It's funny how you point to an element that discredits this "scientific analysis" and fluff it off as an adjustment. At the beginning of the season, they rated the Habs as bottom feeders. They tell you as much. To date, however, the Habs have played as a playoff bubble team and it shows in the standings. So how pucking accurate is this "analysis"? To date, they've shown their analysis is useless when it comes to the Habs. They start with an inaccurate assumption and make calculations and predictions with that assumption. Is that what you did with your thesis? And that's just the Habs. How accurate are they with the other 31 teams? Which compounds their inaccuracy exponentially.
Scientific analysis requires you to keep an open mind and pivot as new evidence becomes available that may prove your original thesis wrong. As an example in inorganic chemistry half the “laws” literally only apply in specific cases which suggests that we don’t have all the information yet. Same reason we haven’t been able to rectify the serious flaws that exist between quantum mechanics and gravity or kinetics. all it suggests is that we don’t have all the information yet. So yes, all thesis start out with assumptions some right some wrong, that’s literally what a thesis is you go about trying to prove it and if the evidence says otherwise you adjust.
Their strength of schedule analysis has stated the “guru” ranks were from the start of the season. If they don’t plan to update their model between now and the end of the season then they need to maintain it for a full season and see how inaccurate it is only at the end. This is the only visible flaw likely in the model because it doesn’t account for any changes to regular seasons performance. However on the flip side it also doesn’t allow for a bias in the data as “gurus don’t all agree on midseason ranking drop dates. Which is how I understand it as I look at it. However if you go to points thet have a “preseason projection” and a current projection. Which suggests there’s some level of correcting going on.
we also don’t have all of the information on their model because the model is there properitary information. Sharing more would allow others to copy without putting in the work. If I was in there shoes I’d reveal as little as possible. With that said I’d definitely have changing variables as time goes on and there weighted value as a function of time through the season.
Sample basic calc
Exp Pts pct(EPP) =((pts pct by guru rank)(ft) + actual pts pct(1-ft))
Where in the above the guru rank avg then takes the avg of that finisher over a 5 year time look back.
All weighted to reflect time in season, in my head I’d likely go with
first 10 games 100%,
72-50 75%,
50-25, 25%
<25 0%.
So that as the season goes on the results are worth more and the guru rank is worth less but best way to determine that and which gurus to use is back calculate and use regression based on which analysts are most accurate.
Then the model turns into a long string of text but it would essentially boil down to.
Total pts = ((EPP habs)+(1-EPP sens))x avg pts per game. W
In the above this would be strung out for a full 82 games but the result would be if the habs and sens are both expected to be 0.500 teams then the model spits back a 1. it’s a toss up game so the habs should expect to get either a loss or a win. However let’s say the 0.500 habs are playing 0.700 devils. Now that number is 0.8, less then 1 which means a loss is more likely Where it gets complicated is NHL avg pts, because it introduces a third pt only sometimes into the equation to get final points for the year you’d land take all these integers and multiply them by avg nhl pts per game I.e. total nhl pts / total nhl games
Final result is a pts projection for the season where the QOC is simply the average of the teams remaining to be played.
