## List of Stats and Correlations

I thought it would be interesting to list all the correlation statistics in one place. These are based on all 32 teams from each of the 5 seasons between 2002-2006. Each variable is listed here with its correlation with season wins, offensive points scored, and defensive points allowed. (0.15 for p=0.05 significance) The list is sorted 3 ways according to the strength of each correlation. Click on the tabs at the bottom of the page to sort the list according to the correlation you wish.

One caveat: some of the stats that are highly correlated with wins are not that useful. For example, extra points attempted' only tells us how many TDs a team scored. We already know that TDs are how you win in football. 'Kick return yards allowed' would be another example. The more TDs a team scores, the more total yards it will have given up on kickoffs.

### 9 Responses to “List of Stats and Correlations”

1. David B. says:

Interesting, and very useful.

However, I would add one VERY CRUCIAL stat. NET POINTS.

2. Happy says:

Brian,

Have you done much research on what causes teams to deviate from expectation? For example, say you use the strongest correlations to formulate your weekly probabilities. Now suppose that over a period of time (season?) a given team loses three games they were expected to win and wins two games they were expected to lose.

Let's say the three games they lost were games where they faced short stocky running backs while the two games they won were against tall lanky running backs. I would be interested in studying what causes deviations from the simple model. These delta functions could then be used as refinements.

Things such as height to weight ratio of primary running back, total weight of offensive and defensive lines, height of #1 and #2 receivers, etc. Obviously, it couldn't be too complex and it would be hard to find statistically distinct relationships based on sample size (unless they could be extended across all teams).

Just a thought. Curious if you've studied it. You've obviously done a lot more work on this than I have. Great site.

Happy

3. Robert says:

It would be interesting to run a stepwise regression and find 3-4 variables which jointly predict wins effectively. Neat correlations; I knew turnovers were important, but wow.

4. Anonymous says:

5. John says:

Need this link to get working!

6. Matt says:

7. Anonymous says:

"Each variable is listed here with its correlation with season wins." There is a broken link to your list.

8. Anonymous says:

The link to the list is not working. We've been trying to find the correlation between turnover ratio and winning percentage which you seem to have but we can't see it.

9. CEO says:

I've found the correlation to be much stronger between turnover DIFFERENCE, rather than turnover RATIO.

For 3,312 regular season games between 2000 and 2012, the (Pearson) correlation between TO diff (= vis TO - home TO) and win pct was 0.964 (0.977 if you band together the infrequently occurring extremes into "more than/less than" caps).

The Spearman Rank Coeff is 0.985 and 1.000 under the same conditions, respectively.

There are some other really interesting pieces of information you can discover once you start digging into game-by-game turnovers...