The Leafs need a good backup goalie.

Did you say yes to that? You’re not alone, as it might be the only thing Leafs fans agree on, but the conversation in hockey is changing from “you need a good backup”, to “tandems are the future”

A team can roll along not caring who the backup is as long as they’re busy rebuilding, but when you move from playoff team to wannabe contender you need to hit some minimum level of value at that position. The ideal situation is to have a young prospect who is a lock to be a starter someday that you can bring along at just the right time to play better than your usual run of backups on balance. If you’re lucky, they can play 30 games or more. The history of new starters coming into the NHL says that most teams aren’t going to have that ideal, and they’ll be relying on a backup who is always going to be a backup, who isn’t young, and who has a history in the NHL.

But right now there is a lot of talk that the future is going to be extremely different to the present. Chris Johnston recently talked goalies in his Beyond the Headlines segment, which is what they said on HNIC only in an article.

Consider this a luxury most NHL teams don’t have: The Boston Bruins can literally roll out either of their goaltenders on any given night and expect to get an above-average performance.

Not only do the Bruins have the NHL’s best team save percentage at .934, they’ve done it with this sequence of starters through 13 games: Rask-Halak-Rask-Halak-Rask-Halak-Rask-Halak-Rask-Rask-Halak-Rask-Rask.

This is the future.

Gone are the days when a goaltender could make 70 (or even 60) regular-season starts and still be counted on to perform in the playoffs. Rask was the biggest reason Boston reached Game 7 of the Stanley Cup Final last spring and would have claimed the Conn Smythe Trophy if they knocked off St. Louis that night.

That came after a season where he started just 46 games — a season where both he and Halak won more than 20 times apiece for Boston.

The more extreme version of this vision of the future is that teams should just go get a cheap pair of goalies to run as a tandem. That story goes like this: All goalie performance fluctuates, and there’s lots of examples where teams have mistaken a hot backup run for a guy who should be a starter and overpaid him. So, instead of falling for a darling of a season, just go get a tandem of goalies no one else has snapped up. This idea rests on the belief that there are a lot of above average goalies at all times and they are easy to identify and acquire.  I am skeptical of all of that. So it seems is Johnston himself to a point, or he wouldn’t have called the (not cheap) Boston Tandem both “a luxury” and “the future”.

How many good goalies are there, really?

To test that skepticism, I looked at the six seasons since the lockout year to see what the distribution of goalie quality really is. I included every season for a goalie who saw at least  300 Fenwick Against (unblocked shots against), which weeds out the five-game tryouts, but includes all starters, regular or semi-regular backups, and injury-shortened seasons.

I measured by Fenwick Save Percentage over Expected (all-situations from Evolving Hockey) which gives a truer picture of the goalie’s actual performance than just raw Save % on Shots on Goal by accounting for the quality and the true quantity of the shots the team allows.

Understanding the chart: each boxplot is one season and the box portion shows half of the goalies in that season, with the thick line marking the median point, so half are above that and half below, 25% are above the box and 25% below it. The top line is the maximum, and the bottom is the minimum of the default range that r sets for constructing these plots. The circles are the outlier results, and the zero line tells you where a goalie who was league average by the expected goals model would fall.

One notable thing is that last season was not a good year for goalies. It was a good year for shooters, and they outperformed expected goals no matter which model you looked at. Any Expected Goals model is created using years of data, more than I’m showing on this chart, and by using thousands of shots to figure out how to weight each type for its chance of becoming a goal, these year-by-year variations get smoothed out.

There are just over 70 goalies per season that made the cut in each year on that chart, and while there are variations from year to year, the basic structure of the distribution is the same. Most of the time, less than half of the goalies are above the Expected rate of an average goalie. Overall in my sample, that figure is 45%. The top 25%, or about 18 goalies per year, have seasons that range from good to excellent. Most years, there is a large number of goalies, nearly the full middle shown by the box on the plot, who are around the average mark and so cover the range of poor starter to poor backup. The rest are terrible and shouldn’t have been playing.

This makes sense. Teams play their best goalie more, so we should expect to see fewer than half of the goalies who have played in a given year, playing most of the minutes.

This reality makes it harder to believe that a team can just go get a backup that’s better than average whenever they want one. It calls this “tandems are the future” idea into even more serious question. Those ideas assumes that the non-elite goalies are spread out in a nice even way towards average, and that there are a host of backups who can be promoted to a 30-40 game goalie. In reality, they actually tend to clump up around the below average territory.

How many good backups are there?

I took the same set of data and cut it off at 30 games played to see how the actual goalies, who were played like backups, performed:

So, about last year...

This is very stark. Most of the time, the backups that are in the NHL are not very good. There’s a few goalies playing very well in a few games, but some of those are injured starters, some others are prospects, and the overwhelming majority of backups are really not even average. This isn’t news, it is well understood that for goalies, replacement level is quite a bit worse than average most of the time.

