someone say that they're 95% confident that something will pass, and then the handful of times where that other 5% occurs they point and go "SEE! You're statistics are nonsense and obviously don't work!" Which is baloney. Models like this only tell us what we should anticipate to happen not what is going to happen with 100% certainty. Stj has sort of touched on it. If the record you're predicting doesn't occur it likely is not because this model is inherently wrong, but because something we thought unlikely occurred instead. Such as the team just flat out under performing in spite of its talent. I will admit that there are bad statistics out there, but no one is trying to hide anything here. We can all see the model and decide if its something we can get behind.

I actually get the resistance to models that predict something that we are so invested in on an emotional level. I still have not fully come to terms with how I can tell you the outcome of about 60-70% of the college football games on any given week and yet I still watch, unsure of the outcome. In fact, when I am asked to predict Tennessee's games, I find that I depart from the known probable outcomes and go with my gut, or my heart. So, I get that part.

As you mentioned, this isn't a "trust me" sort of statistical model where I tell you the outcome and you can either believe me or not. This is a tested model, where I have given everyone my methodology and have given a website to someone who has been doing the same thing (unbeknownst to me until last night) for almost a decade. The method and the findings are all there yet people want to talk about how it can't work because they feel it shouldn't. That part blows my mind.

 

[TreeGreen]

As you mentioned, this isn't a "trust me" sort of statistical model where I tell you the outcome and you can either believe me or not. This is a tested model, where I have everyone my methodology and have given a website to someone who has been doing the same thing (unbeknownst to me until last night) for almost a decade. The method and the findings are all there yet people want to talk about how it can't work because they feel it shouldn't. That part blows my mind.

I think that's just because although statistics are not necessarily difficult, they do involve some math and lots of people are either never exposed to their underpinnings or get some cursory introduction during college. So, a lot of people in the general population just think they're voodoo magic or something.

 

[daj2576]

What's your picks for the 13 season then?


Come on bro, put your money where your mouth is.

I do but you can't use just this system to beat the spread enough to make it lucrative. You can pick winners/losers at a rate of 60-70 percent and still only beat the spread by the tiniest of margins.

There are ways that you can tweak and refine this data to make it even more predictive and lucrative. I don't believe, however, that my employer would be happy if I was giving that away on a message board.

 

[daj2576]

I think that's just because although statistics are not necessarily difficult, they do involve some math and lots of people are either never exposed to their underpinnings or get some cursory introduction during college. So, a lot of people in the general population just think they're voodoo magic or something.

The scariest thing, to me, is that we used to be a nation that respected knowledge and education. Now it seems that we are scared and frightened by it to the point of irrationally distrusting those things.

Things like this should be questioned, but to vehemently disagree without any sort of substantive argument is beyond me.

 

[TreeGreen]

I do but you can't use just this system to beat the spread enough to make it lucrative. You can pick winners/losers at a rate of 60-70 percent and still only beat the spread by the tiniest of margins.

There are ways that you can tweak and refine this data to make it even more predictive and lucrative. I don't believe, however, that my employer would be happy if I was giving that away on a message board.

I would think to start beating the spread more often you would need significantly more variables included in the model. Maybe not, because the talent gap tends to be pretty significant at times in college, but I know for looking at games during the NFL season I'm paying attention to east coast teams flying to play west coast team night games. Especially considering when the west coast teams win 60% of their night games played against an east coast opponent by fairly wide point margins (thank god we get Oregon during the afternoon, right?).

 

[daj2576]

I would think to start beating the spread more often you would need significantly more variables included in the model. Maybe not, because the talent gap tends to be pretty significant at times in college, but I know for looking at games during the NFL season I'm paying attention to east coast teams flying to play west coast team night games. Especially considering when the west coast teams win 60% of their night games played against an east coast opponent by fairly wide point margins (thank god we get Oregon during the afternoon, right?).

:thumbsup:

 

[TheGreenCloaca]

The scariest thing, to me, is that we used to be a nation that respected knowledge and education. Now it seems that we are scared and frightened by it to the point of irrationally distrusting those things.

Things like this should be questioned, but to vehemently disagree without any sort of substantive argument is beyond me.

The regularity with which "talent" quantification comports with reality is the issue, I guess.

Curious -- does you analysis suggest a score in the UT/Oregon game?

 

[Rifleman]

The regularity with which "talent" quantification comports with reality is the issue, I guess.

Curious -- does you analysis suggest a score in the UT/Oregon game?

Models which use similar formulas say it will be a closer than expected loss for us, but I think most freely available on the internet are not sophisticated enough to really pin down scoring margins.

The original betting line was Oregon by 25 and I believe that has already come down a bit.

One complexity in college football is that a very close game often becomes a blowout in a matter of minutes on the field due to conditioning and talent disparities. A less talented team must put forth more effort to match a more talented team, but at a certain point the less talented team red lines unless they possess superior conditioning.

That very well could happen here, but I want to believe that our conditioning, underrated talent and preparation will allow us to be in it all the way.

Another note: Stanford plays a 3-4, but they kept a linebacker/DE hybrid (often called Leo) at the line against Oregon last year to slow the run game and it worked!

