Preseason v Regular Season Correlation (1 Viewer)

[captain_amazing]

So, I was thinking that I would like to see if there is any statistical significance between NFL preseason and regular season records in the 2000's. I could have gone way back, but have no good way to pull data, and put all the data together myself using various websites. I am also interested in breaking down the data into different categories, to see if there is anything significant to be found. I had no hypothesis going into this, just thought it might be interesting.

I know there are various sources on the web that have done such a thing, and would love if others could post some of those sources in this thread. One can be found here http://www.twominutewarning.com/preseason-matter.htm.

I found that there is no statistical significance between preseason and regular season records from 2000-2009. I used alpha=0.5, df=31 (I left out Houston, as they were not in existence for the entire data range) and two-tailed test. The correlation coefficient needed to be greater than .355in order to be significant. R=0.19.

What this shows is that, while winning preseason games is generally good for teams' regular season success, it is not a significant indicator.

In the next week I am going to post other categories of data that will hopefully point out something significant. Would like to know everyone's comments!

 

[Sinn Fein]

So, I was thinking that I would like to see if there is any statistical significance between NFL preseason and regular season records in the 2000's. I could have gone way back, but have no good way to pull data, and put all the data together myself using various websites. I am also interested in breaking down the data into different categories, to see if there is anything significant to be found. I had no hypothesis going into this, just thought it might be interesting.

I know there are various sources on the web that have done such a thing, and would love if others could post some of those sources in this thread. One can be found here http://www.twominutewarning.com/preseason-matter.htm.

I found that there is no statistical significance between preseason and regular season records from 2000-2009. I used alpha=0.5, df=31 (I left out Houston, as they were not in existence for the entire data range) and two-tailed test. The correlation coefficient needed to be greater than .355in order to be significant. R=0.19.

What this shows is that, while winning preseason games is generally good for teams' regular season success, it is not a significant indicator.

In the next week I am going to post other categories of data that will hopefully point out something significant. Would like to know everyone's comments!

Hope you did not spend a lot of time on this. Really trying to compare apples and oranges here. Teams have different agendas in preseason and regular season, and make decisions accordingly.

 

[captain_amazing]

Hope you did not spend a lot of time on this. Really trying to compare apples and oranges here. Teams have different agendas in preseason and regular season, and make decisions accordingly.

Not a lot of time. I realize that teams do have different agendas and take on preseason games differently, which was especially apparent for certain teams, where a very strong positive or very negative correlation existed between preseason and regular season records (among other things). Regardless of that, I was mostly just curious, overall, what the data would show.

 

[Winky the tunnel ferret]

I'm less concerned with team w/l records but I'd be curious to see if there is any correlation to player performances.

Ie sleepers, touchdowns, etc.

 

[captain_amazing]

I'm less concerned with team w/l records but I'd be curious to see if there is any correlation to player performances. Ie sleepers, touchdowns, etc.

I'll attempt to add that as I break down the data further. I think that would be very interesting.

 

[hook014]

So, I was thinking that I would like to see if there is any statistical significance between NFL preseason and regular season records in the 2000's. I could have gone way back, but have no good way to pull data, and put all the data together myself using various websites. I am also interested in breaking down the data into different categories, to see if there is anything significant to be found. I had no hypothesis going into this, just thought it might be interesting.

I know there are various sources on the web that have done such a thing, and would love if others could post some of those sources in this thread. One can be found here http://www.twominutewarning.com/preseason-matter.htm.

I found that there is no statistical significance between preseason and regular season records from 2000-2009. I used alpha=0.5, df=31 (I left out Houston, as they were not in existence for the entire data range) and two-tailed test. The correlation coefficient needed to be greater than .355in order to be significant. R=0.19.

What this shows is that, while winning preseason games is generally good for teams' regular season success, it is not a significant indicator.

In the next week I am going to post other categories of data that will hopefully point out something significant. Would like to know everyone's comments!

You know what would really make you "Captain Amazing"?

Explaining two-tailed test, correlation coefficient, and alpha, Df so that we could all understand it. Cause I'm looking at a chinese menu here and thinking a picture of a chicken would be helpful.

 

[FreeBaGeL]

I'm less concerned with team w/l records but I'd be curious to see if there is any correlation to player performances.

Ie sleepers, touchdowns, etc.

We're trying to track this for the future here.

 

[captain_amazing]

You know what would really make you "Captain Amazing"?

Explaining two-tailed test, correlation coefficient, and alpha, Df so that we could all understand it. Cause I'm looking at a chinese menu here and thinking a picture of a chicken would be helpful.

I will still be captain_amazing, with or without this, but I hear ya.In statistics, in testing the correlation of two variables (variable X and variable Y) in a given data set, you can calculate the correlation coefficient (the number that generates a regression line, which takes into account the entire data set, and sticks a line in that indicates whether the relationships of the variables are positive or negative).

The correlation coefficient is between -1 and 1, where negative numbers indicate negative correlations and positive numbers indicate positive correlations. For example, if the correlation coefficient for our data set is 0.85, then what that says is that the more preseason wins a team gets the higher their regular season wins will be, overall.

