FiveThirtyEight giving the Celtics an 83% chance to win finals

[DroseReturnChi]

yep i told u this stat is the dumbest ever. 83% are you kidding me?

[Saints14]

538 has a seriously terrible basketball model

[TinmanZBoy]

538 in Chinese means “I am a man who likes gossiping” yes, that is all what they do

[JayMKE]

lol always funny when the condescending nerds at 538 embarrass themselves, Im sure they’ll pivot and say they were correct about the odds the entire time or whatever their usual cope answer

[MrGoat]

Yeah this deserved a bump. 538 was always better at election predictions. Some of their sports models are... questionable

[Archerbro]

538 is pretty much garbage.
I learned that back in 2016 and from subsequent elections (including midterms).

Nate peaked along time ago.

It's useful for sure, but not even close in many years. and the model will be off (and not slightly off, off big time) and then "adjust," making it appear like it was a good one to begin with.

[Capn'O]

538 has two models: the ELO model and the RAPTOR model.

Their RAPTOR model gives the Celtics an 83% chance of winning and their ELO model gives the Celtics a 68% chance of winning.

The Celtics have played amazingly well both offensively and defensively overall for a long period of time and 538 RAPTOR weighs that heavily.

Most betting sites are giving the Warriors a 55-59% shot to win the title.

When I first started diving into metrics, my friend who was teaching me had me start with baseball (2014). And he clued me in to something called the Mike Trout rule. And the Mike Trout rule was simple: if you are making an overall metric, and Mike Trout isnt at the top of it, you **** up

As the years have gone on, I've done something similar except with Vegas. If your predictive model is in wild conflict with Vegas... you **** up. But as I said on the other thread, predictive models for individual series in continuous clock sports are a trash concept to begin with - unlike oft misused advanced metrics like PER and RPM, I dont think they carry much value at all, other than to generate clicks by giving an (or any) answer)

I would kill to know how Vegas does their oddsmaking so accurately across such a wide scope of sports.. but I kinda get the feeling that you'd actually have to, to find out

Warriors aren't a bad analogue to Mike Trout. If they're in the mix and you're giving them a 17% chance something is awry.

There was never a mere 17% chance that the Warriors would win the series. They were the better team with the best player in the series.

[sp6r=underrated]

This thread is a great window into people who think about probabilities vs people who think in binary terms. For the later this series completely invalidated the model for the former you need to look at overall performance of the model in every series and beginning of season predictions.

[EArl]

538 has a seriously terrible basketball model

They also predicted Clinton would win and we ended up getting four year of Trump. I would say that its flawed in more than basketball.

[liquidswords]

Data is a tool and not the end game because we're talking about people with emotions. Did the model predict Tatum would have a meltdown?

[Effigy]

Yeah this deserved a bump. 538 was always better at election predictions. Some of their sports models are... questionable

Didn't their model say Hilary was going to beat Trump pretty handily?

[Capn'O]

This thread is a great window into people who think about probabilities vs people who think in binary terms. For the later this series completely invalidated the model for the former you need to look at overall performance of the model in every series and beginning of season predictions.

But there's a third category of "thinking in basketball." Watching the Warriors in these playoffs... they were just dominant. They waxed two MVP candidates and, aside from a real clunker, convincingly beat a very strong team in the Grizzlies. They spent most of the year getting healthy and entered the series at full tilt. Then you have the experience factor. Meanwhile, the Celtics struggled with a team that was basically the Nuggets/Mavs in the Bucks without Middleton.

I'm not saying throw the baby out with the bathwater but the human factor ENTERING this series could easily see where this prediction was way off. There was absolutely no way the Celtics were a 5-1 favorite. If it had been a nailbiter, I'd have a little less conviction in this but _MY_ models had the Warriors as a heavy favorite as simply the better team.

[hoosierdaddy34]

Yeah this deserved a bump. 538 was always better at election predictions. Some of their sports models are... questionable

Didn't their model say Hilary was going to beat Trump pretty handily?

How many times do we freaking have to go through this? The concept of probabilities cannot be this hard to understand. They gave Trump a 30% chance of winning, which was much higher than any other model. And given Hillary was polling 4 pts higher and won the popular vote by around 4%, that was pretty freaking astute of them.

[Eagle4]

Yeah this deserved a bump. 538 was always better at election predictions. Some of their sports models are... questionable

Didn't their model say Hilary was going to beat Trump pretty handily?

How many times do we freaking have to go through this? The concept of probabilities cannot be this hard to understand. They gave Trump a 30% chance of winning, which was much higher than any other model. And given Hillary was polling 4 pts higher and won the popular vote by around 4%, that was pretty freaking astute of them.

Hey 538 damage control representative.

[Nate505]

I'm sure their models for basketball are terrible and all, but beating 83% odds isn't all that impressive or mind blowing. It's the equivalent of saying you'll roll a 6 (or any number) on a die and that number hitting.

[Capn'O]

Yeah this deserved a bump. 538 was always better at election predictions. Some of their sports models are... questionable

Didn't their model say Hilary was going to beat Trump pretty handily?

In terms of electoral votes, yes, but there were a bunch of states where the predicted margin was very close (they had Hillary slightly ahead) and Trump won them all. Their political model, which was roughly an amalgamation of poll results, was very good but the game changed on them there.

[hoosierdaddy34]

Didn't their model say Hilary was going to beat Trump pretty handily?

How many times do we freaking have to go through this? The concept of probabilities cannot be this hard to understand. They gave Trump a 30% chance of winning, which was much higher than any other model. And given Hillary was polling 4 pts higher and won the popular vote by around 4%, that was pretty freaking astute of them.

Hey 538 damage control representative.

No damage control needed. Just helping the less fortunate understand simple concepts such as probabilities. It’s my good deed for the day educating those that desperately need it.

[FNQ]

538 has a seriously terrible basketball model

No one would have a good one that isnt supplemented with subjective data, almost to the point where the subjective data overwhelms the objective

Predictive models are just toys for the most part. But most especially the free ones… the best Joker had it right - if you’re good at something, never do it for free. Free metrics are used a ton here as proofs in argument.. but if that data was valuable, it would be protected

[Capn'O]

538 has a seriously terrible basketball model

No one would have a good one that isnt supplemented with subjective data, almost to the point where the subjective data overwhelms the objective

Predictive models are just toys for the most part. But most especially the free ones… the best Joker had it right - if you’re good at something, never do it for free. Free metrics are used a ton here as proofs in argument.. but if that data was valuable, it would be protected

Yes, yes, and yes. We're pulling stuff from bbref and not Synergy, for example. And certainly not the kind of data that teams use which is much more granular to certain shots/movements.

Even bbref has paywalled their fun stuff

[shoresy69]

Didn't their model say Hilary was going to beat Trump pretty handily?

How many times do we freaking have to go through this? The concept of probabilities cannot be this hard to understand. They gave Trump a 30% chance of winning, which was much higher than any other model. And given Hillary was polling 4 pts higher and won the popular vote by around 4%, that was pretty freaking astute of them.

Hey 538 damage control representative.

It's not damage control, I remember their model showing Trump's chances climb up to 30% the week of the election and getting a sinking feeling in my stomach. 30% was wayyyy higher than anyone else was reporting at the time.