APBRmetrics

Should I be adding additional factors such as the number of days since the last game into my model? I'm curious if anyone knows about factors that have either been shown or that people believe impact performance on a team level. Distance traveled by teams to away games is one I've found.

Have there been studies regarding what factors impact home court advantage? I have to apologize for being horribly ignorant of the analytical research that has been done for the NBA.

-Chris

I don't know the exact numbers but # of rest days has an impact, and whether games the night before went into OT has an impact. Also, different teams have different HCA

In terms of rest the strongest statistically significant fact (I believe) is that if one team played the night before and the other didn't, then the one that didn't gains about 1.8 on the expected score difference for the game.

We have that number at around -2.5 if its the visiting team and around -1.5 if its the home team (ceteris paribus).

We also show milder effects for things such as having played 3 out of the last 4, or for having 'too much' rest.

Obv another issue is that Denver/Utah get an extra 1-1.5 points for altitude.

There's also a jetlag effect for teams flying West to East by more than 1 timezone.

Much more too but those are some of the main ones.

talkingpractice wrote:We have that number at around -2.5 if its the visiting team and around -1.5 if its the home team

That's quite interesting. I tried using separate identifiers for home and away back-to-backs but found that whilst there was about a 0.15 difference (actually worse if the home team is unrested in my case) that when the BIC was considered for each model that a combined value was preferable.

Wow, I'm very interested by your result.

1, We're AIC folk, but I can't see how the AIC vs BIC distinction could be the cause of such a different result.
2, We used a dataset from 1999-2010... did your data overlap?
3, One other thing that maybe is giving us different results is that we considered many 'bad fatigue' situations separately but then found (similar to what you said re home and road) that we're better off with one model for all of those situations (ie 3in4 b2b, 4in5, etc).

My dataset was 2002-2012 so yes they will be mostly the same. Strength estimates in each game were created through a player-ratings system incorporating box-score data and game outcomes, though similar tests using team-level ratings came up with the same result. I should say I only tested back to backs in this instance and have't yet settled on the best way to handle 4 in 5 and other fatigue situations, so my guess would be that that is where the difference is appearing. And yes, the AIC was also preferable for a more compact model.

Here is what I remember and could find quickly on the topic:


http://www.nbastuffer.com/2009-2010_NBA ... Stats.html
http://www.nbastuffer.com/rest_days



http://wagesofwins.com/2012/01/13/artur ... ba-season/

http://arturogalletti.wordpress.com/201 ... advantage/


viewtopic.php?f=2&t=7879&p=11141&hilit=altitude#p11141


http://stat.wharton.upenn.edu/~dsmall/n ... mitted.pdf
http://teensneedsleep.files.wordpress.c ... layers.pdf

We originally followed the methodology in this article: http://stat.wharton.upenn.edu/~dsmall/n ... mitted.pdf, and then (we think) improved on that method afterwards.

Yeah I've read that one several times, it's a nice paper, very useful - anybody interested in analysing rest would do well to use it.

I'm currently looking into using an exponentially smoothed method to consider situations other than back-to-back.

Excellent, adding in whether teams had a game the previous day gives me a nice boost in predictive power. I get that you score 1% more points per possession and allow 0.8% fewer points per possession if you had a rest day vs playing the day before.

-Chris

Do you guys think rest is already "built in" to home court advantage? For example, I usually look at days of rest for each team, but generally home teams have more rest than away teams.

So I guess my real question is, if I already have adjusted each team's ORtg and DRtg for days of rest, is taking into account home court advantage as well an instance of double counting or are the two independent?

I did some work on rest days and impact on pace and efficiencies a couple of years ago (see the last post here: http://godismyjudgeok.com/DStats/APBRme ... =2661.html )

Interested in seeing where this topic goes.

evana1234 wrote:Do you guys think rest is already "built in" to home court advantage? For example, I usually look at days of rest for each team, but generally home teams have more rest than away teams.

So I guess my real question is, if I already have adjusted each team's ORtg and DRtg for days of rest, is taking into account home court advantage as well an instance of double counting or are the two independent?

I would describe home court advantage as the advantage that a home team has when rest is already taken account of. These are probably referee bias and general player comfort issues (fight or flight). The fact that home teams are usually better rested can confound the issues, but as long as rest appears in your regression as separate from HCA then the regression should do a decent job of separating them.

v-zero wrote:

evana1234 wrote:Do you guys think rest is already "built in" to home court advantage? For example, I usually look at days of rest for each team, but generally home teams have more rest than away teams.

So I guess my real question is, if I already have adjusted each team's ORtg and DRtg for days of rest, is taking into account home court advantage as well an instance of double counting or are the two independent?

I would describe home court advantage as the advantage that a home team has when rest is already taken account of. These are probably referee bias and general player comfort issues (fight or flight). The fact that home teams are usually better rested can confound the issues, but as long as rest appears in your regression as separate from HCA then the regression should do a decent job of separating them.

That's precisely what I did in my analysis.

One confounding issue I failed to deal with: I accounted for team strength over the course of the season. Individual games coaches *may* strategically concede/give up on early--like Popovich does overtly. In other words--the team may not be as good on a 4th in 5 game, purely because they are resting players somewhat. This would exaggerate the effect that I found. To account for that, I would need to account for exact minutes played and use a player rating system to estimate actual team strength for the given game to account for rest accurately.