Page 3 of 4
Re: No correlation between salary and wins?
Posted: Tue Oct 25, 2011 4:00 am
by Crow
A quick, very small check suggests that team spending increased a little in the year teams first went over 45 wins (about $2 million) followed by a larger increase in the following year ($5-6 million).
Re: No correlation between salary and wins?
Posted: Tue Oct 25, 2011 7:04 am
by Mike G
For 413 team-seasons from 1997-98 to 2010-11, avg correlation of team payroll to wins is .296 .
This is after converting total payroll to multiple of average for each season.
It doesn't get a lot better to consider multi-year intervals.
Code: Select all
interval corr.
1 yr .296
2 yr .305
3 yr .302
4 yr .288
5 yr .266
7 yr .232
9 yr .210
11yr .229
13yr .250
The 2-year interval has the highest correlation at .305
At 9 seasons, correlations are lowest. This suggests some teams are just chronically overpaying, while others are models of efficient spending.
There's no real argument for just outspending everyone, especially if you can't afford it. "Spending more" can also be relative to your own team history.
Within each team, the average correlations are strikingly different from the above table.
Code: Select all
tm yr1 yr2 yr3 yr4 yr5 yr6 yr7 yr8 yr9 yr10
Bos .58 .69 .83 .89 .89 .95 .94 .87 .92 1.00
Dal .73 .77 .83 .88 .92 .93 .92 .91 .93 .99
Hou -.05 .20 .39 .65 .84 .90 .68 .41 .32 .27
Cle .61 .66 .73 .76 .82 .90 .97 1.00 .94 .90
GSW .24 .65 .80 .92 .89 .89 .89 .89 .92 .65
Chi .63 .69 .74 .83 .81 .85 .72 .70 .61 .66
Mil .33 .44 .55 .70 .81 .79 .61 .52 .21 -.27
Ind .54 .61 .64 .63 .65 .79 .92 .95 .99 .99
Okl .26 .06 .37 .63 .80 .77 .61 .64 .53 .61
Sac .57 .63 .68 .68 .71 .75 .69 .50 .23 .17
Uta .28 .51 .73 .85 .90 .74 .24 -.20 -.37 -.91
Orl .17 .23 .17 .22 .42 .70 .92 .92 .65 .35
Den .39 .43 .47 .54 .58 .67 .88 .95 .93 .97
Mia .13 .28 .28 .40 .50 .65 .72 .67 .71 .52
Por .31 .31 .36 .44 .53 .62 .67 .68 .68 .91
tm yr1 yr2 yr3 yr4 yr5 yr6 yr7 yr8 yr9 yr10
Mem .62 .71 .73 .75 .72 .59 .34 -.17 -.56 -.32
LAC .14 .17 .26 .35 .49 .54 .52 .49 .63 .98
Det .31 .31 .31 .46 .51 .51 .40 .27 -.01 -.01
NJN .29 .35 .32 .30 .33 .41 .51 .61 .80 .84
Min .50 .48 .44 .39 .36 .32 .21 .03 -.05 -.02
NOH -.10 .18 .42 .54 .57 .26 -.15 .00 -.08 -.27
Atl .55 .58 .50 .34 .21 .16 .25 .35 .52 .54
LAL .05 .10 .18 .21 .11 .02 -.16 -.09 -.28 -.11
NYK -.13 -.02 .00 -.01 -.08 .00 .19 .50 .82 .98
Tor -.30 -.67 -.75 -.71 -.77 -.36 -.22 -.57 -.50 -.43
Phl .02 -.15 -.30 -.38 -.42 -.52 -.37 -.19 -.11 -.24
SAS -.19 -.41 -.63 -.75 -.86 -.71 -.74 -.72 -.02 .33
Phx -.49 -.47 -.49 -.48 -.54 -.72 -.81 -.64 -.58 .66
Was -.15 -.39 -.50 -.66 -.72 -.76 -.88 -.63 -.07 .75
Cha .89 .97 .99 .97 .97
NBA yr1 yr2 yr3 yr4 yr5 yr6 yr7 yr8 yr9 yr10
med .28 .31 .39 .46 .53 .62 .52 .50 .52 .54
avg .235 .273 .312 .358 .378 .402 .360 .333 .336 .396
wCha .257 .297 .335 .379 .398
In the bottom line, the Bobcats are averaged in as an equal partner, even though they haven't been around as long.
The 6-yr interval has the highest correlation, on avg, and the teams are ranked by that column.
The Median line (Cha excluded) shows an even sharper peak at 6-yr avg, and a higher number all along. Heavily negative correlations may be not really strong indicators of anything, but they drag the averages down.
I guess this is why I don't deal with 'correlations' much.
Re: No correlation between salary and wins?
Posted: Tue Oct 25, 2011 4:19 pm
by xkonk
What is the one-year correlation of a team within itself? As in, what numbers are you correlating?
