Why is +/- a bad stat?

[Throw More Waffles]

It's a great stat, but like with any other stat, context matters.

If a team is -30 in goal differential on the year, and its best player is +20, that speaks volumes.

Similarly, if a team is+30 in goal differential on the year, and it's worst player is -20, that also speaks volumes.

Using the leafs (my team) as an example, the teams is +40 in goals. Matthews and JVR have a similar number of goals and points. Matthews (famous for being responsible away from the puck) is +24. JVR (famous for being irresponsible away from the puck) is +1.

It's a very useful stat.

 

[CantLoseWithMatthews]

1) It's pretty misleading and inaccurate and doesn't really give you useful information about what has happened
2) it has no predictive value

 

[lomiller1]

1) It's pretty misleading and inaccurate and doesn't really give you useful information about what has happened
2) it has no predictive value

I forget who it was, but someone once showed that it could predict value of free agent signings, but not quite the way it's proponents think. The free agent with the lowest +/- were likely to be the best value signings in the offseason while the ones with the best +/- were the worst.

 

[CantLoseWithMatthews]

I forget who it was, but someone once showed that it could predict value of free agent signings, but not quite the way it's proponents think. The free agent with the lowest +/- were likely to be the best value signings in the offseason while the ones with the best +/- were the worst.

really? That's pretty interesting
for the Leafs, I guess I'd say Hainsey has been bringing better value for us than Marleau, and his +/- was lower

 

[llwyd]

I don't think it's a bad stat but it doesn't really work in isolation, you have to know quite a bit of things about the team for it to make sense.

 

[Mike Lowry]

I've never heard anyone say that the reason +/- didn't accurately predict an outcome was because of luck, sample size, or tell us how it was adjusted. So while it's not a great stat in the overall scheme of things, it is far better than the corsy and fenwich garbage that came out of the basement. The game is won because you score more than the other team, not because you attempted more shots

In 2004 Rick Nash tied for the league lead in goals scored with 41. That stat on it's own might give the impression he had a great year, but when you add to that he only had 57 points, and was a -35, it shows his year was not good. So one stat on it's own doesn't show you the whole story

 

[SotasicA]

The math has conclusively shown that any benefit from including shot quality information is massively outweighed by the noise from seeing eye shots, tips, shots that just make it in the net vs shots that hit the post, etc.

If you want to look at a team/player with luck included use GF%. If you want to look at how they would have performed without luck coming into play use CF%.

No such thing as luck. Putting the puck in the net is not "noise". It's the purpose of the game. There's no right way or wrong way to score or prevent a goal. You don't have to generate as many high scoring area shots as your opposition, if you win the game without them.

What you are talking about is predicting future performance, because generating scoring chances is more repeatable, and thus a team that does it more is more likely to win. But that's not what +/- is for. It's a tangible record of what just happened.

 

[KevinRedkey]

If you need to score more goals... you'll probably scout players with high goal totals, right?
If you need a center that can win faceoffs, you'll probably look at players that win a lot of em, right?

So at what point do you ever look at a player's +/-? What are you looking for in a player, if that's a stat that matters at all? There's nothing you could want (playmaker, goal scorer, fighter, defensive forward, Dman, faceoff specialist, etc...) where +/- would be even close to the first indicator of whether or not a player is a fit for the hole you need to fill in your roster.

Fun fact:
In 2009-10, If all I cared about was +/-, I'd have thought Jeff Schultz was a superstar, and Scott Niedermayer was a scrub.

 

[Spirit of 67]

Yeah but you could do the same and get a secondary assist

Yeah but you could do the same and get a goal.

 

[lomiller1]

really? That's pretty interesting

It implies that NHL teams were paying for +/-. When someone is willing to overpay for something there is value to be had doing the opposite. It’s similar to "smart money" betting against the public favorites, or shorting in the stock market.

for the Leafs, I guess I'd say Hainsey has been bringing better value for us than Marleau, and his +/- was lower

There is a significant body of work going back to his days with the Jets that Hainsey has been one of the more underrated D-men in the NHL for years. He’s easily the best defensive D-man the Leafs have had in a long time and makes an ideal partner for just about any offensive puck moving D-man.

