stats correlations in the NFL

[MIJB#19]

Correlations below are all calculated with the CORREL function in WinExcel by using the 2002 regular season stats, compared to the # of wins, as found on NFL.com.

Some testing showed that replacing the ties for Pittsburgh and Atlanta with wins or losses would change correlation by at most 0.005.

The most intruging correlations I found for winning are below, under the Top 10 and Bottom 10.

Top 10
0.795 least # of kickoffs against
0.772 # of Fieldgoal attempts
0.767 least time of possession for opponents
0.766 time of possession on offense
0.745 least # of rushing attempts against
0.743 least # of kickoff returns
0.736 # of Fieldgoals made
0.711 least # of points again
0.684 # of kickoffs kicked
0.676 # of points scored

Bottom 10
0.000 % of 30-39yd fieldgoals scored against
0.002 # of punt returns
0.007 least # of 4th down conversions made by offense (more is worse!)
0.009 least # of 50+yd Fieldgoals scored against
0.016 least # of punt return Touchdowns (more is worse!)
0.018 # of 50+yd Fieldgoals scored
0.019 % of 40-49yd Fieldgoals made
0.026 # of penalties called on team
0.030 least # of kickoff return Touchdowns allowed
0.034 # of kickoff return Touchdowns


Interesting selected correlations
0.673 # of 30-39yd fieldgoal attempts
0.670 # of 30-39yd fieldgoals made
0.105 % of 30-39yd fieldgoals scored

0.651 best turnover margin
0.596 # of turnovers committed by opponent
0.474 # of turnovers committed by own team


0.589 # of sacks made
0.414 least # of sacks allowed
0.515 # of interceptions made
0.474 least # of interceptions thrown

0.558 # of total yards per game on offense
0.511 least # of total yards per game on defense
0.513 # of yards per play on offense
0.383 least # of total yards per game on defense

0.478 # of passing yards per game
0.350 least # of passing yards per game allowed
0.040 least yards per reception (more is worse!)
0.413 least yards per reception allowed
0.626 # of passing Touchdowns
0.439 least # of passing Touchdowns allowed
0.586 % of pass attempts completed
0.330 least % of pass attempts completed by opponent

0.311 # of rushing yards per game
0.427 least # of rushing yards per game allowed
0.110 rushing yards per attempt
0.132 least rushing yards per attempt allowed
0.440 # of rushing attempts
0.745 least # of rushing attempts against

0.520 least # of 20+yd receptions allowed
0.324 # of 20+yd receptions made
0.095 least # of 40+yd receptions allowed
0.049 # of 40+ yard receptions made
0.081 # of 20+yd rushes made
0.181 least # of 20+yd rushes allowed

0.002 # of punt returns allowed
0.470 least # of punt returns

0.365 # of 2-pt conversions made
0.038 # of 2-pt conversions allowed

0.409 # of receptions
0.084 # of receptions allowed

For other correlations, just ask for either the correlations or the spreadsheet with correlations.

[MIJB#19]

I changed my mind, for those interested, you can find the spreadsheet at my website.

[Fritz]

thanks!

[Blackadar]

Ummm...how can this be?

0.676 # of points scored

That should be a 1.000 correlation.

[MIJB#19]

What surprised me so much are some additional calculations I made figuring out whter some stats categories do care for team success on offense but do not care on defense.

On offense receptions count a bit (0.4), yards per reception don't (0.0).
On defense yards per reception count a bit (0.4), receptions don't (0.1).

Any thought on the plausibleness (sp?) of some correlations?


EDIT: I hope the spreadsheet is readable...

[QuikSand]

Originally posted by MIJB#19
What surprised me so much are some additional calculations I made figuring out whter some stats categories do care for team success on offense but do not care on defense.

On offense receptions count a bit (0.4), yards per reception don't (0.0).
On defense yards per reception count a bit (0.4), receptions don't (0.1).

Any thought on the plausibleness (sp?) of some correlations?

Perhaps the general notion that a team in control of a game tends to become more conservative in its offensive playcalling helps explain this. If you're winning, you tend yo shorten up your passing game... and therefore your yds/rec would come down-- perhaps offsetting what might have otherwise been a positive correlation. I can see parallel explanations for every facet of the quoted selections above.

[Fritz]

Does one season produce a dataset that is valid for drawing conclusions?

[tucker342]

Good point Quiksand, that would make a lot of sense...

[MIJB#19]

Originally posted by Fritz
Does one season produce a dataset that is valid for drawing conclusions?

I'm not sure.
These correlations have only 32 values, though they do measure the full population for one given year.
Personally, I think unexpected results scream for more research.

A small problem is that the NFL is said to change very quikly based on tendencies seen in the previous season, meaning that if half of the teams start. That could mean that making 2001 stats count as 2002 stats reflects the tendencies for two seasons, not just one.

QuikSand, can you explain the almost non-existence of a relationship between (rushing) yards per carry and # of wins?

Personally, I've never made a big deal out of the yards per carry stat, but other people obviously do. Reading articles at site like ESPN's and NFL's, it seems yards per carry are called important.