2014 NBA Playoffs Conference Finals Preview

After an excellent first round, my model had a much less successful second round, as I was 5-5-0 against the spread using my model with subjective adjustments, making me 25-19 for the entire playoffs.  Here are the full results:

LAC 6 times: 2-4-0
WAS 3 times: 3-1-0

Granted, this was an extremely small sample size, and adding my second-round results to my first-round results doesn't significantly increase the likelihood a coin-flip strategy would match my record.  The bigger issue is the 2-4 record betting on the Clippers given the confidence I had in that bet.  Specifically, a 57%-weighted coin would be just as likely to finish with a record as poor as 2-4 as a fair coin would be to finish with a record as good as 25-19.

Subjectively, I’m not sure I would actually adjust my Clippers-Thunder prediction much in hindsight.  The Thunder only outscored the Clippers by five total points in the series (or an average of 0.8 points per game), and while I actually predicted the Clippers to outscore the Thunder by three points per game, variance can often explain such a difference.  For the Clippers, Barnes (23/59 = 39% from the field, 11/35 = 31% from three) and Crawford (30/85 =35% from the field, 10/34 = 29% from three) didn’t shoot particularly well (total: 53/144 = 37% from the field, 21/69 = 30% from three), especially considering their shots were largely wide open (technically, not so much for Crawford, but he had Caron Butler defending him on isolations so it had the same effect).  Sure, there were some unexpected matchup issues, in particular Collison being unable to defend Westbrook/Jackson and Davis being unable to defend anything, but Rivers at least eventually realized near the end of the series that Collison was a huge liability.  The more important point I would make is that based on the Game 1 line, Vegas essentially had the Thunder as a 2-point favorite on a neutral court, and the large discrepancy between that estimate and mine meant that there was a lot of room for error for me to still be correct about betting the Clippers against that Vegas spread.  And in terms of point differential, I was technically correct (albeit by a much smaller margin), as the Thunder only outscored the Clippers by less than one point per game; the reason this conflicts with my 2-4 record is the additional variable of the timing/distribution of that point differential.  In all honesty, I believe I was more wrong about the Pacers-Wizards series even though I was 3-1 (which admittedly includes two abstained games in which the Wizards lost against the spread), due to the underestimated impact Hibbert would have on Wall (as I will discuss in further detail when discussing the Pacers).

I will now proceed to provide a similar analysis of the Conference Finals.  As with before, there will be new minutes allocations, but this time, there are also updated RPM numbers from ESPN (the numbers weren't updated between Round 1 and Round 2 but were updated before the Conference Finals).  Similar to before, all of the computation was done in R and the script I ran will be included.

Series Previews:

East: (1) Indiana vs. (2) Miami

	Model	Vegas*
Game 1 Probability	IND: 49% / MIA: 51%	IND: +135 / MIA: -155
Game 1 Spread	IND: +0.9 / MIA: -0.9	IND: +3.5 (-115) / MIA: -3.5 (-105)
Series Probability	IND: 29% / MIA: 71%	IND: +310 / MIA: -420

Based solely on season-long numbers (so that each regular-season minute is weighted the same as each playoff minute), my raw model suggests betting the Pacers.  It would appear that the Pacers were much better against Washington, and to some extent, this was true, especially with regards to Hibbert.  However, they still only outscored the Wizards by a total of 13 points in the series (or an average of 2.2 points/game), at about the same level that my raw model predicted, even with their significant matchup advantages.  Some of those matchup advantages were obvious before the series, in particular Nene and Gortat's limited shooting range, which allows Hibbert to stay close to the rim.  The other matchup issue is that no Wizard has an effective mid-range game (defined as anything other than a three or a layup, so this includes both two-point jumpers and floaters/runners).  In particular, Wall (and by extension, spot-up three-point shooters like Ariza and Webster) is completely ineffective if he can't drive all the way to the hoop for layups, and this is the specific type of offense Hibbert is designed to stop, especially when he doesn't have to rotate all the way from the three-point line.  In this regard, Washington was probably on the other end of the spectrum from Atlanta in terms of Indiana matchups.