Re: No correlation between salary and wins?
Posted: Tue Oct 25, 2011 6:37 pm
by Mike G
In 14 seasons, 1998-2011, there's positive correlation between team
salary and
wins.
Teams have tended to win more in years when they have had higher player payroll, relative to the league.
The correlation is smaller among single seasons, and it tends to maximize at 6 years.
For example, the Mavs have the best year-to-year correlation, at .73
Code: Select all
.yr Payroll 2011$ *Avg W
1998 27,268,948 56.0 .83 20
1999 35,250,000 62.2 .93 31
2000 39,479,500 58.8 .88 40
2001 51,905,574 67.6 1.01 53
2002 57,363,430 71.9 1.07 57
2003 72,155,176 84.5 1.26 60
2004 79,286,720 92.7 1.38 52
2005 91,553,496 104.3 1.55 58
2006 97,881,086 104.2 1.55 60
2007 88,531,846 92.2 1.37 67
2008 101,021,688 99.0 1.47 51
2009 94,830,398 88.8 1.32 50
2010 88,435,117 84.7 1.26 55
2011 84,643,496 84.6 1.26 57
Middle column is millions of dollars, relative to avg payroll in 2011.
Rising and sinking payroll is correlated to higher/lower win totals, especially over multiple seasons.
There are (5) 10-year intervals here, and the correlation between the last 2 columns, summed over each 10 years, is .993 :
Code: Select all
from thru *avg WAvg
1998 2007 1.18 49.8
1999 2008 1.25 52.9
2000 2009 1.29 54.8
2001 2010 1.33 56.3
2002 2011 1.35 56.7
*avg: Dallas' 10-yr averages have been 1.18 to 1.35 times the league avg payroll.
Similarly, there are (10) 5-year intervals, and the correlation is .923.
Chronologically, and then sorted by higher payroll avg (and wins, over 5 years)
Code: Select all
from thru *avg WAvg *avg WAvg
1998 2002 .94 40.2 1.47 57.6
1999 2003 1.03 48.2 1.46 57.2
2000 2004 1.12 52.4 1.42 59.4
2001 2005 1.25 56.0 1.40 56.6
2002 2006 1.36 57.4 1.36 57.4
2003 2007 1.42 59.4 1.34 56.0
2004 2008 1.47 57.6 1.25 56.0
2005 2009 1.46 57.2 1.12 52.4
2006 2010 1.40 56.6 1.03 48.2
2007 2011 1.34 56.0 .94 40.2
Note: In the descending-$ sequence at right, correlation is just .84 .
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 11:26 am
by Mike G
With the majority of teams above .50 correlation on sequences of 5 years and over, can we confidently say the premise has been refuted?
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 3:20 pm
by xkonk
The five and ten year correlations (really, any of the multiple year correlations) have issues in that their data include some of the same information. The Dallas 98 to 02 and 99 to 03, for example, both include 99 to 02. It isn't too surprising that those numbers line up better than the single-year values; Dallas would have had to have done something fairly striking in 98 or 03 to move the five-year average enough to get the correlation out of line. Generally you don't want to combine overlapping data to create your variables because you lose any independence in the observations.
In regards to the premise being refuted, maybe we should take a slightly different tact? What exactly do we really want to say if there is a correlation between salary and winning? We don't want to say that spending more causes winning, right? Because spending by itself is obviously useless; you have to spend the money on good players, and good players win. Do we want to say that spending more allows teams to get 'unfair' numbers of good players? Do we want to say that spending more allows teams to recover from mistakes? Do we want to make arguments about teams' ability to identify good players, a la Wages of Wins (I assume not, given the general perspective here). Let's assume everyone was willing to agree that there's a weak-to-middling league-wide correlation between salary and wins. Now what?
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 4:20 pm
by Mike G
You can create non-overlapping intervals, say 3 years, 1998-00, 2001-03, etc, and get different correlations. But if you also do 1999-01, 2002-04, etc; and 2000-02, 2003-05, etc; then you have 3 different correlations covering the same 14 years.
And when you average them, voila, you get the same total correlation as you get just using every 3-season interval, overlapping and not.
If all overlap created higher correlation, then longer intervals would have higher correlation, yes?
But each team is different. Bos peaks at 6 yrs and 10 yrs; Dal at 6 and 11; Cle at 8 yr; GS at 4 and at 9; Atl at 2 and 10, abruptly falling off at 11.
See the table of teams, above.
Above 6 year intervals, maybe teams are ready to 'blow up' what they've got? Teams which have underspent are ready to try heavy spending? Teams which overpaid are self destructing?
Teams (47 of them) which have spent less than .80 of the year's average payroll have averaged 31 wins.
The 50 teams spending at least 1.20 of avg have averaged 48 wins. Without the Knicks, it's 52 wins.
