|
Q: What is the best timeframe to use when looking at a team's stats (season to date, last 4 games, specific home and away numbers, etc)?
A:
This has been a long debated question among handicappers, and unfortunately there is no easy answer since the particular statistics you are looking at will have a great deal to do with which data range is optimal. Also of importance is the amount of adjustment you make to the raw numbers to reflect various issues (playing location, strength of opponent, etc). Having said that, we will take a first look at this question by focusing on one stat -- net points per game -- in a variety of breakouts. "Net points per game" is simply the average margin per game of a team: if a team has played two games winning one by 6 points and another by 3 points, their "net points per game" over the 2 game span would be +4.5
We've elected to use the following ranges for this analysis: Season-to-date, last game, last 2 games, last 4 games, last 5 games, last 8 games, and finally home/away specific. Since some of these measures require a lot of games, we will only begin looking at matchups from week 10 on in the season (and include the playoffs) since that way all of the ranges are applicable in each game and none are duplicated. The database covers the years from 1983-2000.
To analyze the predictive value of each range we take the net points per game for each team, add in a constant home field advantage value (with the exception of the home/away specific category), and thereby create a projected margin. We next select a "pick" for each game based on the difference between the net points "line" and the actual line. We will then show the results for all games, and then only those games with at least a 4 point and 8 point difference between the projected score and the line.
Examing different data ranges using NET POINTS
| Range |
All games W-L |
All Win% |
4+ Overlay W-L |
4+ Win% |
8+ Overlay W-L |
8+ Win% |
| Last 8 games |
1052 - 1020 |
50.8% |
535 - 492 |
52.1% |
190 - 167 |
53.2% |
| Last 5 games |
1050 - 1027 |
50.6% |
664 - 612 |
52.0% |
343 - 312 |
52.4% |
| Season To Date |
1047 - 1035 |
50.3% |
426 - 391 |
52.1% |
104 - 77 |
57.5% |
| Last 4 games |
1033 - 1028 |
50.1% |
715 - 696 |
50.7% |
404 - 379 |
51.6% |
| Last game |
1027 - 1032 |
49.9% |
863 - 887 |
49.3% |
693 - 717 |
49.1% |
| Last 2 games |
1019 - 1036 |
49.6% |
811 - 812 |
50.0% |
570 - 568 |
50.1% |
| Home/Away |
1024 - 1058 |
49.2% |
607 - 614 |
49.7% |
291 - 282 |
50.8% |
ANALYSIS:
So what we have found is that the best range to use for this "net points per game" number would probably be last 8, followed closely by last 5 and season to date, which all show slightly better than 50% overall, and 52% or better on the 4+ difference games. Interestingly the "Home/Away" specific data set performs the worst overall, while the last game number is pretty poor, as would be expected.
This is by no means conclusive of course for other factors, but the general gist of the above appears to be that the more "recency focused" categories are less predictive, at least without some further adjustments. The "Last 5 games" is an interesting range since in almost all cases a team will have at least 2 games both at home and away, while a "last 4 games" range can often have a skewed home/away basis (eg 3 home games, 1 away or vice versa).
Based on preliminary research of our drive chart and play-by-play ratings, we found that "season-to-date" numbers outperformed the other ranges we examined. As we get more years worth of data, we may find that this is no longer the case, and so we now provide a "Last 5" breakout as well for comparison purposes.
Additionally, we will return to this question again, with a look at some other factors.
|
