Yup, I'm with you agentkirb87. I'm trying not to bust on DD, since clutchfans is not truly a statgeek site, and I don't expect everybody here to be fluent with statistical modeling. For those who do actually want to learn more, the essential debate revolves around the concept of maximum likelihood, i.e. DD wants to know if THE prediction from the model is right. Well, there's no single prediction from the model. One workaround is to consider the maximally likely prediction from the model, and see if it's right. But judging the model from only that maximally likely prediction is arbitrary.
I'm not really trying to bust on DD, either. I feel like I've probably been really aggressive towards his stance. He'll say one sentence and I'll respond with 2-3 heated paragraphs. Perhaps that's not fair, but he's getting a few fundamental things about statistical/probability analysis wrong. I'm a Computer Science major/Math Minor in college. I do things like this on my free time (although not exactly this, and not near as advanced). So its almost insulting to me when someone comes in and talks down to it, questions how good it is.
Agent, I understand probabilities......and just because one is the highest likelyhood does not make it a prediction. I am just CURIOUS as to how it all played out based upon the first run of the model, and if the OP thinks there might be some ways to tweak it for more data. For instance, in cards, the math is pretty straight forward, but in this instance there are a lot more variables to consider, how to weight home vs away, do you weight more against good teams at home versus bad ones? What about injuries, how much do they weigh into the equation. What is annoying though is that people are complaining about whether or not people understand probabilities......just because I want to look at it from a different angle, does not mean I have no understanding. In your AA holdem scenario....AA is about 83%....but that means there is still about a 17% chance of it being beaten, it is pure math. In this equation there are WAY more variables, and I am intersted in seeing if the model he is running will be tweaked. I am NOT saying this simulator sucks....I am more interested in the variables and whether they can be tweaked, which to me gives a greater accuracy (and yes, hard to quantify because 1% still means there is a possibility of that happening). DD
You say you are interested and curious... what are you waiting for? Given the information you have right now you can make an educated guess as to the accuracy of the formula. You don't need to wait for the playoffs to play out to test out the results for the regular season (to be honest, I have no faith in the accuracy of the "predictions" for who wins the title because the playoffs are a completely different animal... I don't think you can make a formula for the playoffs).
agreed. however, even though there is no such thing as judgement of whether the tool that BBR developed was "right", there is definitely an argument as to how good it is. the tool is really just a forecast of future events based on history. how that history is input makes an enormous difference in its ability to correctly forecast the likelihood of various scenarios. since valid statistical sampling is always a challenge in basketball analysis (as opposed to baseball for example), the assumptions made about the input further magnify the range of potential models that can come out. things like scheduling difficulties (b2bs, and 3in4s) are usually not captured in these types of models. historical over/underperformance vs. certain teams is usually not captured. home vs. road is usually not captured. add all of that up and you can have swings of up to and over +/-6pts vs. what a win% based model would say. baking in a home vs. road assumption alone is worth 3 to 4 points on average, with some teams (like the jazz) being even more extreme. you can see how including that information could really alter the chance to win of a team going to play in Utah, which works its way into the possibilities that develop from there. if 100 of us built models like this, we'd see the predictions clump around an average, and the results of the models would likely be normally distributed on an individual datapoint basis. all because of these types of assumptions and how you bake them in. that being said, i really do appreciate the fact that BBR did all of this, really to no benefit for him/her, and purely as a way for us to see the playoff possibilities in the NBA evolve as the schedule counts to zero. it's been fun to check this thing every morning and get the latest context around upcoming games. so leaving the argument about how detailed it needs to be or how right it is aside, it looks like a lot of people appreciated the effort and got some good perspective out of the information. that's all that matters really.
Wow. So the simulator thinks we have a 53% chance of winning the Dallas game? Cool. If we lose to Dallas, we only have 11% chance of staying in the #2/#3 bracket. If we win, we are guaranteed to stay there. This is why we MUST win the last game.
