Quantcast
Wednesday, May 29th, 2013 at 5:54 pm  |  18 responses

Don’t Call It Luck

One improbable stat after another in Heat-Pacers Game 4.

After blessing SLAMonline with the nERD Power Rankings throughout the season, our guys at numberFire are back to help us look at the postseason in an analytics-based way. So what do the algorithmic models say about how statistically unlikely some of the numbers from Game 4 were? NumberFire’s Chief Editor is here with the answer.—Ed.

by Zach Warren / @ZachWWarren

The Pacers took Game 4. They did so by topping the Heat’s effective field goal percentage (eFG%) by .079, collecting a staggering 45.5 percent of offensive rebounds, and committing 2.7 percent fewer turnovers than their Playoff average.

Does that all sound a little bit off to you? Us too. The Pacers may have only won Game 4 by seven points, but it took an incredible amount of statistically improbable outcomes to get there.

We decided to run some of last night’s numbers through the stats machine, just to see how likely they would have been given each team’s regular-season averages. These odds are assuming the Heat or Pacers played a League-average squad, so, for instance, Miami’s projected field-goal percentage might actually be a bit lower because of the opponent. But these odds still give a good indication of just how unlikely a lot of the results really were given their previous play.

Heat fans shouldn’t worry quite yet. Miami’s poor shooting, Indiana’s strong shooting, and Indiana’s absurd offensive rebounding look a lot like outliers when compared with past play.

How Did That Happen?

Miami Shoots .442 eFG% or Less: 6.90 percent
The Pacers may hold the League’s number one defensive eFG% at .453, but for Miami, these types of games just never happen. In 82 regular-season games, the Heat shot below 40 percent from the field exactly twice: November 11 against the Grizzlies and January 27 against the Celtics. In both of those games and in this one, Dwyane Wade was the ice cold one, going 3-15 against Memphis, 6-20 against Boston, and 5-15 last night. Given 15 shots, the odds Wade would make five or less are only at 11 percent.

Indiana Makes 35 or More Shots on 70 Attempts: 16.86 percent
While Indiana shooting 50 percent may be more likely than Miami shooting less than 40 percent, it’s still not the most likely scenario around. If the defensively minded Pacers played games against a league average team into all eternity (Mike D’Antoni’s version of slow-paced hell), Indiana would shoot this well around once every six times. Now figure that the Heat are better than the League-average team—their defensive eFG% ranked ninth this year—and you realize that seeing a repeat Game 5 Indiana performance isn’t likely.

Indiana Collects 45.5 Percent or More of Offensive Rebounds: 2.30 percent
Yeah, we know the Heat can’t exactly rebound: Miami collected 73.0 percent of defensive boards during the regular season, just 24th in the League. But allowing Indiana to collect nearly half of the available rebounds on their offensive end? That’s a new low. Strangely enough, I may put this one at the feet of LeBron. He collected 20.8 percent of available defensive boards during the regular season, but only 6.7 percent last night.

The New Series Odds

Yesterday, when I took a look at San Antonio’s chances against the Heat and Pacers, I mentioned that Indiana only held an 17 percent chance of winning the series. Naturally, those odds have gone up, but they’re not out of the woods yet.

  4 Games 5 Games 6 Games 7 Games Total Win Odds
Miami 0.00% 0.00% 35.39% 35.64% 71.03%
Indiana 0.00% 0.00% 14.61% 14.36% 28.97%

This is what happens when you win so improbably: The odds aren’t good that you’ll do it again. As it stands, the most likely series outcome is Heat in 7, but that’s only by percentage points over Heat in 6. There is currently a greater chance of the Heat taking both of the next two games than there is of the Pacers winning at all.

That Game 1 Layup hurts even more now, doesn’t it? Poor Pacers fans. Their best team since the Reggie Miller days, and they run into the South Beach buzzsaw. 29 percent isn’t terrible odds, but it’s a substantial amount to attempt to overcome.

NumberFire is a sports analytics platform that uses algorithmic modeling to better understand sports. Follow Nik Bonaddio at @numberfire, Keith Goldner at @drivebyfootball, and Zach Warren at @ZachWWarren. Check out numberFire on Facebook.

