How many games for past performance?

**Nookx** · 12-13-10, 01:00 PM

It depends on what your looking at so there is no one way set an exact number of games.

**Wrecktangle** · 12-14-10, 08:11 AM

For the NBA, you should have a db of the last 3 years at a minimum, 5 is better, at least 10 should be your goal.

If you are talking about smoothing data, there are all sorts of ways to do this, and methods vary according to smoothing window. No one smoothing window works for all instances, shorter windows are good in Reg Season, longer ones are better for Playoffs.

**alling** · 12-15-10, 09:04 AM

Originally posted by Wrecktangle

For the NBA, you should have a db of the last 3 years at a minimum, 5 is better, at least 10 should be your goal.

If you are talking about smoothing data, there are all sorts of ways to do this, and methods vary according to smoothing window. No one smoothing window works for all instances, shorter windows are good in Reg Season, longer ones are better for Playoffs.

Im interested to know the number of games in a matchup.

**Inspirited** · 12-15-10, 04:01 PM

I've been using 5 games mainly just because I can easily get stats from the last 5 games from NBA.com all ready made up for me. I've found some interesting stuff, but who knows if these resulsts will hold up.

**jscol** · 12-15-10, 08:49 PM

I am going to be doing last 5 games as an addition to my model, but haven't found last home/away split or something like it. I think the home/away split is important. Does anyone know where this could be listed?

**CaptainPrice** · 12-15-10, 09:10 PM

from a stats standpoint I look at past 4-8 games

**Richards** · 12-16-10, 06:41 PM

I've been experimenting with a model that uses 5 games, and instead of making direct adjustments for home court advantage, I just weight the ratio of recent home games to recent away games to one side or the other. e.g last 3 home games and last 2 away games for the home team. Vice versa for the visiting team take their last 3 away games and last 2 home games.

Why not just take the last 5 home games for the home team? I think the data that is weeks or months old is too stale at that point. I'm a big believer in recent form being a better predictor than month(s) old performances.

However, I DO add in a sixth "game" taking season average numbers added in as a 1/6 of the predictor...so the older data is there it's just weighted much less.

**Whitefin14** · 12-17-10, 10:16 AM

5 games is usually the way to go.

**Wrecktangle** · 12-18-10, 08:29 AM

Originally posted by alling

Im interested to know the number of games in a matchup.

Between the same teams over time? We need some precision here, 10 posts into this thread and we still haven't a clue as to the question.

**roasthawg** · 12-18-10, 11:59 AM

Season to date.

**classhandicapper** · 12-18-10, 07:18 PM

It seems to me that teams often randomly have good and bad stretches of 5 games. Quite honestly, I can't imagine using such a small sample UNLESS I have reason to believe that a teams is either permanently better or worse than the games further back. For example, it's easy for me to dismiss Miami's early games and look at their last few because they had so many new players and they may be just starting to click now. But for most teams though, their last 5 games isn't enough. The Hornets were great, then terrible, but now they are starting to click again. So how good are they really? It's probably the long term average and not any of the streaks.