I totally agree, I am just more curious about the process etc, and what data gets pumped in.....and would more data create different results, and by how much? DD
its not jsut about more data - its also what data, and in how much detail, as i highlighted above. and yes, different/more/more detailed data will create different results, and the degree to which the results would change would depend entirely on the assumptions you used to build the models aroudn the new dataset.
Right, so what data would make this an even BETTER model? That is what I am interested in...... How well did this model do? Can it be tweaked, or should it? DD
it all depends on the undelying mathematical model. And that's how betting agencies make their money... so please, stop criticizing. If you don't understand statistics, move on.
i'm not criticizing. the question is being asked as to what makes these models better and that's the information i was typing out.
i dont think thats up to any of us to decide. its not our work. if you are interested in exploring "better" models, by all means give it a shot. you can tweak and compare results to your heart's content.
I tend to agree with this. The model should be evaluated based on the inputs used to create it because that's where human judgement comes into play. Just curious but why is this kept a secret? BigRed do you make any money from these probabilities?
If we lose, Denver will be guaranteed #2 and will let Portland win. Portland will get 54, Rockets will only be #4/#5, can't be #3. If we win and San Antonio loses, Portland will pass SAS to grab #4, and will not fight Denver. Denver win to get #2, we will be #3. So, I shifted BBR's numbers a bit to get the following: Code: Last Game Final Rank Hou Por SA Prob Den Hou Por SA === === === ==== === === === === 0 0 0 0% 2 3 4 5 0 0 1 0% 2 4 5 3 0 1 0 17% 2 4 3 5 0 1 1 36% 2 5 4 3 1 0 0 15% 2 3 4 5 1 0 1 11% 2 3 5 4 1 1 0 0% 3 2 4 5 1 1 1 21% 3 2 4 5 Then the percentages should be something like this: Denver (2/3): (69% 31%) -> (79% 21%) Houston(2/3/4/5): (31% 22% 24% 24%) -> (21% 26% 17% 36%) San Antonio (3/4/5): (36% 11% 53%) no change Portland (3/4/5): (11% 65% 23%) -> (17% 72% 11% )
No-one can answer this question until you define what you mean by a model "doing well." Examples: a model does well if and only if gets all the playoff matchups correct, or a model's quality is proportional to the number of matchups it gets correct, or a model's quality is proportional to how closely it predicts team's final records, etc. Unfortunately, the measure that most people are interested in, i.e. who we are going to face in the playoffs, has extremely high variance in the model. So if you try to evaluate the model based on that outcome, then you won't get a statistically useful evaluation of the model. You will do better if you evaluate on something that averages out like "the number of games the rockets will win by the end of the season." But even this has an uncertainty of several games. As a practical matter, I'm simply accepting the fact that the model takes into account most things you'd want it to take into account. That's not to trivialize the model, though, as it handles many things that are difficult to keep track of just by intuition (e.g. which teams play each game, the tie breakers, etc.), which makes it a really nice program. Props to the OP.
I don't really understand where you are going with this. You don't know his exact formula and what factors he considers. His quote from page 1: I'm sure if you poke around, you can find your own formula, program it into the simulator, maybe add some tweaks of your own, and see how it matches with his. Thats about the best you can do. Another thing you can do, like I said, is use the results he projected from the first page and decide for yourself if it makes sense. Here's the first batch of WC wins. Code: Team Wins ==== ==== Lakers 64.9 Spurs 53.8 Nuggets 53.3 Rockets 52.4 Blazers 51.7 Jazz 50.9 Hornets 50.5 Mavericks 48.3 Suns 45.1 The Jazz are about 2 games lower than the prediction, but I believe they have had 2-3 upset victories down the stretch and have just been struggling in general. The Hornets are about where they are, the Mavs are about where they are. The Spurs are a few games below where they should be, but when you lose Ginobli late in the season, thats what happens. The Rockets are a game or so above where they are here. The Blazers are a couple games above this projection. Nuggets are about where they are. Everything is slightly off, but only the Jazz are really far off on the prediction. Now you have to ask yourself. Does this prediction make sense to you considering the information known at the time? I would say it does. No one is really far off of their win total. You have the Spurs/Nuggets/Blazers/Rockets all sort of grouped together. And the Jazz/Hornets/Mavs grouped together. The difference being that there isn't as big of a hole between 5 and 6.
