https://projects.fivethirtyeight.com/2018-nba-predictions/

We've had conversions on this board of the accuracy of FiveThirtyEight. Now they say the Rockets are highly likely to beat the Warriors (79%).

So if (when) the Warriors beat the Rockets in 5 or 6 games, will you say they were still accurate because they left a 21% chance of the Warriors winning?

I think the statistical models of this organization are wrong.

GSW are favored to win the series at all of the betting sites.

Hou has home court and are favored to win Games 1&2 by -2.

If Rockets beat GSW, will you trust 538 more?

Yes.

538 is just a model and the models like Houston because they were the strongest regular season team and they have home court.

Warriors at -195 in Vegas. That means almost two to one they are going to win the series. And that is people putting money on the games. About 35% chance the Rockets will win. I actually like the Warriors at these odds. In fact, these odds of winning this one series are better than if you had bet them to win the championship before the season even started. Their odds of winning the championship are 5/8. Which means you don't get that much more if you bet them to win the next two series. This 538 projection is flawed in that it does not take into account the Warriors getting healthier and more motivated during the postseason. And they are not putting any of their own money at stake so what the hell do they care.

They put their money at stake by making a public prediction. It's their fucking business. When they get things wrong, it costs them a whole lot of money. They are based almost exclusively on their reputation.

Does that mean if you bet on the Warriors to win, you will win $100 but could lose $195?

Yes exactly. You bet $195 to win $100. Last years finals the best odds I could find were Warriors -270. The 538 people are not really concerned about their sports predictions. That isn't what they are getting paid for. Its just entertainment to drive more traffic to their web site. The Vegas people are the ones who put their money where their mouth is.

So why not qualify the prediction by saying they don't believe it or it could be flawed? They aren't saying that. If you go to the website they are saying flat out that Houston will win with a 78-79% probability. They are far more confident of Houston winning than the other prediction (Celtics winning with 56% probability - which is basically a toss up).

So if Houston loses, will people here say they were still correct because they did not say 100% probability? Or will you say the model is flawed?

The problem though is that the Rockets could be 40% likely to win (as most people think) instead of 79% and there is no way to prove them wrong. They could say Rockets are 99% likely to win and still claim to be correct because they left a 1% chance.

I doubt that 538 even accounted for the fact that Ricky Rubio missed the entire series against the Rockets, that Exum got injured in game four first half and did not play again, and that Mitchell got injured in the second half of game five and did not return to the game. They just run their metrics as if the Rockets had beaten a healthy Jazz team. While the Warriors actually did beat a healthy Pelicans team and improved their health considerably by getting Curry back. If they don't account for what actually happens and why the metrics don't really tell the whole story then they are not making accurate predictions. And no bookie would pay them to tell them that the Rockets have a 79% chance of beating the Warriors. Bettors who believe this can go to Vegas and bet the Rockets at +170 instead of at -400 which is an enormous profit margin based on the odds. But the 538 odds are complete crap based on obsolete and incomplete metrics.

Vegas is not in the business of predicting the outcome, they are in the business of predicting which way people will bet. They need half the betters to bet on one side and the other half to bet on the other, and they get their cut when the game is over. That's why the line changes as the betting progresses, if too many more people bet on one side, they adjust the line to even things out. They need to end up with half the people betting on one team and the other half betting on the other team. If, for example, 60% of the betters bet on one team and that team wins, Vegas loses money. Vegas doesn't lose money.

Vegas is not in the business of predicting the outcome, they are in the business of predicting which way people will bet. 538 is not in the betting business, they are in the statistical analysis business.

This debate makes no sense.

+1. I think the model is wrong and Houston's has less than 50% chance of winning. If you believe in 538's model though, this is an excellent opporunity to go to Vegas and place a huge bet.

Vegas absolutely does exactly what Corkscrew described. They have to start someplace though and right now they starting the betting as the Warriors being almost a two to one favorite to beat the Rockets. Now that line could move up or down but it isn't going to move by much. If the line moves too much people can bet both sides of the bet and win both sides. So they have no interest in having the line move and even will take big losses to prevent it from moving too much. So if effect by making the Warriors 2 to 1 favorites they are predicting a Warriors advantage and expect equal action on both sides of the bet. If people believe the 538 projection they would go to Vegas and place heavy bets on the Rockets. So Vegas and 538 are both making predictions for different purposes. The debate is whether the 538 prediction is accurate and what metrics they are using to make it.

The Vegas line sort of represents an aggregation of subjective analysis of every one betting and for the most part that can be a good predictor.

