Boys on the Bus

Fixing the 2019-20 NHL Play-ins

Posted on by Michael Parkatti

Last Friday I published a thread on Twitter decrying the NHL’s proposed Play-in 24-team format for the 2019-20 Stanley Cup playoffs:

In summary, the chosen format punishes some teams (EDM and PIT in particular) while unduly rewarding others (cough, CHI, cough, MTL). Approximately 70 genuine games of an NHL season were played. Players were drafted, developed, signed, trained, coached, and finally executed their way into those results. It seems unfortunate that the NHL is trying to ignore those very instructive results by choosing such obvious winners and losers with this new format.

Why do I care? Well, I am an Oilers fan. We’ve made the post-season once in the last 13 seasons and I’d dearly like them to make it twice in 14. With their performance this season (90%+ odds of making the playoffs by all accounts), I’m a fan who very acutely understands how precious these seasons of actual ‘goodness’ are.

So, what’s the remedy here? In the thread, I suggested that the NHL explore alternative formats to their (ahem) alternative formats. Namely, instead of making this Play-in round to be best 3 of 5 series’, why not explore changing the number of wins necessary by each team to better match their pre-COVID odds? This way, the league can give the longer-odds teams their shot of making the playoffs while respecting each of these teams’ approximate chances before the shutdown.

In this post, I’ll be giving specific suggestions on how this could be accomplished.

Setting the odds

Firstly, we’ll need some reasonable basis to set the pre-COVID playoff odds of each of the 5-12 seeds. I’ve found 6 different Google-PageRank-worthy websites who published daily chances of making the playoffs right up until March 12 (when the season was suspended): SportclubStats, Hockey Reference, MoneyPuck, the Athletic, PlayoffStatus, and PowerRankingsGuru. Here’s a summary of what I found:

The table is sorted by Conference by NHL playoff seeding. The average of the 6 sources is available in the far right column. The sources mostly agree with surprising precision: only 3 teams had different sources put them both above and below 50% probability (WPG, MIN, and NYI). The sum of all 16 teams’ average probability is 8.06 — pretty darn close to the 8 teams that will actually walk out of this Play-in to the Playoffs.

My proposal is that we use these average probabilities as the most accurate reflection of these teams’ expected chances of making the playoffs pre-COVID. This should be relatively non-controversial.

Potential Series Formats

Let’s assume that the NHL is in love with the idea of a 5-game series. Let’s explore the options you have to give the higher-seeded team a higher chance of winning the series than 50/50:

  1. Play 5 games, but the higher seed needs to win fewer games than 3 to win the series and the lower seed needs to win more than 3 games to win the series. The two new possibilities here are: a) high seed needs to win 1 in 5 / low seed needs to win 5 in 5, and b) win 2 in 5 high seed / win 4 in 5 low seed.
  2. Spot the higher seeded team one win automatically and play up to 4 additional games. There are again two options here: a) win 1 in 4 high seed / win 4 in 4 low seed, and b) win 2 in 4 high seed / win 3 in 4 low seed.
  3. Spot the higher seeded team two wins automatically and play up to 3 additional games. Here there is only one option that actually ‘helps’ the high seed: win 1 in 3 high seed / win 3 in 3 low seed.

I’ll be using the trusty binomial distribution here to help us answer what the probability of the high seed and low seed winning these various new formats is. It’s great for situations where you have a number of trials (read: number of games) and want to know what the probability of observing a number of successes within those trials is (number of wins in those number games). For simplicity, I’m going to assume the high seed has a 50% shot of winning any given game. Don’t concentrate too much on this probability as changing this doesn’t significantly alter my conclusions.

Ok, so assuming the higher seed has a 50% chance of winning any single game, what’s the probability the higher seed wins these various suggested series formats?

The conventional “win 3 in 5” approach is highlighted — of course, if both teams are equally likely to win any single game they are equally likely to win a 3 in 5 series. But you can see the influence these alternative formats start to have in terms of win probabilities — requiring the high seed to only win 2 in 5 games gives them an 81% probability of winning the series, while requiring them to win only 1 in 5 games gives them a 97% probability of winning the series.

You can likely see where I’m going here — whichever series format best fits the pre-COVID odds for the high and low seeds in any given series also best replicates the chances those teams actually had of making the playoffs when play was suspended. The NHL is already doing this to some degree — the top 4 seeds in each Conference mostly had odds right near 100% of making the playoffs. Why even make these teams play games? The bottom 7 teams in the league had odds approaching 0% — what’s the point of making them play games? In both cases the outcomes chosen by the NHL conform closely to those teams’ playoffs chances. All I’m suggesting here is that the league extend this courtesy to the teams whose playoff chances were somewhere between 0% and 100%.

Applying this to EDM-CHI

Next we’ll come up with a way to decide which series format best fits a specific proposed match-up. I’ll use the Edmonton-Chicago 5 vs 12 seed match-up in the Western Conference to illustrate this initially.

First, we’ll take the sum of the squared errors between each series format and the two teams’ pre-COVID chances. Second, we’ll then select the series format the minimizes this sum of squared errors (or, we’ll choose the series format whose chances most closely resemble the pre-COVID playoff chances).

We can see from the above table that the ‘1 in 4’ series format best suits the Edmonton-Chicago series. Edmonton’s pre-COVID odds of making the playoffs was 94% while Chicago’s was 3%. The ‘1 in 4’ series format would give Edmonton a 94% chance of winning the series while giving Chicago a 6% chance of winning the series (and thus making the playoffs). Intuitively, matching these two sets of chances as close as possible is the fairest way of conducting this Play-in series.

What does a 1-in-4 series actually mean? You’d start by giving Edmonton a 1-0 lead in the 5 game series. Then, you’d set it so that Edmonton only has to win one more game before ‘winning’ the series, while Chicago would have to win all 4 of the 4 remaining games to ‘win’ the series. If this sounds like a tall order for Chicago, just remember that on March 12 they had a 1 in 30-ish shot of making the playoffs while Edmonton had a 19 in 20-ish shot of making the playoffs. Winning this series should be a tall order for Chicago!

Recommendation for all Play-in series

Now that you’ve seen this to one of the eight proposed Play-in series, let’s extend this exercise to all 8 series. After the chips fall, this is what I would recommend:

Western Conference

Eastern Conference

Conclusion & Notes

Hopefully this paints a useful picture on how we can accomplish the NHL’s business imperative of including large-market longshot teams in this exercise while also respecting the pre-COVID chances of the 5-12 seeds in each conference.

The above assumes the high seed has a 50% chance of winning any game. This might seem low to you. Let me assure you — even if we increase these odds for the higher- seed team to win a given game, it will have a marginal impact on the series format recommended. For instance, assuming the higher seed has a 55% chance of winning a given game only changes one series format recommendation (CAR-NYR moves from a 2 in 5 format to a 2 in 4 format for CAR to win, for instance).

Uncategorized