The tightwads have gotten 15% more wins per dollar, than have the free spenders.
How about this for a statement: You can't expect to win in the NBA, if you aren't willing to spend for it.
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 4:44 pm
by EvanZ
Mike G wrote:
How about this for a statement: You can't expect to win in the NBA, if you aren't willing to spend for it.
I would amend this to say you can't expect to win, but if you do win, it won't be for very long.
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 4:58 pm
by Mike G
xkonk wrote:The five and ten year correlations (really, any of the multiple year correlations) have issues in that their data include some of the same information. The Dallas 98 to 02 and 99 to 03, for example, both include 99 to 02. It isn't too surprising that those numbers line up better than the single-year values; Dallas would have had to have done something fairly striking in 98 or 03 to move the five-year average enough to get the correlation out of line. ..
Honestly, I've waked up at night and thought about just this. From one interval to the next, the Mavs lose one year (1998 in your example) and gain another (2003).
If 1998 offers a negative correlation, and 2003 is positively correlated, then the later interval has a higher correlation than the earlier one; and vise-versa.
To compare, say, a 4-year interval to a 5-year interval, again it depends on whether that 5th year is positively or negatively correlated.
Every negatively-correlated year that's added to the interval reduces the correlation of the interval. The magnitude of the change is muted, but not the direction of the change.
Multi-year intervals offer additional insight, in that they spread out the changes in wins and salaries, which may not coincide in a given year, but often take some years to pay off.
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 10:20 pm
by Chilltown
Anyone have thoughts on Tom Haberstroh's article which attempts to assign relative importance to drafting and spending in terms of producing wins?
http://espn.go.com/blog/truehoop/post/_ ... lance-myth
Re: No correlation between salary and wins?
Posted: Wed Oct 26, 2011 10:39 pm
by Crow
Did Tom do a multi-variable regression analysis? It sounds like he may have but the description is quite brief. Perhaps he will clarify this? That is what is called for here though how you handle scoring draft performance and other things (was coaching / analysis included as its own variable and if so how?) will affect the outcome for those variables and indirectly for salary.
What accounts for the other 59% of the explanation of team performance besides drafting and "spending"? Is that total spending or spending apart from drafting and apart from anything else also included as separate variables? I am not sure how you fairly separate "player talent / performance" from spending (if it is in some fashion) or management of that talent.
Re: No correlation between salary and wins?
Posted: Thu Oct 27, 2011 12:56 am
by EvanZ
I don't think anyone doubts that the draft is important. That seems like a strawman to me. After your rookie contracts expire, you still need to re-sign them, and you need money to attract other elite free agents to pair with them. What I don't know that I'd like to know is whether certain owners are simply being cheap or really don't have the money.
Re: No correlation between salary and wins?
Posted: Thu Oct 27, 2011 2:14 am
by Crow
Using the reported net worths of the owners here
http://hoopshype.com/owners.htm
I computed the correlation between last 6 years of spending and net worth for those owners who had been owners for the full period and there is almost no correlation (+.02).
There was a modest (less than +.2) correlation between years of ownership and 6 year spending and a roughly similar sized correlation between conference titles and recent spending.
Using financial data from Forbes, for the 2008-9 season there was a .69 correlation between team revenue and spending on player salary. But only a .10 correlation between net income and spending. Profit and loss didn't seem to much hardly at all for spending.
Re: No correlation between salary and wins?
Posted: Thu Oct 27, 2011 2:36 pm
by EvanZ
Anybody read
this interview with economist Kevin Murphy. He seems to have a similar position to Zimbalist. But I found this quote interesting (and somewhat puzzling):
KM: The relationship between salaries and the number of wins in a season is positive, but it's pretty weak. It certainly is not going to have a dramatic change in the distribution of outcomes. It might change who the winners are and who the losers are, but you're still going to have some teams that are much better than others. Because some people spend their money much more wisely than others do.
Isn't the bolded part kind of important? Maybe I'm missing his point.
Re: No correlation between salary and wins?
Posted: Thu Oct 27, 2011 5:40 pm
by Crow
He might have meant that it could affect playoff winners and losers at the margin amongst the best teams but there is still going to be good and bad groups (with salary range within them).
But looking at the last 6 years, the top half spending teams won 2/3rds of the playoff series and using my weighted value system for different rounds of the playoffs they also won more than 3/4ths of the points available. Exclude the Knicks and just the next 5 biggest spending teams won nearly half the playoff series and nearly 2/3rds of the weighted playoff value. The top five spenders besides the Knicks won more than 3 times the playoff series and more than 3 times the playoff value of the next 10 teams combined. So at group level spending appears to be very correlated to playoff success between the top half of spenders and the bottom half, and between the top third of smart spenders (excluding the Knicks) amongst above average spenders and the rest of that group.