- --- In APBR_analysis@y..., "Michael K. Tamada" <tamada@o...> wrote:
> won-loss records to calculate a player's probability of not being ahypothesis: "Player
> replacement level player (I forget what won-loss percent he used as
> replacement level). I.e. he literally created a null
> X is a replacement level player, with a won-loss % of .abc" andtested
> those 5-2, 8-8, or whatever records against that null hypothesis.would be
> Players who were close to .500, but who played a lot of games,
> very unlikely to be replacement level players. Ditto players whosegames.
> records were far above .500, and who had a reasonable number of
> A clever idea, a nice try, but James eventually dropped it, I
> because he found the same thing that I did: the resultingprobabilities,
> if used to rank players, resulted in strange-looking, not very goodsolve
> rankings. Fiddling around with different replacement levels didn't
> the problem.I tried it, too. I forgot about it until you mentioned it. I like
your other idea, though, a little better.
> A better solution would be something like this: figure out what the
> replacement level is (something like .415 or something like that
> be popular in baseball, it is likely to be a different number inrecord,
> basketball). And simply calculate wins above replacement. A 5-2
> compared to the 2.9-4.1 record that a replacement level playerwould have,
> is 2.1 wins above replacement. An 8-8 record, compared to the 6.6-9.4
> that a replacement player would have, is 2.4 wins above replacement.incorrect, the
> So in your example question, my off-the-cuff estimate was
> 8-8 player contributed more to the team.OK. Defining replacement value is an interesting challenge. James
used 0.350. I always felt like basketball's number was lower, maybe
about 0.200-0.250. If baseball's worst teams win about 35% of their
games and that corresponds to replacement level, then shouldn't
basketball's worst teams set replacement level? Basically an average
player on the worst team is probably a bum. Think of all the guys on
those rebuilding Chicago teams as pretty much replacement level (that
includes, interestingly, Brent Barry).
Conceptually, I'd think this replacement level number goes down with
increasing league size. More players means more bad players, in
general. Does that make sense?
I really do like this idea, though, MikeT. If we say that 0.25 is
replacement level, then Duncan's 17.1-0.7 record is 12.7 games above
replacement level. McGrady was 8.7 games above replacement level.
Ewing was 0.3 games above replacement level. Troy Hudson was 0.1
games above (though I know a lot of guys like him more than that, he
has generally been considered pretty close to replacement level).
And by this measure, sure enough, I do find that Erick Dampier has
edged out Michael Olowokandi for farthest below replacement level (-
2.0 to -1.5). Actually, I didn't do the calcs for all players, but I
did a few others and even Jason Spazz Williams didn't quite match
> So in your example question, my off-the-cuff estimate was
> 8-8 player contributed more to the team.No, you were actually originally right. 8 - 6.6 = 1.4, not 2.4 as
you wrote down. The breakeven point at which these 2 guys are
equivalently above the replacement level is 0.330.
- Yeah, these are things to keep in mind as we try to zero
in on the replacement level. Some so-called replacement
level players are clearly better than others. Are they
thus truly "freely available"? There's only one Rod
Strickland; once he's signed, it's not possible for any
other team to go out and sign him. On the other hand,
he was in a sense available to any team that wanted him.
I suspect we'll want to look at a bunch of players
identified as replacement level and find their average value,
to even out the occasional Strickland who brings up the
average and the occasional Oakley or whoever who brings down
In other words, something more akin to KevinP's look at the
average performance of free agents, instead of the 290th
man approach. Although it turns out that they yielded
very similar values.
From: John Hollinger [mailto:alleyoop2@...]
Sent: Monday, March 22, 2004 8:20 AM
Subject: [APBR_analysis] Re: - wins/Tendex replacement
I think at this time of year there's also multiple levels of "freely
available" -- Level 1 is the guys who get waived in March (Dion
Glover, Rod Strickland, etc.) so they can sign with a playoff
contender. Level 2 is the guys the bad teams sign to replace them
(Britton Johnsen, Josh Davis, etc.).... and Level 5,831 is Charles
--- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
> -----Original Message-----times,
> From: Kevin Pelton [mailto:kpelton08@h...]
> Sent: Wednesday, March 17, 2004 11:45 AM
> >> > I'm not sure it will be easy to identify a replacement level.