  • Add a Comment
  • Share
  • RSS

Tags: , , , ,

  • http://twitter.com/sooperfadeaway nbk

    This is awesome

  • i_ball

    Yeah but they still need to explain what “league average” means

  • http://twitter.com/sooperfadeaway nbk

    League average would be a team of statistical averages. It’s just all the teams stats combined then divided by 30

  • Kid-Canada

    Interesting stuff. Don’t forget about Game 2 though. Indiana managed to win collecting 33.3% of offensive rebounds and with Miami shooting an eFG% of 0.514.

  • berkamore

    Guys,

    Seems to me that what the Heat did against the League is kind of irrelevant. You should be looking at what they did against the Pacers in the last 2-3 years to get a better idea of what to expect.

    If you don’t have enough data points, just make some assumptions based on the games played and iterate 1,000 times. (You can use a Monte Carlo simulation) That should give you a better idea of what’s what. (Law of Large Numbers and all).

    When you have that, then you can go, yeah, the numbers should look like………… and this game is a wide deviation from the average (sorry if i am a little technical here) and yada yada……………………..Your method works if you want to know what to expect at the beginning of the season. (And you should probably be careful with the league average thing because each team doesn’t play every opponent the same amount of time…you get the idea)

  • berkamore

    …same amount of TIMES (you play more games against some teams than others)

  • http://twitter.com/sooperfadeaway nbk

    It’s makes more sense to you for them to take 12 games, from 3 different teams that share common players, multiply it by 1,000 and get their data from that? And this sounds more intelligent then using 82 games played by these teams this season? Fa real?

  • berkamore

    I recommend that you look up what a Monte Carlo simulation is. It is certainly NOT multiplying by 1000. LOL.

  • http://twitter.com/sooperfadeaway nbk

    Do you know anything about Monte Carlo simulation aside from the basic concept? For instance, considering you have statistics, or measurable variables, how would you identify random variables for probability distribution?

    .

    You use a Monte Carlo Simulation on a situation that you can’t quantify. But since we have actual statistics, there is no need for a Monte Carlo style simulation, because we can just run a direct simulation. Which is what this is. And it is much more accurate. Because you aren’t making up numbers and variables and hoping they give you the results your looking for.

  • berkamore

    You run a Monte Carlo simulation when you don’t have enough data points to make a decision.

    Now what are we looking to do here? We are trying to understand/ predict what is likely to happen when the Heat play the Pacers during the playoffs That’s it. Not the rest of the league, not some team in Europe. Heat vs. Pacers and preferably in a Playoff situation.

    What is our problem here? We don’t have enough data points to predict what can/should/will happen. That’s why we use a simulation. And I am sure that you understand that

    I grant you that the Heat and Pacers have changed/grown in the past 2-3 years but those data points are the least imperfect that we have. Again, what LeBron did vs the Raptors or Blazers or any other team apart from the Pacers is kind of irrelevant.

    And just for the record, when you run a simulation, you have assumptions but don’t “pick” numbers. You run a program that picks them randomly. That’s actually the whole point.

    But again, you don’t have to believe me. Print the whole page and the comments, show it to a College Statistics Professor and see what she tells you.

  • http://twitter.com/sooperfadeaway nbk

    ugh. the simulation can’t just pick numbers randomly. you have to stage the simulation. Or else it will grab completely irrelevant numbers.

    .

    a simulation doesn’t know points are more important than rebounds, unless you tell it so. it can “randomly” pick a statistic that is completely irrelevant.

    .

    and it still doesn’t make sense to use data from previous years in that small of a sample size. not at all.

    .

    first, you can’t value the Pacers in previous seasons anything remotely close to how they are today. They have grown by leaps and bounds each of the last 2 years.

    .

    second, your monte carlo simulation in 12 games (that’s the largest sample you can use under your suggestion) can’t take into account any context. Because your sample size is so small, any assumed/projected/random variable hinges on a completely unreliable subset.

    .

    In 2010-11, the first year of the simulation YOU want to run, the Pacers went 37-45. Their starting lineup features ONE player from this years starting lineup. That is a completely different team. A terrible, non playoff team.

    .

    And you think it makes sense to use that team, from 3 years ago, to help create random variables, to simulate what would happen between these teams, in a playoff series.

    .

    You can’t honestly think this sounds more logical then using a larger set of data, from these teams, this season.

    .

    Monte Carlo simulations are not usually used to predict the most likely outcome of an event anyway. It’s used to show all possible outcomes, and the likelihood of each outcome assuming certain decisions. So unless you have a way to know what decisions will be made, all of them, over 48 minutes, by 12 different players, 2 different coaches, and 4 referees, it makes absolutely no sense to use a Monte Carlo Simulation to predict what will happen in a basketball series when you have a large data set to use in a direct simulation.