The other thing to notice is that the range of how bad a backup can be is much larger than the range of how good. This is just part of the deal with being a goalie. The distance from average to perfect is shorter than the vast chasm of horrible that is open on the other side of the zero line.

Tandem dreaming

You can get a tandem of above average goalies like the Bruins have achieved. The New York Islanders got it last year. But it’s hard to try to do it on purpose. The Buffalo Sabres seemed to want that, and they did not get that:

Remember the total number of goalies above average enough to give you 40 starts each is well below enough for even half the teams in the NHL to go this route and succeed. Most teams will have a skill imbalance, and they absolutely will play the better goalie a lot more than the poorer one. They’d be foolish not to.

Backup shopping is hard on a budget

Forget searching for the luxury of a tandem for a few years while Frederik Andersen is a high quality starter. The job of just finding a good backup is difficult enough.

Compounding the issue for the Maple Leafs this season is that there isn’t a lot of wiggle room in the cap space to pay more than Michael Hutchinson’s $700,000. It wouldn’t be impossible to pay $925,000 for a goalie, but it would be difficult. If the Leafs really want to run a 20-man roster all season, they can get to over a million by a little.

This is the full list of goalies who have played at least one game this season and who have a cap hit of $1.5 million or less:

The method by which most backup goalies change teams is either as a claim off of waivers or via a trade right after they’ve cleared waivers. Ideally, the Leafs would like to have someone waiting in the wings in the AHL to swap places with Hutchinson and to play some of the games if Hutchinson doesn’t work out. This is absolutely a case of trying to find the goalie who is on an hot streak and riding him until he gets cold.

Riding the Hot Hand Fallacy to the bank

That sounds like a bad way to run a hockey team, but it isn’t. It’s just realistic. Don’t try this with your starter — glances at the Marlies playoffs last year, and the year before, and... — but do it with your backup. Teams have goalie coaches, scouts and analysts who can look at all the available players, their own goalies, and make decisions about who to try in net, and some teams are better than others at it, but no one has a crystal ball to tell them how any goalie is going to perform in 10-30 games in a given season.

Two years ago in the playoffs, we all got a demonstration of how big the range of goalie performance can be, and how desire alone isn’t enough to get you a great result when you need it.

Both Frederik Andersen and Tuukka Rask were pulled in a game, neither played well, both were inconsistent, and each only had a couple of games befitting NHL starters, and yet all of their performances fit in their individual historical records and were not unprecedented. The really horrible Andersen game was just barely not his worst ever, but absolutely none of those results were more than just very unlikely.

For a backup playing once in a while, the range of likely outcomes is much greater than it is for Rask and Andersen. Backups wouldn’t be backups if they were more consistent. If you look at Garret Sparks’ performances last year, you see a wide range of results. From game to game, it was impossible to know where he’d fall on that continuum.

I’ve switched to Goals Saved Above Expected here, because it’s easier to get for a single season, but the concept is similar. Looking at that range of results, which isn’t good on average, it’s hard to guess what Sparks would produce in another season of games. His historical record is sparse, but doesn’t make him look better.

Hutchinson fits neatly within Sparks’ performance range last year, and in fact, he fits within every goalie performance ever with this set of results this season. His historical record, which is fairly substantial, says he’s more likely to improve than get worse. So it’s not an absurd idea for the Leafs to simply keep throwing him in and instructing the rest of the team to actually play a little hockey in front of him.

I noticed in putting this together, that Hutchinson is being tanked by his PK results. His even-strength play has averaged out closer to Andersen’s so far, although Andersen’s stats are trending up quickly.

Garret Sparks had a soft honeymoon period as the Leafs backup because the team happened to play much better than usual in front of him for his first run of games. And then they reverted to normal, and the wheels came off the Sparkmobile. Hutchinson is more likely to get the opposite scenario where the team stops being particularly dreadful when he’s in net.

Keep trying until you get it right

But exactly how Hutchinson is doing right now is not important.

Most teams play three goalies in a season, and the Leafs should be looking now for a goalie who, with a favourable roll of the dice, can be better than replacement level and even rise above league average backup for a stretch. Even if they plan to keep playing Hutchinson, assuming they’re more likely to see better results than not, they should still be looking.

They tried Michal Neuvirth, and that didn’t work, but they shouldn’t quit, even if the cap situation makes this an extra difficult problem. Kyle Dubas is meant to be an elite-level problem solver, not an average one. Time for a clutch game, Kyle. Sometimes you just need a General Manager to find a player when you need one. Just don’t expect a starter in sheep’s clothing, if he does find someone else to try. The tandem dream is a luxury only a few teams get, but there’s a large gap between a starter playing 70 games and what the Bruins have going. The Leafs can comfortably fit in there somewhere.