Miami, on the other hand, is somewhere in between Atlanta and Washington on that spectrum.  LeBron has had similar issues with Hibbert as Wall in the past, but he also has a much better jumper, which likely means the Hibbert effect would be slightly smaller against the Heat (Wade also has an outstanding mid-range/floater game).  The Heat will likely have one three-point shooting big (Bosh) on the floor at all times, instead of two like Atlanta or zero like Washington.  Some people may wonder why Miami can't just follow Atlanta's blueprint of spreading out Indiana's defense completely, but Miami just doesn't have the bulky three-point shooter that the Hawks do in Antic, because Bosh, for all his strengths, just isn't strong enough to prevent Hibbert from establishing deep post position.  Andersen likely isn't strong enough to defend Hibbert either, which likely means a lot of Udonis Haslem in this series, who's significantly worse offensively than Andersen, Bosh, or any small-ball PF the Heat might play.  Whether Haslem can consistently hit his mid-range jumpers might be especially important in this series, because if jumper is falling, Miami will likely be able to score enough in the half court to be able to defend Indiana conservatively by dropping their bigs (in other words, like any other team would) instead of having to resort to their usual aggressive traps.  While the Heat traps are generally successful (especially since they lead to fast breaks for probably the best fast-breaking team in the league), they also lead to volatility, because an offense that executes well can shred it for easy baskets.  This is exactly what happened during last year's Conference Finals, as Indiana scored significantly above its season averages in the series.

Personally, I believe that Miami will choose to play Haslem/Andersen most of the time and defend Indiana conservatively.  As a result, I would adjust the margin of victory for each game by approximately 3 points each game towards Miami for recency adjustments and by approximately 2 points each game towards Indiana for matchup issues.  This moves Miami's expected margin of victory in Game 1 to 2 and makes Miami a 77% favorite in the series.  These estimates are close enough to the Vegas lines that I would abstain from this series for now.

West: (1) San Antonio vs. (2) Oklahoma City

	Model	Vegas*
Game 1 Probability	SAS: 78% / OKC: 22%	SAS: -245 / OKC: +205
Game 1 Spread	SAS: -9.3 / OKC: +9.3	SAS: -5.5 (-115) / OKC: +5.5 (-105)
Series Probability	SAS: 89% / OKC: 11%	SAS: -200 / OKC: +170

Oklahoma City has definitely provided matchup problems for San Antonio ever since the Spurs won the first two games of their 2012 Conference Finals matchup, as they then won the next four games of that series and have gone 6-2 in the regular season the last two years.  However, these matchup issues are not as extreme as these records might indicate.  In this year's season series that the Thunder swept, both teams were significantly undermanned, with matchups as ridiculous as Belinelli on Durant, making the results from those games much less meaningful.  Furthermore, this is the best Spurs team in years (perhaps ever), and even though the Thunder are just as good as they were two years ago, they no longer have Harden, who provided huge issues for a Spurs team that had no place to hide Parker on defense since both he and Westbrook could just overpower Parker.  As a result, I wouldn't place much stock in these past results.

It's no secret that I'm not the biggest Ibaka fan given that his box-score stats drastically overrate his impact.  Defensively, he tends to overhelp, takes poor routes, and lacks the strength or discipline to defend the post.  Offensively, he offers nothing beyond 45-50% 18-footers (and occasional corner threes), given that he's a poor passer and screener and his jumper doesn't actually space the floor much for his teammates as defenses are fine letting him shoot those long twos.  As a result, even with his absence limiting the potential switching strategies the Thunder can employ to limit San Antonio's pick-and-roll attack, I'm not convinced the injury is as impactful as most people think.  As long as Brooks gives most of those minutes to Collison (rather than playing Perkins and Adams together too much), there shouldn't be a significant dropoff.