 

[ulvvf]

Whether you care about past goals or future goals is key. GF% most accurately describes what goals occurred, CF% outperforms it for predicting what goals will be scored. +/- is not GF%, it’s just a broken useless statistic. Don’t use it period.

I would say it depends on the sample size then, if the sample size is small then corsi maybe become more telling about the future than more regular +/-, because in 1 game, it can be little a coincidence who scores against who. But if the sample size is big, then it is less coincidence who scores most., therefore +/- tells more about future success than corsi does when sample size is big.

Look at this way, if we would just focus on the plus side, on both corsi and +/-. If we have 2 players, Player 1 scores 110p in 82games, while player 2, scores, 50p in 82 games. But player 2 have more often been on the ice when his teams shoot. Is corsi then a good way to measure future success? Is it more likely that player 2 will score most points the next season? No when the sample size becomes big, it is more telling who actually get the puck in the net not who shoot most.

But both are team stats, and therefore pretty pointless when you are judging individual players.

But overall i think people like corsi more because it sounds more fancy and advanced

 

[IamNotADancer]

But can't the same thing be said of Corsi and Fenwick?


m.

Well, not sure I'm the right person to ask. I know how the basic stats work but generally prefer to watch the game itself not the stats.

 

[IamNotADancer]

Not totally true, a SH goal earns a plus.

Apologies, I was a bit too broad with my statement. You are correct.

 

[Flowah]

really? That's pretty interesting
for the Leafs, I guess I'd say Hainsey has been bringing better value for us than Marleau, and his +/- was lower

Seems easy to explain.

Bad +/- player who's not bad but on a bad team gets traded/signed by a better team. The bad +/- keeps their price a little lower. The inverse can also be true. Player from a good team has inflated +/- from being on a good team and has an inflated price. Ends up not being as good as desired. Or at least not as good in value as the bad +/- player.

 

[SotasicA]

If you get too many minuses, you're on the ice too often when the other team scores.

Bobby Orr was +124 one season. That's a complete player. If you were on the other team, you did not want to be on the ice when he was on.

 

[Rants Mulliniks]

Plus minus is a good stat only if used in the right context. For example looking at McDavid's +/- is extremely impressive this season when compared to the other top scorers in the league. The majority of the others play on better teams and teams that have much better goal differentials, gives a good indication of how dominant he is on the ice with less support. Last season on the Oilers for example Adam Larsson and Oscar Klefbom played as a defensive unit almost the entire season. Larsson finished +21 and Klefbom only +7. It certainly matched the eye test as Klefbom certainly had deficiencies in his defensive game last season, not just this season. Most of the success of that defensive unit defensively was due to Larsson playing better defensively compared to Klefbom, which was certainly recognizable when watching them. Even this season Larsson is +13 in a rough season on a team with -22 goal differential. People can say what they want about the Hall trade but Larsson absolutely makes the Oilers a better team. Leads the league in hits per game and is solid most nights.

Plus minus used in the scope of the team and the player usage is still a solid stat.

That sounds like a misunderstanding of how +/- works as those teams with better goal differentials have PP goals counted in those totals. As an example, Tampa has 30 more PP goals than Edmonton. That's 30 goals towards their differential, none of which count toward player +/-.

Edit - that's just 5 on 4 goals for.

 

[RickP]

People who like to use +/- to know if a player is good defensively should REALLY look at 5-on-5 Goals Against/60 (GA/60). It is a flawed stat, just like +/-, but at least it tells you how many goals a team has allowed when player X was on the ice, per 60 minutes played.

+/- includes way too many different situations, you can't really make sense of it.

 

[lomiller1]

No such thing as luck.

Have you ever watched a hockey game? Even without the math co firming it, it’s pretty obvious luck plays a huge role.

Putting the puck in the net is not "noise". It's the purpose of the game..

Purpose of the game or not most goals are not repeatable and depending on things, like the puck bouncing just so, that are simply not things plays have much control over.

You don't have to generate as many high scoring area shots as your opposition, if you win the game without them.

Correct, ad the reason for this is that each chance is essentiality a random role of the dice. Some may be a 1 in 5 chance some may be 1 in 20 for becoming a goal. The way you overcome this is to get lots of chances.