What statisticians will do is take the correlation coefficient and find what its significance is (whether, in the given data set, based on several factors, the coefficient is strong enough to be significant in predicating outcomes). So statisticians will do either a one-tailed (if they have a strong hypothesis as to what the outcome of the test will be) or two tailed test (if there is a weak or no hypothesis, as in my case).

I won't bother explaining the test itself, or alpha or df (check out this link for more info on these two http://www.socialresearchmethods.net/kb/statcorr.php, it could explain it better than I), but in the end, you will end up with a number between 0 and 1. That number will indicate the threshold, where if your coefficient falls below that number, it is not significant. IF it is above that threshold, it is significant, and can be used to successfully predict other outcomes.

I hope that helps a bit. Might be even more confusing

We're trying to track this for the future here.

I wonder if we could find out, generally, who the sleepers and hyped busts where in the 2000's? We could back track and measure, against some of the data I have collected and will develop, progressions of player performance based on some interesting factors. That thread is great, BTW!

 

[Sammy Traveller]

will this end the preseason silliness once and for all?

look up Ben Tate and Montario Hardesty.

 

[hook014]

You know what would really make you "Captain Amazing"?

Explaining two-tailed test, correlation coefficient, and alpha, Df so that we could all understand it. Cause I'm looking at a chinese menu here and thinking a picture of a chicken would be helpful.

I will still be captain_amazing, with or without this, but I hear ya.In statistics, in testing the correlation of two variables (variable X and variable Y) in a given data set, you can calculate the correlation coefficient (the number that generates a regression line, which takes into account the entire data set, and sticks a line in that indicates whether the relationships of the variables are positive or negative).

The correlation coefficient is between -1 and 1, where negative numbers indicate negative correlations and positive numbers indicate positive correlations. For example, if the correlation coefficient for our data set is 0.85, then what that says is that the more preseason wins a team gets the higher their regular season wins will be, overall.

What statisticians will do is take the correlation coefficient and find what its significance is (whether, in the given data set, based on several factors, the coefficient is strong enough to be significant in predicating outcomes). So statisticians will do either a one-tailed (if they have a strong hypothesis as to what the outcome of the test will be) or two tailed test (if there is a weak or no hypothesis, as in my case).

I won't bother explaining the test itself, or alpha or df (check out this link for more info on these two http://www.socialresearchmethods.net/kb/statcorr.php, it could explain it better than I), but in the end, you will end up with a number between 0 and 1. That number will indicate the threshold, where if your coefficient falls below that number, it is not significant. IF it is above that threshold, it is significant, and can be used to successfully predict other outcomes.

I hope that helps a bit. Might be even more confusing

We're trying to track this for the future here.

I wonder if we could find out, generally, who the sleepers and hyped busts where in the 2000's? We could back track and measure, against some of the data I have collected and will develop, progressions of player performance based on some interesting factors. That thread is great, BTW!

That's cool. I respect mathematical studies and I learned something here. Mainly some jargon I can use to impress a cute but nerdy chick I met. So it's all good.

But Sammy makes a good point, one torn ligiment and it all goes out the window.

OTOH if you find something useful let us know.

 

[captain_amazing]

Nothing of hard FF usefulness just yet (but I am working there), but I found something I thought was rather interesting. While you could not predict, overall, NFL teams regular season records based on their preseason record in a way that was statistically sound, there was one team that you CAN statistically predict their regular season record with their preseason record: Denver.

They had such a strong correlation between their preseason record and regular season record between 2000-2009, that there was only a 2% probability that the correlation was just chance. The correlation is a positive one, meaning as they win more preseason games, they win more regular season games.

So with their 2-2 preseason record this year, you could predict, statistically, that they have a strong chance of ending up 8-8 +-1 game, based on their record from 2000-2009.

 

[CalBear]

It's always good to check widely-held assumptions (such as "preseason records are meaningless") against actual data, so thanks for doing this (even though it just confirmed the assumption, more or less.)

I think you're probably being facetious with your Denver comment; you seem to know enough about statistics to know that if there's a 2% chance that something will be true about one specific team, there's almost a 50% chance (47.6%) that it will be true about one of the teams in the league. There's an equal chance that you'll have one team with just as strong a negative correlation (so that wins in the pre-season equate to losses in the regular season, and vice-versa).

 

[captain_amazing]

It's always good to check widely-held assumptions (such as "preseason records are meaningless") against actual data, so thanks for doing this (even though it just confirmed the assumption, more or less.)

I think you're probably being facetious with your Denver comment; you seem to know enough about statistics to know that if there's a 2% chance that something will be true about one specific team, there's almost a 50% chance (47.6%) that it will be true about one of the teams in the league. There's an equal chance that you'll have one team with just as strong a negative correlation (so that wins in the pre-season equate to losses in the regular season, and vice-versa).

Yes, you are right, except I was not being facetious. Based on my wording, I can see how you concluded that, which is what I would have done if I read my statement. Let me clarify: there is not a 2% chance that there will be or could be a statistically significant correlation by any one specific team (which is how it sounded the way I worded it above). Looking at that specific correlation, the odds are 2 out of 100 that this a chance occurrence. So, its not a 2% chance that it could be true, it is a 2% probability that the occurrence was just by chance, and is not statistically significant.

I am not saying you were wrong, because you were right, based on how I worded my statement. So sorry for being misleading.