Latest NBA results used: Apr 14 Western Conference Average Wins Code: Team Wins ==== ==== Lakers 65.0 Nuggets 54.3 Spurs 53.7 Blazers 53.7 Rockets 53.5 Mavericks 49.5 Hornets 49.3 Jazz 48.0 Western Conference Wins (Percent) Code: Team 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 ==== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== Lakers 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100.0 Nuggets 0.0 0.0 0.0 0.0 0.0 0.0 65.9 34.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Spurs 0.0 0.0 0.0 0.0 0.0 31.1 68.9 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Blazers 0.0 0.0 0.0 0.0 0.0 34.1 65.9 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Rockets 0.0 0.0 0.0 0.0 0.0 53.2 46.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Mavericks 0.0 46.8 53.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Hornets 0.0 68.9 31.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Jazz 100.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Western Conference Placement (Percent) Code: Team 1st 2nd 3rd 4th 5th 6th 7th 8th out ==== ===== ===== ===== ===== ===== ===== ===== ===== ===== Lakers 100.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Nuggets 0.0 69.2 30.8 0.0 0.0 0.0 0.0 0.0 0.0 Spurs 0.0 0.0 36.6 11.0 52.4 0.0 0.0 0.0 0.0 Blazers 0.0 0.0 10.9 65.7 23.4 0.0 0.0 0.0 0.0 Rockets 0.0 30.8 21.7 23.3 24.2 0.0 0.0 0.0 0.0 Mavericks 0.0 0.0 0.0 0.0 0.0 36.6 63.4 0.0 0.0 Hornets 0.0 0.0 0.0 0.0 0.0 63.4 36.6 0.0 0.0 Jazz 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100.0 0.0 Western Conference Division Title (Percent) Code: Team Northwest Title Percent ==== ====================== Nuggets 100.0 Blazers 0.0 Jazz 0.0 Team Pacific Title Percent ==== ====================== Lakers 100.0 Team Southwest Title Percent ==== ====================== Spurs 36.6 Rockets 63.4 Mavericks 0.0 Hornets 0.0 Western Conference Playoff Advancement (Percent) Code: Team To 1st To 2nd To Conf To Fnls Champs ==== ======= ======= ======= ======= ======= Lakers 100.0 86.2 65.8 53.7 23.6 Nuggets 100.0 69.7 37.8 11.4 1.9 Spurs 100.0 49.0 19.2 6.6 1.1 Blazers 100.0 62.6 23.2 12.8 3.0 Rockets 100.0 59.2 28.4 10.3 2.0 Mavericks 100.0 30.7 11.3 1.9 0.2 Hornets 100.0 28.7 10.0 1.7 0.1 Jazz 100.0 13.8 4.4 1.7 0.2 Eastern Conference Average Wins Code: Team Wins ==== ==== Cavaliers 66.9 Celtics 61.9 Magic 58.8 Hawks 47.6 Heat 42.6 Bulls 41.7 76ers 40.1 Pistons 39.4 Eastern Conference Wins (Percent) Code: Team 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 ==== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== Cavaliers 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 14.0 86.0 Celtics 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.3 93.7 0.0 0.0 0.0 0.0 0.0 Magic 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 16.9 83.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Hawks 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 35.2 64.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Heat 0.0 0.0 0.0 36.3 63.7 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Bulls 0.0 0.0 28.9 71.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 76ers 0.0 86.0 14.