But that's not what +/- is for. It's a tangible record of what just happened.

Read my previous posts. GF% is useful that way, +/- is not because the formula it uses s broken.


Generally speaking I don’t need a "tangible record of what just happened". I can go directly to the game score and standings and that’s all you really need to know. What’s far more interesting is the question of whether a team or player is going to be successful going forward if they continue to play as they have. The best way to answer this question (non-public xGoals models notwithstanding), is Corsi and Rel Corsi.

 

[AvroArrow]

It's a team stat, not individual player. Kessel didn't magically become some defensive specialist overnight when he got traded to Pittsburgh. Roman Polak is not better than Morgan Rielly.

It's a useless stat when comparing individual players, it's indicative of team play not players. The problem on hf is too many clowns try and use +- as a way to justify their argument that Player X > Player Y , don't believe me look at their +-

I.E. player comes on for a line change, 1 second later gets scored on and he gets the minus. What does that tell you ?

Goalie lets in a fluttery puck from 200ft away, all 5 guys get a minus, what does that tell you ?

Team allows 10 breakaways in a game, goalie saves all of them, what does that tell you ?

Your defense man turns it over on a simple play in the slot and you get scored on, all 5 guys get a minus. what does that tell you ?

Absolutely nothing, it's such a vague stat and 99% of the time used out of context by some random clown trying to strengthen his argument.

 

[tarheelhockey]

If nothing else, it's corrupted for absolutely no reason by nonsense like including SHGs and ENGs. There's simply no reason whatsoever to include those numbers, but we do anyway.

The best way I've heard it said is: if someone proposed +/- as a brand new stat in 2018, they'd be laughed out of the room. It's just an obviously bad way to think about hockey. We could just as easily create a (somewhat) better version by applying simple subtraction to information we already have at hand: ESGF - ESGA. Right off the bat, you have a better statistic than +/- and you barely had to do anything.

 

[lomiller1]

Look at this way, if we would just focus on the plus side, on both corsi and +/-. If we have 2 players, Player 1 scores 110p in 82games, while player 2, scores, 50p in 82 games. But player 2 have more often been on the ice when his teams shoot. Is corsi then a good way to measure future success?

Again, don’t try to use +/- that way because the formula is broken.

Assuming you means GF% rather then +/-, the part of the problem is that an NHL season isn’t long enough to get the sample sizes you need. Since teams get shuffled every year, players age, etc you can’t just carry over into the next year and even if you could your sample size with Corsi would remain ~15X larger.

Look at this way, if we would just focus on the plus side, on both corsi and +/-. If we have 2 players, Player 1 scores 110p in 82games, while player 2, scores, 50p in 82 games

Nearly everyone who uses CF% will also be looking at players Points per 60 min of ice time along side Rel CF% when evaluating players. Looking at one doesn't preclude looking at the other. xGoals can bridge the two, but it needs to be the right xGoals model.

 

[ImNeverWrong]

It has its benefits. If you're -30 and the rest of your team is +20 better...we know you suck.

 

[Atoyot]

It's a great stat, but like with any other stat, context matters.

If a team is -30 in goal differential on the year, and its best player is +20, that speaks volumes.

Similarly, if a team is+30 in goal differential on the year, and it's worst player is -20, that also speaks volumes.

Using the leafs (my team) as an example, the teams is +40 in goals. Matthews and JVR have a similar number of goals and points. Matthews (famous for being responsible away from the puck) is +24. JVR (famous for being irresponsible away from the puck) is +1.

It's a very useful stat.

So you have the best shutdown line in the league on the highest scoring team in the league, the line is not expected to score and will occasionally get scored on, but much less than others would playing against top players. They are -20, the team is +30. What has +/- told you in this situation?

 

[Atoyot]

+/- doesn't exist during the PP

It does if you score short handed.

 

[Flukeshot]

I think over a full season it provides good context to a players offensive and defensive capability. Definitely leans towards being a team stat and is heavily impacted by player usage i.e. is it the best D on a bad team, playing a 25 min a night = OEL.

I would be interested to know if NHL coaches/GMs/scouts value +/- at all.