However they don't always set their line ups to get bets to line up prefectly and they do sometimes purposely skew their lines just a little to "bet" on a certain team if they think the betting market is irrational. For example in this case, lets say computer prediction was accurate (which it's not), then ideally for Vegas as many people would bet on the Warriors as possible, so if they were willing to take the risk they might slightly increase the payout for the Warriors to win bets to encourage it, but so much they get completely creamed if the Warriors do win; some bookies do this more than others. The with way the betting is set up and the amount of data the casino has, they can combine more educated that the general population betting and still take the vig the process, rather than pay one.

Tullabye do you believe that Houston will win with a 79% likelihood?

If I tell you there is a 50-50 chance of flipping a coin heads instead of tales, and you flip a coing 10 times and it come up tails, you have not proven me wrong, my statistical analysis is still correct, it is still 50-50. If you flip the coin 100 times or 1,000 times and they all come up tails, my analysis is still correct.

Of course, flipping tails 10 times in a row is a rare occurrence (1024 to 1, actually), and 100 times in a row even rarer (1,267,650,600,228,230,000,000,000,000,000 to 1). But still, each single flip is a 50-50 flip. That number is still always correct because there are two possible outcomes and a single flip.

Statistics are not predictive. The human inference of the statistical analysis can be predictive, but the numbers by themselves are just that, numbers, calculations,

If the Warriors win the series they have not proven 538 wrong, and if the Rockets win the series they have not proven 538 right.

This debate is silly. It would be much more interesting to ask how 538 arrived at that number, and maybe debate the methodoligy. But to pass a verdict on 538's credibility based on whether or not one team or another wins the series is absurd.

That's partly my point. They can claim 99% probability Houston will win and you can't prove them wrong. But nobody here would say Houston has a 99% probability of winning or even a 79% probability of winning.

People go to them to see who is likely to win. And, in a larger sense though, they are actually calling a winner and we'll see if that is correct (as it was not during the presidential election).

People can make up any probabilities that they want but they don't mean a damn thing if it does not cost them any money if they are wrong. On a fair coin toss everybody will concede that it is 50/50 whether the toss comes up heads or tails. Now if I said it would come up heads 80% of the time you would say that is bullshit and you would be right. So if 538 says the Rockets have a 79% chance of winning against the Warriors people can call them out that their numbers are bullshit and be right. The proof is the Vegas odds which have the Warriors as almost a two to one favorite. To accept odds based on faulty metrics is really dumb. I can prove by the coin flip example that coins are not going to come up heads 80% of the time even though past metrics may point to the fact that it will. You have to exercise some common sense and say 538 is wrong regardless of what their metrics are telling you.

The Vegas line gives you a consensus of what oddsmakers think of who will win the series. And they try and keep it pretty consistent throughout all the casinos. And their opinion usually is representative of what the betting public thinks to produce an equal amount of bets on both sides like mannycoon stated. So yes the 538 site can't be proven wrong with their projection but the fact that they are not taking any bets from people who disagree with them and no bookie in the country is paying for their projection means they are just clowns doing this for entertainment and web clicks. And the Cavs are -290 favorites against the Celtics. Which means they are almost three to one favorites to beat them. So a 25% chance of a Celtics win is a far cry from the 56% that the 538 site gives them. So 538 is not flipping coins so you can't exactly quantify their predictions but anybody with even a shred of basketball knowledge could determine they are complete bullshit.

I suspect the 538 methodology is based on regular season performance, team performance metrics, homecourt advantage, win differential, etc... I suspect it does not account for the fact that the Cavs and Warriors becoming a different type of beast once the playoffs start. And they probably don't account for the playoff experience of the players on both teams which the Cavs and Warriors have big advantages in. The proof that 538 is wrong is that their metrics used in their predictions are wrong. Just like if I flipped the coin 1,000 times and it came up heads 800 times it would not prove that the coin had an 80% chance of coming up heads. It would just prove that my methodology for coming up with that number was wrong.

+1. I totally agree 538 is clickbait.

People don't get the relevance or importance of this argument which is unfortunate. In 2008 people put out metrics that the mortgage failure rate could not exceed more than 5% of the existing mortgages on the market. That is what the metrics and past performance of mortgage products suggested was the maximum failure rate. Then Bush comes in and says everybody should be a homeowner so a bunch of subprime mortgages were issued to people who could not really pay them. But people believed the 5% metric number was good when the actual failure rate went to 25%. Results were trillions of dollars in losses to people left holding the bad mortgages and a financial collapse of several banks and mortgage brokers. Lesson is don't believe metrics and projections that are complete bullshit. Analyze what is actually happening and call out bullshit for what it is. So of course the argument over what constitutes good metrics is an important one that impacts your daily life. What you buy, what companies you invest in, where you want to live.