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Pistons 63.7 36.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Eastern Conference Placement (Percent) Code: Team 1st 2nd 3rd 4th 5th 6th 7th 8th out ==== ===== ===== ===== ===== ===== ===== ===== ===== ===== Cavaliers 100.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Celtics 0.0 100.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Magic 0.0 0.0 100.0 0.0 0.0 0.0 0.0 0.0 0.0 Hawks 0.0 0.0 0.0 100.0 0.0 0.0 0.0 0.0 0.0 Heat 0.0 0.0 0.0 0.0 100.0 0.0 0.0 0.0 0.0 Bulls 0.0 0.0 0.0 0.0 0.0 96.0 4.0 0.0 0.0 76ers 0.0 0.0 0.0 0.0 0.0 4.0 96.0 0.0 0.0 Pistons 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100.0 0.0 Eastern Conference Division Title (Percent) Code: Team Southeast Title Percent ==== ====================== Magic 100.0 Hawks 0.0 Heat 0.0 Team Atlantic Title Percent ==== ====================== Celtics 100.0 76ers 0.0 Team Central Title Percent ==== ====================== Cavaliers 100.0 Bulls 0.0 Pistons 0.0 Eastern Conference Playoff Advancement (Percent) Code: Team To 1st To 2nd To Conf To Fnls Champs ==== ======= ======= ======= ======= ======= Cavaliers 100.0 97.3 92.0 63.5 47.3 Celtics 100.0 93.8 60.7 23.5 13.9 Magic 100.0 91.9 37.9 12.1 6.4 Hawks 100.0 65.6 5.4 0.6 0.1 Heat 100.0 34.4 1.7 0.1 0.0 Bulls 100.0 8.0 0.7 0.0 0.0 76ers 100.0 6.3 0.7 0.0 0.0 Pistons 100.0 2.7 0.9 0.1 0.0 ======================================== Rockets Home Court Playoff Matchups By Round (Percent) Code: Team 1st 2nd Conf Final ==== ===== ===== ===== ===== Mavericks 30.8 4.8 0.7 0.0 Hornets 21.7 6.5 0.6 0.0 Blazers 12.4 0.0 4.7 0.0 Spurs 10.9 0.0 2.1 0.0 Nuggets 0.0 15.4 0.0 0.0 Jazz 0.0 2.9 1.0 0.0 Hawks 0.0 0.0 0.0 0.1 Heat 0.0 0.0 0.0 0.0 76ers 0.0 0.0 0.0 0.0 Pistons 0.0 0.0 0.0 0.0 Rockets Away Playoff Matchups By Round (Percent) Code: Team 1st 2nd Conf Final ==== ===== ===== ===== ===== Blazers 24.2 0.0 0.9 0.0 Lakers 0.0 18.8 14.7 0.0 Nuggets 0.0 11.0 2.2 0.0 Spurs 0.0 0.0 1.5 0.0 Cavaliers 0.0 0.0 0.0 6.6 Celtics 0.0 0.0 0.0 2.4 Magic 0.0 0.0 0.0 1.2 Rockets Combined Playoff Matchups By Round (Percent) Code: Team 1st 2nd Conf Final ==== ===== ===== ===== ===== Blazers 36.6 0.0 5.6 0.0 Mavericks 30.8 4.8 0.7 0.0 Hornets 21.7 6.5 0.6 0.0 Spurs 10.9 0.0 3.6 0.0 Nuggets 0.0 26.4 2.2 0.0 Lakers 0.0 18.8 14.7 0.0 Jazz 0.0 2.9 1.0 0.0 Cavaliers 0.0 0.0 0.0 6.6 Celtics 0.0 0.0 0.0 2.4 Magic 0.0 0.0 0.0 1.2 Hawks 0.0 0.0 0.0 0.1 Heat 0.0 0.0 0.0 0.0 76ers 0.0 0.0 0.0 0.0 Pistons 0.0 0.0 0.0 0.0 Rockets Home Court Advantage By Round Code: Round Percent ===== ======= First 75.8 Second 29.5 Conf 9.1 Finals 0.1 =============================== As many have pointed out, with yesterday's results, we will not be meeting the Jazz in the first round. There is nothing much to add except: Go Rockets!
By the way, I just checked Hollinger's playoff odds and his is wacky. It has the Lakers and Utah out of the playoffs! Also, Basketball Reference's model gives the Rockets only a 40% of getting the division. That seems very low given the scenarios. I'd say that my model is currently better!
