Making a line with the "First Down Conversion" numbers

The following research is based on the new rating system discussed in the Accurate First Down Conversion article in the main research section.

The fact that this new scheme had the Bucs as the #1 team in the league, and the Raiders as the #2 team (using regular season data) is a good sign that there may be some legitimate value in the stats as a predictive function.

To set out to test the merits of the stats for forecasting future games, we built a database with the season-to-date averages for these numbers going into a particular matchup. The only thing left to resolve prior to running through the season, was how to apply the stats for projecting a score. To keep things simple, and seeing as how the third best team (Philadelphia at +6.3% when you take overall offensive series conversion minus overall defensive series conversion) compares to the third worst team (Arizona at -9.1%), we figured we'd start by assuming each 1% of net rating is worth one point of spread value.

So as an example, if Philly and Arizona were to meet, we'd say the line is Philadelphia favored by 15.4 points (+/- home field advantage depending who the home team was). This seems reasonable, and given the fourth worst team was -6.4%, you basically are going to have fairly normal spread predictions with the exception of the real extreme teams.

Other details are: we used a rolling season-to-date average for our data set, and all tests include all games from week 5 through the conference championship playoff events.

Comments: not too shabby, hitting 57% playing ALL games from week five on. However, the disappointing element is that the large overlays were not good plays, whereas the tiny overalys were outstanding. The underdogs were not surprisingly excellent picks (62%) given that the dogs as a whole were barking loudly in 2002. As is the case with many of our other statistical indicators, when the selection is an underdog receiving five or more points on the line the results are an electric 21-8 (72%).

As luck would have it, our decision to set the value of 1% = 1 point was a good one, since other weights were not as good (.75 points/% = 56% overall, .5 points/% = 53%, 1.25 points/% = 55%). As the stats are based on a fairly large number of events (eg teams will often have 30 or more down series in a game) we didn't think there would be a need to wait beyond the week five typical stats kick-in phase, however to our surprise, there was a marked improvement if you set the filters to only look at games from week 10 forward:

Comments: in the 2002 season at least, these numbers were incredibly potent in the second half of the season! A 61% mark on every game would be well worth incorporating into the overall handicapping picture if this could prove to be even close to a consistent performance level. What's nice about the above results is that the picks were profitable in every sub-category, including the dreaded away favorites.

Part of the reason why starting the projections up later appears to make sense is that teams do show very high net stats in the week 5 period. If the matchup then was Detroit at Indianapolis, the prediction would have been Indy by close to 40 points...a little overboard to say the least. The difference between best and worst teams was 33 points in week six, 32 points in week seven, 25 points in week eight, 24 points in week nine, and 23 points by the time you get to week ten.

The good news on top of all this initial feedback is that these numbers are data points we can create without bias for another four seasons. Clearly then the next stage will be to calculate the ratings for these previous years and recheck the results against the line. Stay tuned!