For this discussion a better metric than what 538 is using might be that the Warriors and Cavs have already met in the finals the last three years and they have many of the same players. So why wouldn't they both be favorites to advance.

How do you know if you don't know the methodology?

That's not proof at all, as I explained earlier, Vegas odds a only predictors of how the public will bet, not predictors of the outcome.

Rather than suspecting, perhaps check out the methodology first? I may be very wrong or it may be right, but the fact that you are making an assertion based on "suspition" without checking out the methodology first indicates you are making a judgment based on a personal bias. Again, your bias may be correct, but you don't know that without knowing the methodolgy.

You haven't even checked out their metrics. Your "suspicions" magically become facts?

You know a blindfolded monkey drawing colored marbles out of a giant bowl could make more accurate predictions than these guys. Of course you do or you would be draining your bank accounts and placing massive bets in Vegas on the Rockets and the Celtics. So you know the information is bad. You are just making the case that these guys are authoritarian figures so we should take their information seriously unless we can prove them wrong. You would be much better off taking your betting advice from a 10 year old child who had never seen an NBA game in their life. Use a little street smarts and call these guys out for being the sham which they are.

You mis-spelled Clinton.

I wondered who would call him on that one, LOL.

If the Warriors sweep the Rockets or beat them pretty easily in 5 which some people do believe, the 79% will in fact look silly because while you can argue 79% is not 100% so anyone can win you can not reasonably argue the series being uncompetitive in the Warriors favor. If the Rockets win in 5 then maybe we will have to take a long look at their metrics. I feel like it will be a competitive series though so it will still end up being ambiguous in the end.

Thanks for the correction. I honestly thought that Bush was behind the mortgage fiasco since it happened while he was president and he went on national TV and said he thought that it should be a home ownership society where everybody owned their own homes. Did not know it started with Clinton policies. And there is this:

https://www.nytimes.com/2008/12/21/business/worldbusiness/21iht-admin.4.18853088.html

So you could say that Clinton got the ball rolling on the housing crisis but the Bush adjministration did nothing to prevent the mortgage market from collapsing even though they could have. So I would have to give Bush just as much credit for the financial collapse as Clinton if not more. Let's not start a partisan fight over this. Like the Vietnam war. Kennedy started it and Johnson prolonged it for the five years he was in office. Who is more guilty? Johnson for his failure to change the course of what was going on while he had the chance to.

OK. Let's look at what 538 is doing. First they use their ELO rating.

https://en.wikipedia.org/wiki/Elo_rating_system

They rate each player of the two teams and determine their skill level. They combine this with their Carmelo rating system:

https://projects.fivethirtyeight.com/carmelo/

Somehow they have combined these two ratings systems and come out with the projection that the Rockets are 79% favorites to win the series. So obviously the Rockets players got much higher ratings than the Warriors players did. But we all know these ratings go down the toilet when the Rockets players face defenders like Klay Thompson, Draymond Green, Kevin Durant, and Andre Iguodala. Whatever the Rockets players did against lesser defenders does not mean a damn thing once they come up against a team of stout defenders. Skill points built up by Rockets players against poor defensive opponents are negated by the defense. Why do you think we lost to the Warriors 10 times in a row when we were a highly skilled team with a good record that should have beat them a good percentage of the time based on our offensive numbers. It just does not work like that against an elite defensive team. Basically Vegas has it as Warriors in six. I will go with that projection.

It's a tricky series to call, but it's obvious that the Warriors should be the favorites and that they have a better overall team when you compare the top 10 performers. The reason it's tricky is because of matchups and minutes, regular season vs playoffs. Houston has the best top-3 players, but has to give a lot of minutes to below-average players like PJ Tucker, Eric Gordon, and Gerald Green. GSW is equal to the Rockets if you look at the top-5 guys, but after that they need to skip players like Nick Dumb (who Kerr says he is going to play because he is tall), less Livvy (who has really dropped off), and more Looney and West. A big key is whether going without a C continues to work for GSW against a team with a dominant Clint Capela having his way inside. It's very different playing against Clint and his 1.5 points per shot and dominant rebounding than Anthony Davis. More Javale McGee, I would play.

We all know first hand that wearing down Chris Paul so that he's too tired to turn the corner late in games is a big key. Except that here there is James Harden.

