Ritch Winter (@HockeyAgentDad) has joined Twitter and is talking statistics. We’ve had hints before that Winter is a guy with an interest in stats before – the comments to this post about Winter’s interest in stats are worth reviewing.
It turns out that Winter’s quite the analytics guy. Now, you never entirely know when guys are blowing smoke but he’s pretty adamant about the value of hockey analytics.
I don’t quite recall if Marian Hossa had a no-trade clause, although this story suggests not. If he didn’t, I’m not entirely certain how Winter used analytics to put him in Pittsburgh and a Stanley Cup final – my recollection is that he almost ended up in Montreal and that Pierre McGuire wept on the air when he didn’t. Or something like that. As for the hockey analytics and Hasek business, Detroit went 49-20-9-4 for 111 points in the year before they had Hasek, with a goal difference of +51. The goaltending was split between Chris Osgood (.903 in 52 GP) and Manny Legace (.920 in 39 GP). A particularly dull monkey could have foreseen that Detroit was a good place to be.
I have a bit of an aversion to guys who overstate their case, regardless of field. When I happen to have some knowledge of a field and catch a politician or journalist overselling something, it makes me wonder about why I should rely on him when it comes to areas in which I don’t have knowledge of the subject matter. I had a bit of an exchange with Winter about the value of defence independent goalie ratings and he kind of came off as a bit of a touchy guy who doesn’t know his Shakespeare but let’s leave that aside. This was what first caught my eye from him:
A claim like that is always going to get my attention. Frequent commenter @dawgbone98 went after Winter a bit on the point and they had an exchange. Winter clearly believes that there’s something magical about the first twenty games of a season.
An aside: it really only makes sense to deal with the post-lockout era in evaluating this claim. The points structure before the lockout was such that it created different incentives for teams. In particular, the incentive to play for OT is much stronger now, as it guarantees you one point and a 50/50 shot at a second.
Post lockout, Winter’s point is clearly wrong: of the 180 team seasons since the lockout, there have been 56 which saw the team start below .500. Twelve of those saw the team make the playoffs anyway. The 2005-06 Ducks and Lighting were sub-.500 after twenty. The 2006-07 Senators and Canucks were sub-.500. The 2007-08 Capitals, Devils, Penguins and Flames were sub-.500. The 2009-10 Candiens were sub-.500. The 2010-11 Sabres, Rangers and Blackhawks were sub-.500.
Going back isn’t going to make Winter’s case any better. The pre-lockout NHL had fewer points, with even fewer points before 1999 and there was less competition for playoff spots. You don’t make the playoffs with 52 points anymore, 1987-88 Toronto Maple Leafs. The 3% number is a mystery to me – I’d love to know how it’s being calculated.
Bonus tweet from Winter that came after I chided young Jonathan Willis for what I perceived to be a failure to tell truth to power about the dodginess of DIGR:
So let’s see what else the data tells us. Is the first 20 games more tightly correlated with final points than other twenty game stretches post-lockout? No. For that, we look to the famed G29-G48 stretch, which has a correlation coefficient of with points total .72, easily trumping the .66 of the first twenty games.
As it turns out, twenty game stretches of sub-.500 hockey are actually pretty common for playoff teams – of the 96 post-lockout playoff teams, 65 of them have had a twenty game stretch of sub-.500 hockey at some point during the season. If you think through this, you realize that a team has 63 twenty game stretches – 1-20, 2-21, 3-22 etc. As it turns out, the odds of a playoff team having a sub-.500 stretch at some point are high. The graph that follows shows the number of playoff teams playing sub-.500 in their previous twenty games for each of those 63 game segments post-lockout.
So “1″ refers to 1-20, 2 to 2-21, etc. The average is 10.7. So, in fact, on average, more playoff teams go sub-.500 in the first twenty games than they do over the season as a whole. There’s something unusual that happens at the end of the year – playoff teams are far less likely to be sub-.500 over the twenty game stretches that end in the last twenty games of the year. I suspect that this has something to do with teams playing for single points – in the first twenty games, the average pts/game is 2.19 and over the final twenty, it’s 2.25. It would probably be worth checking to see if playoff teams are more likely to play OT games – my money would be on “yes.” If you recalculate it and exclude the final 14 games, you come up with an average of 12.04 playoff teams going sub-.500 in a given twenty game stretch.
Since the lockout, twelve teams have made the playoffs despite being sub-.500 in their first twenty games. Not a lot there to suggest that the first twenty are special.
What if we phrase the question a different way? If you go sub-.500 in a given stretch of twenty games, what are the odds that you’re a playoff team?
Again, this graph looks similar to the one above. 21.4% of teams that were sub-.500 after the first twenty games made the playoffs between 2005-2011. The average for each stretch of games was 20.5% of sub-.500 teams making the playoffs. You were marginally more likely to make the playoffs with a sub-.500 stretch in your first twenty games than the average.
The first twenty games seem determinative to us because twenty games is usually enough to have a decent idea about a team. Odds are, if you’re sub-.500 after twenty games, you aren’t very good. What cannot be supported though is the idea that this is because the first twenty games, as opposed to twenty games at any point in the season are special. Are Calgary, the Islanders, Blue Jackets and Hurricanes finished? Probably. But it’s because they’re bad, not because they haven’t properly defined their season in the first twenty games. They’re just bad.