Overall, the Thunder do still provide some matchup issues.  Leonard has had little success defending Durant, their length allows them to close out effectively and chase the Spurs off threes, and they still have both the big wings and the mobile bigs (even without Ibaka) to switch more successfully than most teams.  In fact, Oklahoma City should probably switch all pick-and-rolls whenever they go small (either with Durant at PF or even potentially at C) to force the Spurs to take off-the-dribble two-point jumpers that ideally would be less efficient than the Thunder's small-ball offense, but I have little faith that Brooks will make such adjustments until late in the series (if at all).  As a result, I would adjust the margin of victory for each game by only approximately one point towards Oklahoma City due to matchup issues, and another point towards Oklahoma City for Spurs home games due to diminishing marginal returns.  This moves San Antonio's expected margin of victory in Game 1 to 7 and makes San Antonio an 83% favorite in the series.  Even though Ibaka's injury lowered San Antonio's value, I would still bet San Antonio.

*All Vegas numbers from Bovada


List of minutes estimates and xRAPM ratings:

Show/Hide

     Spurs:

player	minutes	orpm	drpm	rpm	onet	dnet	net
Boris Diaw	23	0.47	1.03	1.50	0.23	0.49	0.72
Cory Joseph	2	-0.85	-3.28	-4.13	-0.04	-0.14	-0.17
Danny Green	23	0.71	2.47	3.18	0.34	1.18	1.52
Kawhi Leonard	35	0.88	1.98	2.86	0.64	1.44	2.09
Manu Ginobili	29	4.86	0.25	5.11	2.94	0.15	3.09
Marco Belinelli	12	0.98	-3.29	-2.31	0.25	-0.82	-0.58
Matt Bonner	4	1.96	0.26	2.22	0.16	0.02	0.19
Patty Mills	15	3.08	0.47	3.55	0.96	0.15	1.11
Tiago Splitter	27	-1.06	4.74	3.68	-0.60	2.67	2.07
Tim Duncan	35	-0.30	5.25	4.95	-0.22	3.83	3.61
Tony Parker	35	3.23	0.11	3.34	2.36	0.08	2.44

     Thunder:

player	minutes	orpm	drpm	rpm	onet	dnet	net
Caron Butler	19	-0.85	-3.65	-4.50	-0.34	-1.44	-1.78
Derek Fisher	13	-0.98	2.05	1.07	-0.27	0.56	0.29
Hasheem Thabeet	2	-4.70	3.69	-1.01	-0.20	0.15	-0.04
Jeremy Lamb	0	-0.40	-0.92	-1.32	0.00	0.00	0.00
Kendrick Perkins	22	-7.39	3.06	-4.33	-3.39	1.40	-1.98
Kevin Durant	45	6.38	0.05	6.43	5.98	0.05	6.03
Nick Collison	30	2.35	3.45	5.80	1.47	2.16	3.63
Reggie Jackson	26	0.71	0.85	1.56	0.38	0.46	0.85
Russell Westbrook	42	4.66	0.25	4.91	4.08	0.22	4.30
Serge Ibaka	0	1.02	3.19	4.21	0.00	0.00	0.00
Steven Adams	20	-3.16	-0.83	-3.99	-1.32	-0.35	-1.66
Thabo Sefolosha	21	-1.01	1.88	0.87	-0.44	0.82	0.38

     Pacers:

player	minutes	orpm	drpm	rpm	onet	dnet	net
C.J. Watson	15	0.17	-0.40	-0.23	0.05	-0.13	-0.07
Chris Copeland	6	0.55	-1.92	-1.37	0.07	-0.24	-0.17
David West	36	1.20	2.72	3.92	0.90	2.04	2.94
Evan Turner	10	-1.34	-1.44	-2.78	-0.28	-0.30	-0.58
George Hill	37	1.95	0.40	2.35	1.50	0.31	1.81
Ian Mahinmi	12	-5.01	3.67	-1.34	-1.25	0.92	-0.34
Lance Stephenson	37	0.66	0.27	0.93	0.51	0.21	0.72
Luis Scola	10	-3.36	0.59	-2.77	-0.70	0.12	-0.58
Paul George	41	0.30	2.70	3.00	0.26	2.31	2.56
Roy Hibbert	36	-1.53	3.31	1.78	-1.15	2.48	1.34