> >> Wouldn't be easy, but I'd try various techniques: WAG (wild a**
> >> guessing: what number seems reasonable?); looking at actual
> >> transactions of players cut and added, especially those on 10-day
> >> contracts; actual statistics of actual 12th men, etc. These
> >> different techqniques would undoubtedly lead to different
> >> definitions of what the replacement level is, but now at
> >> least the range has been narrowed down.
> >MikeT, I know you've mentioned this kind of study a number of
> >and since I probably make more use of replacement level thananyone
> >else here (Dan being the possible exception), I've wanted to do itwas
> >for a long time and finally got around to it:
> That's really good stuff, exactly the sort of empirical study that
> needed. And I think we've got a good estimate of replacementlevel,
> at least as measured by your efficiency statistic.is a
> The point about in-season vs truly "freely available" free agents
> good one, and your stats show the importance of the distinction.You
> seem to be saying, and I agree, that the higher level, out-of-season,
> freely available free agent replacement level is the better one touse.
> At least for making long run, multiple-season, comparisons ofplayers.
> Teams in the short run may end up with less-than-replacement-levelrun.
> players due to contract restrictions, salary caps, or what not, but
> those conditions will not or at least need not persist in the long
> I think your .425 or .430 estimate of the replacement level is
> a good one, maybe it won't turn out to be 100% accurate but I'llbet it's
> reasonably close to whatever the true figure is. Because you'veused
> multiple techniques to arrive at the same estimate: the 10th playerIntuitively,
> technique, and the free agents' stats technique.
> There is a subtle problem however with the 10th player technique.
> one would think that the 12th player would be a better measure ofthe replacement
> level player. But the stats of these 12th players might actuallybe poor
> measures; they might be BELOW replacement value, and are only onthe roster
> due to guaranteed contracts or what not. Or they might be an 18-year old kid
> being stashed on the roster but not expected to contribute yet. Orthey might
> just have turned out to have a bad year, with horrendous stats thatplummet
> them to 12th.a good
> So those are good reasons why the 12th men's stats are perhaps not
> measure, they're likely to give us a figure that's belowreplacement level.
> So taking the 10th man's stats, or more precisely the stats of the290th
> guy in the league, might lead to better estimates.deck
> The subtle problem with all this is that your technique stacked the
> against the 12th men from the beginning, because you order theplayers
> by efficiency. So the 12th men are guaranteed to have the verylowest
> efficiencies on your rosters, and thus have stats that are likely tohave to
> be below replacement level.
> And thus to come up with a figure that is a better estimate, you
> "move up" the roster to the 10th position.impossible to
> A theoretically better technique (but one that is perhaps
> actually execute) would be to look at the stats of players who weidentify
> as the 12th man *a priori*, i.e. before looking at their stats.Some of
> these 12th men will be the Brian Cardinal types, with some decentstats.
> Some of them will be the Ansu Sesay types, basic deadwood there tofill
> the roster, literal replacement level players. Averaged together,their
> efficiency will be higher than the average efficiency of the 12thman
> calculated by looking at their actual stats and ordering them ("aposteriori"
> stats rather than "a priori"). Indeed, if done correctly, I'dexpect these
> 12th men to have stats slightly above the replacement level, simplybecause
> they were the 12th guys and not the 13th guys, the ones who trulywill be
> the replacements.before
> The trouble with my a priori idea of course is how do we know,
> looking at the stats, who the 12th man is? For some teams, it'sfairly
> easy to identify, but for others, not. Probably we'd have to picka couple
> of players from each team, and estimate the stats of the 11th-12thmen
> collectively.high. It's
> There's another reason why the .425/.430 might be just a little
> possible that the off-season pickups might not be as freelyavailable as
> we might think -- though undrafted, there might be a few teamswilling to
> snatch up these players, for their low cost, low risk, andpotential
> contributions. I.e. there may be some semi-decent free agents outthere,
> whose stats bring up the average, but who weren't really freelyavailable
> because other teams went after them and signed them. They were abit better
> than the truly freely available replacement players.suggests that
> However, you've got a great quote from Brian Cardinal which
> this is not the case.getting us into
> Bottom line: I think this is a good estimate which likely is
> the neighborhood of the replacement level. My guess is that if itturns out
> not to be the correct value, it's because it's a shade high.Yahoo! Groups Links