Many analysts have tried to project players based on their past performance, historical trends, or past performance of similar players. PECOTA, made famous by sort-of-data-journalist Nate Silver, is the most well-known example of the latter. It goes like this: so-and-so has X number of similar players through age 25, here is what they did at and after 26, so that is how we think so-and-so will perform.
It all sounds sensible except for one thing: how do you determine who is “similar” to someone? Well, there’s nothing more similar than having the same first name.
In October and again in January I published lists of junior hockey teams that were higher in my rankings than the CHL top 10, and vice versa. With the season complete we can see which set of rankings, if any, was better at predicting the final outcome. (I am a couple of months late on this, that is true.)
“Rounds” refers to number of playoff rounds won. League champions would be 4.
Higher in my rankings:
Average number of points / rounds won: 95 / 2.5. All three conference champions were in this group.
Higher in Top 10:
Average number of points / rounds won: 77 / 0.6.
We are talking about just five and six teams here; even so, 18 points in the standings over 68 games and advancing through two additional playoff rounds is a pretty big difference.
It’s fair to suggest that the other rankings aren’t meant to be predictive. But they are supposedly identifying the top 10 teams in the CHL, and yet aren’t doing that in any way that is tied to future success. So my question is therefore: what is the purpose of the CHL top 10?
Inspired in part by this, here are how the Jeff Skinner Rankings would have drafted for Vancouver in this year’s draft, compared to their actual selections. Whoever is the highest remaining player in the Skinners is taken at Vancouver’s spot.
1-6. Actual selection: Jake Virtanen. JSR selection: Nikolaj Ehlers.
Both were top-15 in the Central Scouting rankings. Virtanen was 6th among remaining draft-eligible players in the Skinners. Ehlers would go three spots later so this seems to be a fairly close selection.
Assorted facts and factoids from going through the data…
- Liberal support dropped by 2000 or more votes in nine ridings. SW Ontario accounted for eight (!) of them. It went up by 2000 or more in 59 ridings and just six of them were SW. The region was, as always, not representative of the province.
- The top 3 closest 3-way races, and 6 of the top 10, were in SW Ont. The three closest battles not involving the Liberals (so, PC vs. NDP) were in Sarnia, Chatham, and London. That probably means something.
- Liberals had 50%+ of the vote in 24 ridings. NDP in 9. PCs: only 4, all rural. They somehow lost Halton, which I never expected to see. The NDP gained votes in 69 ridings, the PCs 56, and the Liberals in 84 (!). Biggest Liberal drops were, you guessed it, all in SW Ontario.
- Skip a debate at your own risk: the worst performance by a PC candidate was in Thunder Bay–Superior North, 7.2%. Then again, maybe Tim Hudak skipped it because he knew it was a 7% riding, even if it was 17.5% last time.
- Average performance of repeat candidates (in terms of votes added, not a percentage-point change): LIB +19%, NDP +11%, PC -1%. Then there’s GRN +88%. Hmmmm.
- That last stat also shows us that another overlooked element was how poorly Andrea Horwath did in her own riding. NDP repeat candidates had 11% more votes than in 2011 on average, but Horwath lost 9%. Not a great performance from a party leader, especially one who has a long history representing downtown Hamilton. The voters could have assumed she wasn’t going to run again in 2018 (or whenever) and were trying to signal that by supporting other candidates, but I don’t want to read too much into a few data points.
- A vote in Timmins was worth 3.5 times more than one in Markham. Reminder that votes are not equal.
In response to much media speculation, it is with a heavy heart that I must advise I will not seek the leadership of the Progressive Conservative Party of Ontario.