     Heat:

player	minutes	orpm	drpm	rpm	onet	dnet	net
Chris Andersen	15	0.62	4.05	4.67	0.19	1.27	1.46
Chris Bosh	37	0.18	3.58	3.76	0.14	2.76	2.90
Dwyane Wade	36	1.06	0.91	1.97	0.80	0.68	1.48
Greg Oden	2	-3.80	-0.31	-4.11	-0.16	-0.01	-0.17
James Jones	4	-1.09	1.03	-0.06	-0.09	0.09	-0.01
LeBron James	42	8.90	-0.28	8.62	7.79	-0.25	7.54
Mario Chalmers	27	0.53	1.06	1.59	0.30	0.60	0.89
Michael Beasley	0	-2.88	-2.88	-5.76	0.00	0.00	0.00
Norris Cole	16	-1.63	-1.69	-3.32	-0.54	-0.56	-1.11
Rashard Lewis	6	-2.41	-0.51	-2.92	-0.30	-0.06	-0.37
Ray Allen	24	2.22	-2.75	-0.53	1.11	-1.38	-0.27
Shane Battier	15	-0.42	1.39	0.97	-0.13	0.43	0.30
Toney Douglas	0	0.17	-1.53	-1.36	0.00	0.00	0.00
Udonis Haslem	16	-4.43	1.53	-2.90	-1.48	0.51	-0.97

R Script:

Show/Hide

library(XML)
library(RCurl)
v.pages <- 1:11            #change if increase in number of pages on http://espn.go.com/nba/statistics/rpm/_/page/1/sort/RPM

v.column.names <- c("player","team","orpm","drpm","rpm")
v.teams.home <- c("sas","ind")
v.teams.away <- c("okc","mia")
v.teams <- c("sas","okc","ind","mia")

hca <- 3.2                #difference between neutral court and home court; this is total hca adjustment, not the adjustment for each team
avg_eff <- 104            #average league points per 100 possessions

df.rpm <- data.frame(player=character(0),team=character(0),orpm=numeric(0),drpm=numeric(0),rpm=numeric(0))

for(page in v.pages)
{
    v.url.rpm <- paste("http://espn.go.com/nba/statistics/rpm/_/page/",page,"/sort/RPM",sep="")
    table <- readHTMLTable(v.url.rpm)[[1]]
    table <- table[as.character((table[,2]))!="NAME",]
    table2 <- cbind(as.data.frame(matrix(unlist(strsplit(as.character(table[,2]),", ")),ncol=2,byrow=TRUE),stringsAsFactors=FALSE)[,1], as.data.frame(matrix(unlist(table[,3]),ncol=1,byrow=TRUE),stringsAsFactors=FALSE), as.data.frame(matrix(unlist(table[,6]),ncol=1,byrow=TRUE),stringsAsFactors=FALSE),as.data.frame(matrix(unlist(table[,7]),ncol=1,byrow=TRUE),stringsAsFactors=FALSE),as.data.frame(matrix(unlist(table[,8]),ncol=1,byrow=TRUE),stringsAsFactors=FALSE))
    table2[,3] <- as.numeric(table2[,3])
    table2[,4] <- as.numeric(table2[,4])
    table2[,5] <- as.numeric(table2[,5])
    colnames(table2) <- v.column.names
    df.rpm <- rbind(df.rpm,table2)
}

