Clayton Blog: Competition Committee recommends OT changes

[CrazyCowboy]

ROMO will catch that next snap!

 

[Mavs Man]

You draw the line at context. Here is where critical thinking skills come in... E.g., is this result likely to change much over the next year? For 3 games, it could shift considerably. For data based on 30 years, it would be nearly impossible for the observed differences to fade based on a reasonble # of OT games.

I don't have the 2006 stats but going into the 2003 season there had been 342 overtime games, with those winning the toss winning the game 177 times, those losing the toss winning the game 149 times, and 16 ties.

If it's only critical thinking skills (personal judgment call) rather than objective statistics that determines if stats pass the mustard test, then I can't say a 54% chance of "not losing" is much of an advantage to indict the coin toss.

And other factors do make a difference, though I admit I don't have the stats to back it up. But I would be willing to bet that there is a slight difference between home and away, home team winning the toss and winning the game, home team losing the toss but winning the game, etc. If the figures were about the same I would agree it doesn't make a difference, but I would have to see what they were.

Going back to other posts in this thread, I think that the opening possession scoring 29% of the time is much more significant and is a better indicator of OT games being skewed toward the winning toss than overall win/tie percentage. Though, I wouldn't split hairs and would make it the first drive - the percentage of games ending on a pick 6 on the opening drive of OT is most likely only one or two percent, but it's still greater than zero.

 

[burmafrd]

then of course you have to find out how many INTs and how many fumbles happened on that opening drive; then look at major penalties or big sacks.
to me, if it really only affects maybe one or two games a year it is not worth messing with. Sudden Death is great because sometimes it is just THAT.
No one has a reason to complain- they had 60 MINUTES of playing time to win it otherwise.

 

[burmafrd]

By the way- ANYONE that claims that a pick 6 happens on the second possession sounds like a shyster lawyer chasing ambulances.

 

[AbeBeta]

So, 100 coin flips isn't entire population, but 300 football games is entire population? There's no difference. There are potentially infinite coin flips, and there are potentially infinite football games in OT. You can't use 55 out of 100 flips being heads to claim that the coin is rigged in heads favor, just like you can't use 52% wins out of 300 football games to claim that the coin flip decides the game. (Note: Obviously, I think there is some level of influence in the coin flip and I think this should be eliminated, but that's not what we're talking about at this point. My point is that I don't know whether the sameple size is large enough to know how correct the 52% number is.)

Again, this isn't a sample. It is data from all the games that were played. Inferential statistics focus on drawing inferences about populations based on samples. This is not a sample. It is all of the games that went to OT. That's the population. We know exactly how correct the 52% is because it is based on the entire population.

Significance tests are for situations when we estimate the population value from the sample. We have the population value. This isn't a theoretical value. It is right there, measured for us.

I can take a 10 year old company's stock and measure it's performance against the market for 10 years (which would be "an entire population" for the company's stock), and if the cumulative abnormal returns are small enough, then the number will not be statistically significant. I can similarly take 30 years of football games and measure the results against coin flips for 30 years (which is what we're doing here). If the abnormal results are small enough, then the difference will not be statistically significant.

Again, the error in your thinking is that you are taking a population value and applying techniques focused on sampling to it. In this case, the difference either is or isn't there. A better question is that difference large enough to make you concerned. The issue is effect size not significance.

To illustrate the problem, over the next 5 year span, it is entirely possible for the team that loses the coin flip to win every single OT game. That would likely change the results sigificantly. Let's assume that it would change the results to the coin-flip loser winning 55% vs. the coin-flip winner winning the game 40%. At that point, you couldn't make the argument that losing the coin helps win the game, because obviously there is some level of chance in the results. Statistical significance states whether the abnormal results (compared to the coin-flip having absolutely zero impact, i.e., teams spliting 47.5-47.5-5) of the sample size (which is all this is) are meaningful -- in other words, whether they're true.

You are correct about what significance testing does. But incorrect in how you apply it. We don't need hypothesis testing to tell us in this case whether we differ from 47.5-47.5-5. The exact question those tests ask is "would this sample result be surprising if the population looked like this (47.5 etc.)?" We know exactly what the population looks like in this case. You don't apply significance tests to population data. It is totally meaningless.

By your logic, in a close election, we would ask "is the 1000 vote difference between the candidates statistically significant?" It isn't a relevant question, the population is what it is. Significance tests address sampling error. There is no sample so there is no sampling error.

What happens over the next 5 years might change our interpretation -- because that would now be data that was added to the population. That has nothing to do with significance testing.

 

[AbeBeta]

By the way- ANYONE that claims that a pick 6 happens on the second possession sounds like a shyster lawyer chasing ambulances.

It is called a change of possession. If a foul occurred prior to the pick, it would remain the O's ball. If it occurred after, it would remain the D's ball.

How can you have a change of possession and not move from possession 1 to possession 2?

 

[theogt]

Honestly, I can't believe you're being this wrong and this stubborn. I've used significance testing in financial studies. I know what they are.

 

[theogt]

Again, this isn't a sample. It is data from all the games that were played. Inferential statistics focus on drawing inferences about populations based on samples. This is not a sample. It is all of the games that went to OT. That's the population. We know exactly how correct the 52% is because it is based on the entire population.

Significance tests are for situations when we estimate the population value from the sample. We have the population value. This isn't a theoretical value. It is right there, measured for us.

It is a sample of a potentially infinite data set. We're not using a sample of 30 games to determine what the percentage would be in a population of 300 games; rather, we're using 300 games as a sample of what the percentage would be of a population of an infinite number of games. Maybe that helps.

