Predicting the Playoffs: Why Nobody is an Expert

Ben Wendorf
May 25 2012 09:34AM

I'm going to admit my bias: I don't like experts. I mean "experts," really, the people who posit themselves as the go-to person for hockey advice. There are a lot of intelligent and/or informed people out there who can provide equally apt insight into the game and its machinations (and sometimes, its psyche), but I guess it makes for some pretty milquetoast TV to just poll them all the time when you want to make a statement.  

When it comes to the playoffs and predictions, though, people like to pump their own tires a bit. Every group of analysts at all the levels of exposure, pomp, and circumstance truck out their predictions, sometimes all together, sometimes series by series. We scrutinize and agonize the picks, but for any "expert" the assumption is that their pick has elevation above the analysis of a bloke at the bar.

I'm here to tell you that, when it comes to playoff predictions, nobody's an expert.

Monkeys Tossing Darts

Remember when that octopus was picking everything correctly?  That should have been your tip-off.  Or maybe it should have been the fact that, to this point, only one NHL team hasn't had a four-game losing streak this year (the New York Rangers, though they have lost four of five twice). And that's the worst-case scenario, of course...you can be considered successful even if your winning percentage in a 7-game series would have barely put you in the playoffs in the first place (57.14%).

So, let's say you aren't willing to trust your gut (we've yet to prove this is the case with our "experts", but whatever), and wanted to select your teams by the conference rankings. Some years, that's a really good idea, just like it's sometimes a good idea for March Madness, so rock out. If you did that to predict each of the playoff series the last four years, your record would have been 42-30...or 58%.  

Keep in mind a coin toss is 50%.

Then, you decide that the 3 seed is throwing you off, so you shift over to points percentage (similar to win percentage, but remember the Bettman point). Happy that you are reaching a higher level of analysis, your reward over the last four years was 38-29 (in five series, the points percentage was identical)...or 57%. Well what the hell?

Okay, screw points percentage, sometimes the best teams don't always win; now we're going to move to what really matters, what the game is all about: goals. I think that it's really the defence that wins the game, so I'm going to say that the difference in goals allowed between the two teams will determine the winner. They had a hot goalie this season, so surely...ah shit, 58%.

Y'know, it's offence that wins the game, goals-for differential that's right, Gretzky Gretzky Crosby Crosby and all that...59%.

Hmm...well it's about holism here, we're moving into our John Lennon, Hindu-consciousness era and must consider the offence and the defence. Goal-differential...differential. Two become one.  68%, booyah baby!  Just as you're about to order Rosetta Hindi, you check the four years before the previous four years, just in case...

53%, or a full ten percent lower than if you chose by rank during those years.  Crap.

Beaten, bruised, waking up in the gutter, you've got no place left to turn except your gut (currently turning in a different way, so scratch that) or the...gulp...stats people. They're like the mole people only mole people come up from underground every once in a while. Probably to watch the games.

They tell you to use "Fenwick close,"  which you guess is a distant Kentucky cousin of Glenn Close, and you diligently record the numbers (they use decimals, jeebus) and run the results. 56%.

Fetal position, thumb-sucking...there there. It's just the playoffs.

Future Babble

You see, there's a certain level of dissonance between what even the last approach uses and the NHL rewards. Play 30 games against the same opponent and Fenwick Close is your strongest predictor, but put those two teams in a format where a singular result (winning a series) comes from a slightly positive effort (57% winning percentage) over a small sample, and you have a recipe for prediction disasters.

It's not a complete crapshoot, you're likely slightly better than a coin flip if you know the game, but in any year an octopus can put you to shame. There's only one exception, and that's when a big-time Fenwick Close team (55+) matches up against, well, pretty much anyone. Chances are, though, that same team is a 1 seed with high goal differential and is visibly better, so everybody picks that one. Not what you'd call the expert edge.

There are greater implications at work here. For one, teams sometimes use playoff results to shell out greater contracts and fans sometimes attach great significance to "playoff performers".  If any team has a pretty good chance at winning, any skater (and especially any goalie) has a pretty good chance at looking like a quality player. At the team level, this means that your roster decisions should be focused on making the playoffs, not agonizing over playoff results. I don't give a crap if you think Roberto Luongo isn't a playoff performer, he still turned in a regular season that would make the Philadelphia Flyers jealous.

But back to my initial point, to the people that feel they are (or are being presented as) "experts": you are not an expert, none of us are, not even myself.  

I'm over it.  Deal.

E3c4fc2dfa99ac2e872f17c7bb92b731
Benjamin Wendorf was co-manager of the SB Nation Winnipeg Jets blog Arctic Ice Hockey (formerly Behind the Net); he is currently co-editor of Hockey-Graphs.com. He has been writing about NHL analytics for six years. He can be reached via email at wendorf DOT benjamin AT gmail DOT com, and tweets from @BenjaminWendorf.