#minutes distributions and individual xrapm:
df.sas.r3 <- data.frame(player=c("Tony Parker","Tim Duncan","Kawhi Leonard","Marco Belinelli","Boris Diaw","Danny Green","Manu Ginobili","Tiago Splitter","Patty Mills","Cory Joseph","Matt Bonner"),minutes=c(35,35,35,12,23,23,29,27,15,2,4))
df.sas.r3 <- merge(df.sas.r3, df.rpm[,c(1,3,4,5)], by="player", all.x=TRUE)
df.okc.r3 <- data.frame(player=c("Kevin Durant","Serge Ibaka","Russell Westbrook","Reggie Jackson","Caron Butler","Thabo Sefolosha","Jeremy Lamb","Kendrick Perkins","Derek Fisher","Nick Collison","Steven Adams","Hasheem Thabeet"),minutes=c(45,0,42,26,19,21,0,22,13,30,20,2))
df.okc.r3 <- merge(df.okc.r3, df.rpm[,c(1,3,4,5)], by="player", all.x=TRUE)
df.ind.r3 <- data.frame(player=c("Paul George","Lance Stephenson","George Hill","David West","Roy Hibbert","Evan Turner","C.J. Watson","Luis Scola","Ian Mahinmi","Chris Copeland"),minutes=c(41,37,37,36,36,10,15,10,12,6))
df.ind.r3 <- merge(df.ind.r3, df.rpm[,c(1,3,4,5)], by="player", all.x=TRUE)
df.mia.r3 <- data.frame(player=c("LeBron James","Dwyane Wade","Chris Bosh","Mario Chalmers","Ray Allen","Norris Cole","Shane Battier","Chris Andersen","Rashard Lewis","Toney Douglas","Michael Beasley","Udonis Haslem","James Jones","Greg Oden"),minutes=c(42,36,37,27,24,16,15,15,6,0,0,16,4,2))
df.mia.r3 <- merge(df.mia.r3, df.rpm[,c(1,3,4,5)], by="player", all.x=TRUE)

df.sas.r3$onet <- df.sas.r3$minutes * df.sas.r3$orpm / 48
df.okc.r3$onet <- df.okc.r3$minutes * df.okc.r3$orpm / 48
df.ind.r3$onet <- df.ind.r3$minutes * df.ind.r3$orpm / 48
df.mia.r3$onet <- df.mia.r3$minutes * df.mia.r3$orpm / 48

df.sas.r3$dnet <- df.sas.r3$minutes * df.sas.r3$drpm / 48
df.okc.r3$dnet <- df.okc.r3$minutes * df.okc.r3$drpm / 48
df.ind.r3$dnet <- df.ind.r3$minutes * df.ind.r3$drpm / 48
df.mia.r3$dnet <- df.mia.r3$minutes * df.mia.r3$drpm / 48

df.sas.r3$net <- df.sas.r3$minutes * df.sas.r3$rpm / 48
df.okc.r3$net <- df.okc.r3$minutes * df.okc.r3$rpm / 48
df.ind.r3$net <- df.ind.r3$minutes * df.ind.r3$rpm / 48
df.mia.r3$net <- df.mia.r3$minutes * df.mia.r3$rpm / 48

#team-level xrapm
df.all.r3 <- data.frame(team=v.teams)
df.all.r3$onet <- as.vector(t(data.frame(sas=sum(df.sas.r3$onet),okc=sum(df.okc.r3$onet),ind=sum(df.ind.r3$onet),mia=sum(df.mia.r3$onet))))
df.all.r3$dnet <- as.vector(t(data.frame(sas=sum(df.sas.r3$dnet),okc=sum(df.okc.r3$dnet),ind=sum(df.ind.r3$dnet),mia=sum(df.mia.r3$dnet))))
df.all.r3$net <- as.vector(t(data.frame(sas=sum(df.sas.r3$net),okc=sum(df.okc.r3$net),ind=sum(df.ind.r3$net),mia=sum(df.mia.r3$net))))
df.all.r3$home_perc <- (avg_eff+df.all.r3$onet+hca/4)^14/((avg_eff+df.all.r3$onet+hca/4)^14+(avg_eff-df.all.r3$dnet-hca/4)^14)
df.all.r3$away_perc <- (avg_eff+df.all.r3$onet-hca/4)^14/((avg_eff+df.all.r3$onet-hca/4)^14+(avg_eff-df.all.r3$dnet+hca/4)^14)