  • berkamore

    You obviously haven’t read my email.

    “The Simulation can’t pick numbers randomly…………”. In my email, i said that you must have assumptions, Read it closely.

    “A simulation doesn’t know that points are more important than rebound unless you tell it so………..” Where in my email did I disagree with that? Beats me.

    “First you can’t value the Pacers yada yada……….”. I granted you that point, I said it was imperfect.

    “Monte Carlo simulation in 12 games…..” Ok, I think that’s where the disconnect is. Reread my email. You use simulation when you don’t have enough data points. If you do, you just use historical data.

    The next few paragraphs are just rehashing old stuff.

    “Monte Carlo simulations are not usually used to predict the most likely outcome of an event anyway”. WHAT???????

    ROFL. Okay, man, you are right, whatever you say. ROFL.

  • LakeShow

  • http://twitter.com/sooperfadeaway nbk

    “Monte Carlo simulation furnishes the decision-maker with a range of possible outcomes and the probabilities they will occur for any choice of action.. It shows the extreme possibilities—the outcomes of going for broke and for the most conservative decision—along with all possible consequences for middle-of-the-road decisions.”
    - http://www.palisade.com/risk/monte_carlo_simulation.asp

    - And i didn’t just say, “Monte Carlo simulations are not usually used to predict the most likely outcome of an event anyway” – way to leave out the part that makes that statement relevant tho, that’s cute.

    .

    .”It’s used to show all possible outcomes, and the likelihood of each outcome assuming certain decisions. So unless you have a way to know what decisions will be made, all of them, over 48 minutes, by 12 different players, 2 different coaches, and 4 referees, it makes absolutely no sense to use a Monte Carlo Simulation to predict what will happen in a basketball series when you have a large data set to use in a direct simulation.”

    .

    QUOTE, “The Simulation can’t pick numbers randomly…………”. In my email, i said that you must have assumptions, Read it closely.”

    QUOTE, “And just for the record, when you run a simulation, you have assumptions but don’t “pick” numbers. You run a program that picks them randomly. That’s actually the whole point.”

    .
    Yeah you said you have assumptions, and the computer simulation picks numbers randomly.
    .
    So you want to make assumptions, based on 6 different teams that share common players, and let the computer pick which ones matter….randomly.
    .
    I don’t understand why you can’t see how stupid that is when you can just use a much larger subset of numbers, for these teams, with these players, this season.
    .
    I know you keep talking about how it’s irrelevant how Miami did against Toronto, etc. But that’s not any more irrelevant than what the 2010-11 Heat did against the 2010-11 Pacers. Actually, it’s less irrelevant. Because at least half of your variables in a direct simulation represent the teams that are actually playing. Instead of using 2/3′s of your variables from teams that have nothing to do with this simulation.

  • http://twitter.com/sooperfadeaway nbk

    this guy has an idea, and is just running with it. he’s wants to run a simulation that must assume data without any constant variables to anchor the simulation.

    .

    in essence, there is no variable that says “if player A makes this decision, action B will take place” because THESE ARE HUMAN BEINGS who are random variables individually, yet he thinks using a smaller data set, for multiple human beings, and making assumptions makes more sense, because basically, the team name hasn’t changed in 3 years.

  • berkamore

    You just put the whole thing in the proper context. Thanks for that. LOL.

  • bustha

    THIS is by far the stupidest statistics i have ever seen. Can we get another article after game 5 with another reason why the heat lost?? It’s hilarious your gonna pick the last game to come to a conclusion. Other analysts have made the same prediction after the Heat were up and then again when the Pacers tied it. Each analyst jumps on the bandwagon with some stupid statistics. Stop trying. Please.

    P.s. the pacer’s didn’t win improbably, only the author of the article is thinking that (he probably don’t even play basketball. ) the Heat didn’t lose improbably either. The heat played like garbage, and the Pacers won. The End.

  • Jeff Willis

    You could make the assumption that the Pacer’s are learning from past mistakes and correcting their defense. When you play the same team over and over again you start to learn weaknesses. To think that the Pacer’s will never hit 35 or over shots on 70 attempts. Where did you even get this statistic from?? It’s so random. Considering if they hit 30 of 60 or 34 of 69 it wouldn’t count toward your statistic at all. Your dumb.

Advertisement