Since we are switching back to 4-3, but it is the Leo variety 4-3, I think a lot can be learned from that Stanford tape. Of course, the Ducks will take some lessons from it too, but I think there are reasons for optimism that we can compete.

My prediction would be Oregon 34 - Tennessee 30.

 

[McDad]

Many of you read my discussions on talent evaluations as a way to predict football games (generally: 2/3 of the time, a four year trailing talent average can predict seasonal outcomes).

Some of you even participated in some threads where I developed and tested this process (thank you for the mature discussion and insight).

Today, I stumbled across a website where someone is using a very similar system to the one that I tripped upon.

Here is his page dedicated to UT evaluations.-->Tennenessee Vols 2013 Football Preview

Notice some similarities to things that we have already discussed, 1) He predicts UT will go 7-5 or 8-4 this year, exactly the same prediction that my numbers show, 2) He says that Jones had a +1 effect at Cincinnati. My numbers suggest it was +2 if you average in the first year where he was +0, or +3 if you consider his first year at Cincy to be an outlyer., 3) He suggests that UT v. Oregon will be a much closer game than most believe. Ditto.

I found this website to be fascinating, and also a little humbling. I thought I had stumbled upon something really unique. This guy has been doing something very similar for several years, apparently. I also laughed when I read through some of his statements about how many people completely reject this system, no matter how strong the data.

He also adds more and other details about the likelihood of success for a first year coach. This is definitely an interesting read.

daj!!!!!!! great to read you again. When you initially posted your matrix last season, I heaped tons of praise on you for it. I thought it was grand. Now, it seems that others are just as smart as you. LOL.

You mentioned that the other site said Sumlin was a one year wonder. What was their reasoning?

 

[daj2576]

daj!!!!!!! great to read you again. When you initially posted your matrix last season, I heaped tons of praise on you for it. I thought it was grand. Now, it seems that others are just as smart as you. LOL.

You mentioned that the other site said Sumlin was a one year wonder. What was their reasoning?

I think the reasoning was just that you can't judge Sumlin for over performance on such a narrow window. Personally I think Sumlin benefited from Arkansas and Auburn being at historically low levels. A similar example was when Ole Miss beat Florida with Tebow at QB, but that was an anomaly. A&M losing to Florida and LSU, beating Bama, and beating historically bad Arkansas and Auburn teams earns Sumlin and A&M undue praise (at least in my view).

Bartoo from CFBmatrix.com sent me a twitter message today. Apparently he has been reading this thread. Someone had asked about the effect of returning QBs on seasonal outcomes. He had the answer, and it will probably surprise everyone (at least it did me).

College Football Returning Starters Kickers and QB 2013

You always flatter me. Others are just as smart as me, and even smarter. This guy is probably smarter, and definitely quicker, as he has been working on this evaluation for a decade.

 

[daj2576]

Models which use similar formulas say it will be a closer than expected loss for us, but I think most freely available on the internet are not sophisticated enough to really pin down scoring margins.

The original betting line was Oregon by 25 and I believe that has already come down a bit.

One complexity in college football is that a very close game often becomes a blowout in a matter of minutes on the field due to conditioning and talent disparities. A less talented team must put forth more effort to match a more talented team, but at a certain point the less talented team red lines unless they possess superior conditioning.

That very well could happen here, but I want to believe that our conditioning, underrated talent and preparation will allow us to be in it all the way.

Another note: Stanford plays a 3-4, but they kept a linebacker/DE hybrid (often called Leo) at the line against Oregon last year to slow the run game and it worked!

Since we are switching back to 4-3, but it is the Leo variety 4-3, I think a lot can be learned from that Stanford tape. Of course, the Ducks will take some lessons from it too, but I think there are reasons for optimism that we can compete.

My prediction would be Oregon 34 - Tennessee 30.

You are correct. The models that are sophisticated enough to predict spreads/scores are extremely complicated and expensive. Make no mistake, they absolutely do exist, and some are even good enough to be incredibly lucrative.

 

[Ten_Titans]

:ermm::ermm: ....................................... :no1:

Tell you what. Let's meet back here in say 3 months and check to see how many teams in the bottom half of the SEC statistically are in the top 2 of their division.

You can mishandle statistics but they provide a reliable tool for establishing probabilities. The better they match history... the better you can expect them to predict the future.


What you posted basically amounts to an assertion that understanding history is not important to making good decisions in the future. I suppose corporate spending on statistics is a waste of time? I guess those marketing guys should stop it too, right?


Archaeology is a bad example btw. By and large, they very often use faulty premises and assumptions about events so far in the distant past that they truly are guessing.

Yes, marketing is a gigantic scam of a degree, and an absolute silly field of study. And yes, corporate spending on statistics is mostly a waste of money.

I think I should probably elaborate before you think I am trying to assert that I am writing off statistics as a whole. That is not the case at all. We use them everyday.

I am at work right now, when I get home I will elaborate on my desk top.

 

[daj2576]

Yes, marketing is a gigantic scam of a degree, and an absolute silly field of study. And yes, corporate spending on statistics is mostly a waste of money.