#specific round 3 matchups
df.r3.matchups.1 <- data.frame(team=v.teams.home, opponent=v.teams.away, stringsAsFactors = FALSE)
df.r3.matchups.1$team_home_perc <- merge(data.frame(team=df.r3.matchups.1$team),df.all.r3[,c(1,5)], by="team", all.x=TRUE, sort=FALSE)[,2]
df.r3.matchups.1$team_away_perc <- merge(data.frame(team=df.r3.matchups.1$team),df.all.r3[,c(1,6)], by="team", all.x=TRUE, sort=FALSE)[,2]
df.r3.matchups.1$opponent_home_perc <- merge(data.frame(team=df.r3.matchups.1$opponent),df.all.r3[,c(1,5)], by="team", all.x=TRUE, sort=FALSE)[,2]
df.r3.matchups.1$opponent_away_perc <- merge(data.frame(team=df.r3.matchups.1$opponent),df.all.r3[,c(1,6)], by="team", all.x=TRUE, sort=FALSE)[,2]
df.r3.matchups.1$team_matchup_home_perc <- (df.r3.matchups.1$team_home_perc - (df.r3.matchups.1$team_home_perc*df.r3.matchups.1$opponent_away_perc))/(df.r3.matchups.1$team_home_perc + df.r3.matchups.1$opponent_away_perc - (2*df.r3.matchups.1$team_home_perc*df.r3.matchups.1$opponent_away_perc))
df.r3.matchups.1$team_matchup_away_perc <- (df.r3.matchups.1$team_away_perc - (df.r3.matchups.1$team_away_perc*df.r3.matchups.1$opponent_home_perc))/(df.r3.matchups.1$team_away_perc + df.r3.matchups.1$opponent_home_perc - (2*df.r3.matchups.1$team_away_perc*df.r3.matchups.1$opponent_home_perc))
df.r3.matchups.1$series_perc <- df.r3.matchups.1$team_matchup_home_perc^4 + 4*df.r3.matchups.1$team_matchup_home_perc^3*(1-df.r3.matchups.1$team_matchup_home_perc)*(1-(1-df.r3.matchups.1$team_matchup_away_perc)^3) + 6*df.r3.matchups.1$team_matchup_home_perc^2*(1-df.r3.matchups.1$team_matchup_home_perc)^2*((3*df.r3.matchups.1$team_matchup_away_perc^2*(1-df.r3.matchups.1$team_matchup_away_perc))+df.r3.matchups.1$team_matchup_away_perc^3)+4*
df.r3.matchups.1$team_matchup_home_perc*(1-df.r3.matchups.1$team_matchup_home_perc)^3*df.r3.matchups.1$team_matchup_away_perc^3
df.r3.matchups.2 <- df.r3.matchups.1[,c(2,1)]
colnames(df.r3.matchups.2) <- c("team","opponent")
df.r3.matchups.2$team_home_perc <- df.r3.matchups.1$opponent_home_perc
df.r3.matchups.2$team_away_perc <- df.r3.matchups.1$opponent_away_perc
df.r3.matchups.2$opponent_home_perc <- df.r3.matchups.1$team_home_perc
df.r3.matchups.2$opponent_away_perc <- df.r3.matchups.1$team_away_perc
df.r3.matchups.2$team_matchup_home_perc <- 1-df.r3.matchups.1$team_matchup_away_perc
df.r3.matchups.2$team_matchup_away_perc <- 1-df.r3.matchups.1$team_matchup_home_perc
df.r3.matchups.2$series_perc <- 1-df.r3.matchups.1$series_perc
df.r3.matchups <- rbind(df.r3.matchups.1,df.r3.matchups.2)
df.r3.matchups <- df.r3.matchups[order(df.r3.matchups$series_perc, decreasing=TRUE),]



#print results
df.r3.matchups
df.all.r3[order(df.all.r3$net,decreasing=TRUE),c(1:4)]
df.sas.r3
df.okc.r3
df.ind.r3
df.mia.r3