FiveThirtyEight: Rockets 79% likelihood of beating Warriors

61 posts / 0 new
Last post
May 14, 2018 2:02pm
mannycoon
Online
Last seen: 8 min 34 sec ago
Joined: Nov 2, 2011
I think their sports coverage is largely irrelevant, and I dont think anyone takes their sports projections too seriously. They are first and foremost a politics site, and their record with senate, governor and congressional races is reasonably strong, which is where they get most of any credibilty they have. They might try to hype it up as more, but to me most their sports systems feel like fun toys they get to play around with on the side to their real work. Sports data is good to play around with because there is tons of it, sports are a fun topic for a lot of people and the stakes are pretty low if you make some mistakes in the process and you can learn from those mistakes and apply them to other processes.
May 14, 2018 3:25pm
Clippers1121
Online
Last seen: 2 min 3 sec ago
Joined: Nov 15, 2011

My fix is very simple.  Run the model against the last 10 playoff games that the Rockets and Warriors have played and throw out the regular season statistics they are using.  They way they are measuring the current form and how players are playing against a high level of competition.  I think the reason they don't do this is because they would rather have methodology that was consistent even though it was wrong then change their methodology every time an outlier comes up like the Cavs or the Warriors.  I don't think it is a game to them.  They built a good mathematical model that will work in most cases.  They just refuse to change it for those cases where it does not work.

May 14, 2018 9:37pm
Dyce
Offline
Last seen: 3 hours 6 min ago
Joined: Nov 27, 2011
Two problems with the model. 1. Doesn't take into account that teams play differently in the playoffs. Tighter rotations, more dependency on stars. 2. Doesn't take into account the injuries the Warriors dealt with during the regular season.
May 15, 2018 8:21am
Clippers1121
Online
Last seen: 2 min 3 sec ago
Joined: Nov 15, 2011

After seeing game one I think their model is even more pathetic than I did prior to the game.  If all they had used was their ELO and Carmelo ratings and just added up the values for the top players on each team they would have been OK.  Instead they add in wins produced and point differentials for regular season games and come up with the metric that the Rockets are 10% better than the Warriors since they won 10% more games.  Then they come up with the metric that teams who are 10% better than the teams they are playing against win 80% of seven game playoff series especially with HCA.  Now the Rockets have dropped from a 79% chance of winning the series to 53%.  After seeing the game most sports pronosticators would not even give them a 10% chance of winning the series.  Not that they lost the game but how they lost it would be the decisive factor.  Warriors just had more and better players than they did.

https://projects.fivethirtyeight.com/2018-nba-predictions/?ex_cid=rrpromo

May 15, 2018 9:32am
gilp5
Offline
Last seen: 53 min 35 sec ago
Joined: Dec 5, 2011

Clippers1121 wrote:

After seeing game one I think their model is even more pathetic than I did prior to the game.  If all they had used was their ELO and Carmelo ratings and just added up the values for the top players on each team they would have been OK.  Instead they add in wins produced and point differentials for regular season games and come up with the metric that the Rockets are 10% better than the Warriors since they won 10% more games.  Then they come up with the metric that teams who are 10% better than the teams they are playing against win 80% of seven game playoff series especially with HCA.  Now the Rockets have dropped from a 79% chance of winning the series to 53%.  After seeing the game most sports pronosticators would not even give them a 10% chance of winning the series.  Not that they lost the game but how they lost it would be the decisive factor.  Warriors just had more and better players than they did.

https://projects.fivethirtyeight.com/2018-nba-predictions/?ex_cid=rrpromo

Good post.

May 15, 2018 11:38am
Hitnrun24
Offline
Last seen: 53 min 12 sec ago
Joined: Dec 6, 2011

People are really burying the Rockets. I think the Warriors will get by them and still do, but it's not the kind of loss that I felt was indicative of there being no hope.

 

The Rockets tend to look really bad when they lose because of the iso ball, but they can fix a lot of the stuff they did poorly. They can't keep running the clock as low as they did, can't let Klay have that many open looks, and have to do better on their matchups with KD. I also don't think Steph looks 100%. They have to put Luc or Tucker on Durant more and live with their offense. Chris can play a lot better too, they're not out of it yet.

May 15, 2018 2:42pm
Clippers1121
Online
Last seen: 2 min 3 sec ago
Joined: Nov 15, 2011

But 538 said they had an 79% chance of winning the series.  That would make them have like 90% chance of winning at home and 70% chance of winning on the road.  Somebody with a 90% chance of winning would have to have the opposition shoot really well or basically miss most of the shots they normally make.  Or at 90% both of those things would have to happen.  What did happen was the Warriors played their normal game and the Rockets played their normal game and the Warriors won by 13 points.  That would never happen with a legitimate 90% favorite.  You would need it to be a fluke for a Warriors win.  And that win was no fluke.  This thread is about the whether the 538 NBA playoff prediction model is accurate or not.  I have to say last nights game was proof enough that it isn't regardless of what the Rockets do or don't do in the future.

May 15, 2018 2:49pm
Hitnrun24
Offline
Last seen: 53 min 12 sec ago
Joined: Dec 6, 2011

Is anyone defending the model on this pick? It is what it is, I don't think anyone thinks that model is in any way accurate on this prediction. I'm ok with them not changing their model based on one bad prediction for a team that would be hard to give accurate information on based on regular season performance.

May 15, 2018 2:54pm
gilp5
Offline
Last seen: 53 min 35 sec ago
Joined: Dec 5, 2011

Hitnrun24 wrote:

Is anyone defending the model on this pick? It is what it is, I don't think anyone thinks that model is in any way accurate on this prediction. I'm ok with them not changing their model based on one bad prediction for a team that would be hard to give accurate information on based on regular season performance.

Assuming GSW does win, this is not their first model to incorrectly pick the winner (see 2016 election).

May 15, 2018 7:19pm
mannycoon
Online
Last seen: 8 min 34 sec ago
Joined: Nov 2, 2011
538s reputation mostly comes from 2008, were they were 50 for 50 on senate rates and 49 for 50 on states in presidential election. 2016 didnt go as well for them, but were generally speaking less bad than other most other systems. Their basketball coverage is basically irrelevant.