 

[AbeBeta]

Honestly, I can't believe you're being this wrong and this stubborn. I've used significance testing in financial studies. I know what they are.

OGT - I'm not some hack who took an Intro Stat course. I've got a Ph.D. in Quantitative Methods. I write peer-reviewed articles on the topic. I've used significance testing pretty much every day of my life for the past 15 years. Seriously.

 

[AbeBeta]

It is a sample of a potentially infinite data set. We're not using a sample of 30 games to determine what the percentage would be in a population of 300 games; rather, we're using 300 games as a sample of what the percentage would be of a population of an infinite number of games. Maybe that helps.

Samples are used when the population outcome is immeasurable. Here the outcomes for the population ARE measurable. We don't have an infinite number of games. We have a finite number of games.

 

[zack]

Know how many games? To see if it's statiscally significant?

Regardless, I'd like to see it change. I just don't like sudden death. All it takes is one break away play after 60 minutes of hard fought football.

Or one 50 yard BS pass interference call that sets up the winning field goal. Sorry, but that penalty needs to be addressed!

 

[bbgun]

OGT - I'm not some hack who took an Intro Stat course. I've got a Ph.D. in Quantitative Methods. I write peer-reviewed articles on the topic. I've used significance testing pretty much every day of my life for the past 15 years. Seriously.

This better not devolve into a discussion on geocentrism, Aristotle, sophists, Kant's "Categorical Imperative," etc.

 

[theogt]

Samples are used when the population outcome is immeasurable. Here the outcomes for the population ARE measurable. We don't have an infinite number of games. We have a finite number of games.

We know what the results of the entire eisting population are, of course. What we want to know is what the results would be over 1,000,000 or more games (i.e., infinite). We want to see whether the results of this sample are close enough to the results of a potentially infinite data set from which we obviously cannot determine the "true" results. Obviously you can't determine an exact statistical significance, but I think we can "eyeball" it and say that 52% doesn't truly reflect the effect of the coin-flip on winning OT games.

 

[Mavs Man]

This better not devolve into a discussion on geocentrism, Aristotle, sophists, Kant's "Categorical Imperative," etc.

That reminds me of a story from high school. A friend of mine was arguing with someone (don't even remember what it was about now) and he made the comment "what Kant would say about this is . . ." Before he could finish, the other guy screamed, "Well, Kant says you're gay!" and ran off.

 

[AbeBeta]

We know what the results of the entire eisting population are, of course. What we want to know is what the results would be over 1,000,000 or more games (i.e., infinite). We want to see whether the results of this sample are close enough to the results of a potentially infinite data set from which we obviously cannot determine the "true" results. Obviously you can't determine an exact statistical significance, but I think we can "eyeball" it and say that 52% doesn't truly reflect the effect of the coin-flip on winning OT games.

But you see here, you've walked right into the flaw in this thinking -- there aren't going to be 1,000,000 more games. There is not an infinite data set here. It is a finite data set. The population is exactly as it was measured. Will future events change the population, certainly that is likely. We aren't searching for some population truth based on a sample -- the true value for the population exists and is right in front of our face.

The real questions here are a) is there enough population data available to draw a conclusion and b) is the difference in win/loss % big enough to matter.

Regarding a, we can speak to the stability of the population values. If these data are based on 350 games, this would mean that a streak of about 32 straight OT games ending in a loss for the team who wins the toss would bring us to equal.

Regarding b, we can compare 52% to 43%. Clearly that is a big advantage.

 

[Mavs Man]

If you throw out ties (status quo) then it works out to 54% in favor of winning the toss (46% wins game when losing toss).

 

[Yeagermeister]

That OT change is fine with me just don't change it to that stupid system college football uses. If your defense can't stop a team you don't deserve to win.

 

[Mavs Man]

Again, I think this argument comes down to statistical semantics. For you it's significant, for others it isn't. It's not over 60% or something that would be considerably obvious that only a fool would argue with you, as some of your posts subtly hint at.

But the better stat is the 29% chance of scoring on the opening drive compared to the 1-2% (if that) for the opposing team to score off a turnover on the opening drive.

Unless they're playing against Rex Grossman.

 

[Mavs Man]

That OT change is fine with me just don't change it to that stupid system college football uses. If your defense can't stop a team you don't deserve to win.

Fantasy football leagues might like that - more scoring.

 

[theogt]

But you see here, you've walked right into the flaw in this thinking -- there aren't going to be 1,000,000 more games. There is not an infinite data set here. It is a finite data set. The population is exactly as it was measured. Will future events change the population, certainly that is likely. We aren't searching for some population truth based on a sample -- the true value for the population exists and is right in front of our face.

There is a potentially infinite data set. Just like the coin flip situation. There is no difference. You can flip the coin 350 times , and if the results are 52% heads, then that abnormal result may or may not be statistically significant, meaning it may or may not be sufficient to tell us there is some flaw in the coin causing it to land on heads.

The real questions here are a) is there enough population data available to draw a conclusion and b) is the difference in win/loss % big enough to matter.

Yes, that is exactly what we're asking. Whether this is a large enough sample of a potentially infinite data set to tell us whether the results here are statistically significant.

Regarding b, we can compare 52% to 43%. Clearly that is a big advantage.

Maybe. Maybe not. Obviously there should be some influence. The influence may actually be greater than the stats shown here. I don't think the sample is large enough to really make any meaningful determination one way or the other.