Do you think historic results in certain matchups have predictive value?

[davopnz]

In my model I only use the current season's data. I'm always looking for ways to improve it.

One thought I've had is incorporating historic data, mainly for totals.

Lets say my model spits out a total of 50 using the current season's data, but 20 years of matchup data shows an average of 40 points in this matchup. Do you think that is statistically significant?

One way to potentially validate this would be to compare the matchup total against league average for every season. If the total is regularly below league average then that would suggest it is statistically significant.

Also, I'm betting on small leagues. Obviously if this was NFL it would already be baked in, if statistically significant.

Input would be much appreciated, thanks!

[JacketFan81]

I definitely think the historical matchup data matters. Really depends on the sport, but in a game like say basketball, you might have an opponent that your team or coaches just hate. Game plan might change to playing slower because their transition/tempo always trips up your defense, don't want to make any mistakes or miss opportunities. So I guess what I'm saying is that if the historical matchup is something in the way of a rivalry or an important game, you might as a coach/player make alterations to your team's (for the season) game plan that aren't necessarily accounted for in the numbers. If that makes sense. Just my two cents.

[Ian]

IMO data like this might be helpful for ferreting out individual bets, but for a model I would be extremely cautious.

Coaching and personnel changes could really sabotage something like this. For example, in CBB a coach with either a pressing style or a deliberate motion offense style will drastically effect totals. Once that coach leaves his respective program the data from that coach's era will be all chaff and no wheat for your model.

Also, even without a coaching change a team may change style just based on the players on the field/court at that time. For example, during the peak of the Tom Brady/Randy Moss era of the Patriots dynasty, Belichick had the offense airing it out and snapping the ball with a lot of time on the play clock so that their offense could press their advantage to its fullest as many times as possible. During the end of the Brady years when the O line sucked and the defense was great, Belichick had the Patriots offense milking the clock and lining up in the I formation more than any other team. So even though the Patriots were the same dynasty the whole time, using historical data for totals at any time during that dynasty would put a bettor on the wrong side virtually every game.

Just my opinion...

[Don_Omarion]

depends on the performance of both teams recently in their last matches, beside few care for historic data.. which means if you think team X recently is better but historic data in the side of team Y , You fade the game as an idea ... that is my system and it works well.

[semibluff]

Context is most important. Context is often just current form. However, if it's a local rivalry or a divisional match-up it might be best to completely ignore current form. Sometimes both teams are expected to take a cautious or negative approach to a match-up, (eg Ravens @ Steelers). That can move the totals, and consequently the handicap, away from projections. Sometimes a match-up will go the other way and be over-aggressive, (eg Chiefs @Bills). Sometimes teams are put together to do well in certain conditions, (eg cold weather or in a dome). It's also different if 1 or both teams is out of the running. It's all context.

General answer to your question: Yes. If your model spits out a total of 50 using the current season's data, but 20 years of match-up data shows an average of 40 points in this match-up I would think that was statistically significant. If the model is a long way away from historical data and the current betting line I would ignore the model in those circumstances. Models don't have to work for every situation for them to be useful. Knowing both when they do and don't work is very useful.

[Waterstpub87]

How can you only use this seasons data?

If you are waiting till mid season, I get it.

Of course you need to use historic data. If you want to see how accurate it is, a quick test would be to pull a correlation between years for a statistic.

I wouldn't put much stock into past matches between the exact two teams. You suffer from small sample sizes.

[trytrytry]

rules and approach to game changes too much for that imo

[davopnz]

rules and approach to game changes too much for that imo

That's why I thought it would be sensible to compare to league average for each individual season.