Combining drive chart predictions with turnover difference theory We've been cranking out projected scores based on our exclusive TMW drive chart stats since 1997, but have always wondered if incorporating the turnover difference theory into the projections would improve results...

A promising tool for predicting NFL outcomes

One thing that has troubled us through the years though is that often times the drive chart predictions are in direct conflict with the turnover difference theory recommendations. This is not surprising in that the drive charts rate highly the good statistical teams with strong effectiveness, while turnover difference tends to side with teams of questionable merit from a numbers standpoint in the belief that they are under-rated through their turnover shortcomings.

It is not uncommon to see the drive score projections and the T/O Difference picks aligned almost perfectly against each other in a given week. As both approaches have been potent indicators of point spread success, the question was how to combine them to up the intensity and overall win percentage while reducing the number of games passed on due to "conflicting information."

Attempts to settle matters by simple consensus (both the DC and T/O agree, etc) were not especially exciting, and so we went back to the drawing board and added some significant changes to the basic drive chart projected score formula to include the aspects of turnover difference we believe to be most predictive. It should be noted that this proprietary algorithm for projecting scores was created in advance of testing the predictions against actual historical data.

We have a five year database of complete drive information (1998 through the 2002 season) and when we ran our new scheme against the games from week five forward (under the assumption that you want teams to have played three or four times before taking the numbers too seriously) we produced the following cumulative results:

Wow! While we knew going into the test that the drive chart predictions had been solid for this five year span, and likewise turnover difference had as a whole maintained strong results in the core week 5 to week 12 range of games, we did not expect that our new score algorithm would yield a prediction tool that had hit 57.8% playing all games from week five forward!

The level of overlay is not particularly important, as even the smallest overlays (1/2 point to 1.5 points) hit at a 55.6% clip. These numbers are all against a neutral, consensus line, which means that a bit of line-shopping could up the overall percentages by another point or two!

In an era where underdogs were significant better bets, the dog picks were stronger, although favorites selected by this new score system were a more than respectable 56% against the consensus line. Also of note is that the predictions were at their best in games foreseen to be close -- when the favorite was giving 4 1/2 points or less to the underdog -- where the record was a blazing 59.7% against the spread.

All this again to be emphatic is based on using a new formula that was not adjusted or manipulated to backfit the data (other than knowing that both elements of the drive stats and turnover difference had held reasonably well in their own right).

Consequently we will be addding a new feature to our 2003 weekly coverage, which will be these DCTO Projected Scores and if the results for this coming season are anything like the ones for the past five years, then it should be a very welcome addition.

See Also:
DCTO Predictions by Year (Subscriber Feature)
DCTO Predictions by week number (Subscriber Feature)