What I don't like about the 538 metrics is that in game 5 of the semi-finals with the Rockets playing the Jazz an elite defender like Rudy Gobert was able to take away Capela rolling to the basket and getting easy hoops. Capela isn't elite offensively against an elite defender. And the 538 methodology does not account for that. Only way the Rockets won that game was Donovan Mitchell getting injured, Ricky Rubio and Dante Exum already injured, and CP3 launching and hitting a bunch of three pointers in the fourth quarter that a team like the Warriors could easily take away from him. If your methodology is based on rating players playing against average defenses that don't have time to prepare for your offensive sets then those metrics are not going to hold true against elite defenders that do have time to prepare. Plus the metrics used by 538 are high for the Rockets because they played their asses off in the regular season while the Warriors were the defending champs who were not chasing anything so they loafed through the regular season which would lower their ratings.

I still say the 538 projection is bullshit even after having looked at their methodology and that I would go with the blindfolded monkey drawing marbles or the 10 year old who has never seen an NBA game over their projections.

I guess I will come out and say it - if I were forced to bet on this series, I'd bet on the Rockets.

I know that's not popular. I know it might make me look foolish in retrospect in a couple of weeks, but that's how I'd bet if I had to.

Its fine to bet the Rockets if you are getting +170. But 538 is saying you should be getting -400. So the question isn't whether you would bet the Rockets but would you bet the Rockets if you had to put up $400 to win $100.

Celtics just creamed the Cavs which means 538 is looking good so far. But whatever the outcomes of the games are it doesn't make their methodology wrong or right. The games don't prove anything as Corkscrew pointed out anybody can flip a coin and have it land heads four times in a row even though the odds are against it. My problem with their methodology is that it gives great weight to Harden being able to score 50 points in a regular season game and no weight to Klay Thompson being able to prevent him from scoring those fifty. It gives great weight to Capela being the roll man in their pick-and-roll offense and no weight to somebody like Rudy Gobert being able to stop that play and knock down Capela's offensive rating. Their methodology uses objective metrics but are useless because they are based on regular season performance. What they should have done is just use playoff statistics. That way they would give the Warriors at least a 50% chance of beating them. That probably isn't close to the true odds of a Warriors win but it is a lot closer than their prediction.

They have home court, and they are favored at home. If they win game 1, their spread grows. So it's not a bad bet

They should use regular season and playoff stats of specific players guarding specific players.

https://projects.fivethirtyeight.com/complete-history-of-the-nba/#warriors

It is strange that last years Warriors team got the highest ELO rating from them in NBA history at 1865. It beat the 1996 Chicago Bulls team that got a rating of 1853. Now the Warriors have the experience of winning a title, many of the same players coming back, and just won eight of their last 10 playoff games this year and they only give them a 20% chance of beating the Rockets. Nobody is calling this Rockets team the greatest in NBA history like last years Warriors team. They have seriously screwed up here. Gave way too much weight to their Carmelo ratings with their BTM stats and their RPM stats. I guess the Rockets produced a bunch of good statistical games this season and got a very high Carmelo rating. Whatever. Mbah a Moute may be able to get a +/- of 57 in a game but that isn't going to help them against the Warriors. They will still let Curry guard him because he can't shoot. Or put Curry on Ariza. Thompson on Harden and Iguodala on CP3. Maybe Durant on Capela. Or maybe Green.

Hey I just can't get there. Regardless of what statistics are used the Warriors players still should come very close to the Rockets players in both ELO and Carmelo rankings. I mean an 80% is a prediction you would expect in the first round or something. Even the worst mathematical model would give the Warriors at least a 50% chance of winning just based on the players they have on their roster, their record in the playoffs so far, their record in the regular season, and their ELO rating and title from last year. The model is broken and needs to be fixed.

What metrics would you use to fix the model though? That's the point most are making, we know there are nuances that can't be measured but how do you tell a computer how much of a slip in defense is because of championship fatigue?

Personally I would put it at about 65% for the Warriors because I just see less ways to beat them. The versatility in their frontcourt allows them to play in any style and having Green allows them to make a lot of Centers less effective, Capela is a big X factor. While I think Gobert is better, Capela's skill set is better suited to translate on the court vs the Warriors.

So you admit the model appears to be flawed, as your prediction is vastly different than 538.

Why can't we criticize the model without proposing how to fix it? It is perfectly fine to say this model is flawed without proposing a new model.

My fix would be to not have a so-called statistical model at all - but then 538 wouldn't be able to make clickbait money.

Sure, it's flawed. But I'd also guess that in most series their model probably works pretty well. Golden State and Cleveland are 2 outlier teams that play very different in the postseason, but I'd be curious to see how it does with other teams and series where there is more useful traditional data.

Their ability to stay in business depends predominantly on their credibility and track record. Their motivation is to maintain these at a high level, which they have done exceptionally well over the years. Let them fail for a period of time and there goes their business. Do you actually think they would risk their reputation by falsly formulating a 79% probability just to create "click bait."?