I think I should probably elaborate before you think I am trying to assert that I am writing off statistics as a whole. That is not the case at all. We use them everyday.

I am at work right now, when I get home I will elaborate on my desk top.

I have a sister that works in the social services field, and she has a favorite quote from Mark Twain as well. It goes something like this: "Better to be silent and thought a fool than to open one's mouth and remove all doubt."

 

[sjt18]

Yes, marketing is a gigantic scam of a degree, and an absolute silly field of study.

Which is proven by Microsoft vs Apple, right? Oh wait... the company with great marketing holds over 80% of market share while the company with the superior product does not...

And that is hardly a unique example. Statistics tell marketing people what moves consumers... and it works. What you said is (and I do apologize for being blunt) abjectly stupid. If marketing did not work... GM would probably be out of business because they produced inferior vehicles for the better part of 25 years.

Another great example was the stupid move by the former Google exec to change JC Penney's marketing approach. He ignored his demographic and what was proven to move that demographic. He stopped coupons and sales and hired a lesbian to be one of his main spokespeople. While much of the country does not care about that last tidbit... the unique demographic that made JCP money for years does- married with kids and generally thought of as "traditional" families.

And yes, corporate spending on statistics is mostly a waste of money.

I will remember that next time we make money while our competition does not because we have better data and statistical evaluations than any of them. And we use them in a commodity market which has a diverse set of factors.

Past performance does not "guarantee" an outcome but it does help determine the probability of something happening in the future. You are on average going to position yourself better if you have more information and have collated it effectively- statistics. That is especially true when the thing you are trying to predict has to do with human skills or behaviors. People ARE predictable.

I think I should probably elaborate before you think I am trying to assert that I am writing off statistics as a whole. That is not the case at all. We use them everyday.

I am at work right now, when I get home I will elaborate on my desk top.

If I misunderstood your intent then I apologize. But if you mean what you said... then I did not misunderstand.

 

[DiderotsGhost]

Many of you read my discussions on talent evaluations as a way to predict football games (generally: 2/3 of the time, a four year trailing talent average can predict seasonal outcomes).

Some of you even participated in some threads where I developed and tested this process (thank you for the mature discussion and insight).

Do you have a link to your analysis? Would be interested in reading.

 

[daj2576]

Do you have a link to your analysis? Would be interested in reading.

Somewhere buried on here, there are a few threads from a year or so ago bearing the name "statistics tell the whole story, 1,2 and 3". The evolution of this analysis ends right about where cfbmatrix.com picks up. Go there, he has some great articles and explanations of the fully developed matrix.

 

[VolGee4]

I believe this guy was on Memphis radio last week, and now I see a quote on the website from the station's producer. It was really interesting, especially the coach effect. He had good things to say about CBJ.

I believe he was asked about the Vandy-OM matchup to start the year, and he said that OM was the favorite under his system. Franklin is 0-4 at home against coaches that have the same "coach effect" as Hugh Freeze. Can't quite remember that formula, but it involves recruiting and outperforming/underperforming your talent.

Interesting you looked at some of this, too. I think it shows that we have a good coach, but don't be quick to lose faith when/if we struggle this season. We have some talent, but we are very thin and could lose games due to injuries.

 

[Rifleman]

You are correct. The models that are sophisticated enough to predict spreads/scores are extremely complicated and expensive. Make no mistake, they absolutely do exist, and some are even good enough to be incredibly lucrative.

I can imagine some ways to create a great refined model, but they would all require a lot of data and it would all need to be kept accurate every week.

There are probably some ways of automating some of that, but at the end of the day someone is going to need to check that data for accuracy. Therefore, it is probably a little too much for one hobbyist to make a great model to predict margins in all college football games. In general, if someone goes through that much trouble (or exudes that much cleverness) they are gonna want a pay day.

Is there is a way to crowd-source such a thing online?

I like cfmatrix, though, because he really has it boiled down to the key components. If I made a refined model, I'd start with something similar and add complexity. I think you could test against previous seasons and arrive at something that arrived at +90% on the W/L. Predicting total score and victory margins is where the gold is, though.

 

[DiderotsGhost]

Excellent data and analysis. A dozen thumbs up from me!

The coach effect is the one aspect of this I would question. Obviously, there is a "coach effect", but the problem is that it's a difficult thing to quantify in terms of a 'win differential'.

For instance, Bill Snyder destroys all other coaches on this (+5.25). This is partly because Bill Snyder is a truly excellent coach. It's also partly because Bill Snyder has been a horrendous recruiter (at least on paper) at K State.

Whereas, Nick Saban is +0.20. It's difficult to imagine how Nick Saban could possibly beat his recruiting averages, though; no matter how great of a coach he is. In fact, I think the model has predicted 13 wins for Alabama over the past four years.

I do think 7-5 is about where Tennessee should be when it comes to talent. We're talented enough to beat Georgia or South Carolina; maybe even Florida or Oregon. But I suspect Auburn and / or Missouri could be more difficult than people realize, as well.

 

[Catbone]

We have the talent almost as good as USCjr but we don't have depth which I believe will ultimately kill us this year