this guy is awesome

[Indy Coug]

Originally Posted by ChinoCoug

ha. yeah right. that study was quite robust. keep remaining in denial.

In the interest of time this morning, I'll link this for beginners

http://www.sabernomics.com/sabernomi...eferees-again/

[ChinoCoug]

Originally Posted by Indy Coug

In the interest of time this morning, I'll link this for beginners

http://www.sabernomics.com/sabernomi...eferees-again/

“But most of the players in the leagues are black” or “The tops foulers in the league are white.”

The authors of the study use fixed effects.

"The granular data is not available to the public, so we have to choose to live in a world where he accept to deny the inference of this study…except that the data could be made available by the NBA. The main argument coming from the NBA is that they have done a study with granular data, and that they do not find the same result."

I say we subpoena the NBA for this data.


You can download the study yourself:
http://graphics.nytimes.com/packages...race-study.pdf

[ChinoCoug]

Originally Posted by Cali Coug

That link seems to suggest that the study has merit. It doesn't seem to argue that it is one of the "biggest statistical farces" as you claimed.

Am I missing something?

No, Indy was bluffing. He'll get away from asymmetric information if you let him.

[Indy Coug]

Originally Posted by Cali Coug

That link seems to suggest that the study has merit. It doesn't seem to argue that it is one of the "biggest statistical farces" as you claimed.

Am I missing something?

Insufficient data granularity. Not properly taking into account concentration of race by position.

Are most white players post players that tend to pick up more fouls than black guards and small forwards?

Do players that play in "garbage time" tend to pick up fouls at a higher rate? A majority of white players are bench players.

http://bpp.wharton.upenn.edu/jwolfer...aceSummary.pdf

What this results in is only 4% more fouls? By the biggest change is from 0% white to 33% white refs. After that, the rates are nearly flat (4.317 to 4.322 black players, 4.992 to 4.897 white players for 33% to 100% white referees)

How many games were refereed by all blacks or all whites? Could it be that those cells had much fewer data points than 33% or 67% white refs and therefore resulted in higher variability? Well, if you look at the standard error in the 0% white refs, you can see that it has the largest standard error by a factor of three.

In fact, upon looking at the paper in detail, you see how much smaller the 0% white referee grouping is

0% white - 668 games
33% white - 4,928 games
67% white - 11,580 games
100% white - 7.350 games

http://bpp.wharton.upenn.edu/jwolfer...rs/NBARace.pdf

So in other words, we have a study that has significantly incomplete data, a failure to sufficiently normalize stats and is hanging the bulk of their "Black minus White differential" on a data cell that only accounts for 5% of the total games.

As far as statistical significance goes, you might be able to show statistical significance, but how does that stack up against PRACTICAL significance?

A total of 4% more fouls? For all practical intents and purposes, that's nothing. The authors put in a limp-wristed disclaimer that this study doesn't intend to show racist intent, but obviously from ChinoCoug, et al, that seems lost on them.

Is it worth throwing out a study like this which elicits a stronger public reaction than the data, even if it wasn't flawed, would merit? No.

That's why I had such a strong reaction to this.

[MikeWaters]

I've sat through hymns where people were asked to stand. No one shot me, or grabbed me by the shoulder and escorted me out.

[pelagius]

Originally Posted by Indy Coug

Insufficient data granularity. Not properly taking into account concentration of race by position.

Are most white players post players that tend to pick up more fouls than black guards and small forwards?

Do players that play in "garbage time" tend to pick up fouls at a higher rate? A majority of white players are bench players.

Indy, I don't really care about this paper and maybe the data granularity charge is somewhat fair (although, I am inclined to think this works against the study finding something). However, I don't think your second charge is fair at all. The empirical design does take into account concentration of race by position. The specifications that include player fixed effects specifically take this into account. I don't think mainstream stats guys use fixed effects that often (as opposed to economists), but it really is a good way to control for the problem you raise about "concentration of race by position".

[ChinoCoug]

Originally Posted by Indy Coug

Insufficient data granularity. [Not properly taking into account concentration of race by position.

Are most white players post players that tend to pick up more fouls than black guards and small forwards?

The study only looks at technical fouls. Do we have any theoretical reason to believe there are positions that get more technicals than others?

Originally Posted by Indy Coug

Do players that play in "garbage time" tend to pick up fouls at a higher rate? A majority of white players are bench players.

Again, the study only looks at technical fouls.

[pelagius]

Originally Posted by ChinoCoug

The study only looks at technical fouls. Do we have any theoretical reason to believe there are positions that get more technicals than others?



Again, the study only looks at technical fouls.

Chino, I don't know why Indy attacked this paper so much. The empirical design is pretty solid (although I have some quibbles). But why do you keep repeating the technical foul comment? It is not true. Most of the paper uses all fouls (fouls per 48 minutes). Table 5 does include specifications with "Technical fouls" as the dependent variable but those regressions have no power: the coefficient isn't significant. What am missing?

[ChinoCoug]

Originally Posted by pelagius

Chino, I don't know why Indy attacked this paper so much. The empirical design is pretty solid (although I have some quibbles). But why do you keep repeating the technical foul comment? It is not true. Most of the paper uses all fouls (fouls per 48 minutes). Table 5 does include specifications with "Technical fouls" as the dependent variable but those regressions have no power: the coefficient isn't significant. What am missing?

oops. I misread this from the blog: "However, I would also like to point out that the paper looks at technical fouls." I was being careless. No "only" in there.

[ChinoCoug]

Originally Posted by Indy Coug

As far as statistical significance goes, you might be able to show statistical significance, but how does that stack up against PRACTICAL significance?

A total of 4% more fouls? For all practical intents and purposes, that's nothing. The authors put in a limp-wristed disclaimer that this study doesn't intend to show racist intent, but obviously from ChinoCoug, et al, that seems lost on them.

They found that "The bias in foul-calling is large enough that the probability of a team winning is noticeably affected by the racial composition of the refereeing crew assigned to the game." That's the practical significance.