Yes, your projection is likely wrong because it's statistically highly unlikely that you would accurately project Caufield's goals, assists, and points perfectly. That's a sound statistical foundational theory to adopt and one I'd stand by.
Obviously, exact projection unlikely, bigger the margin you give me the more likely it is to be within that range.
I have yet to hypothesize OHL numbers for him, despite your pleading. Yes, I think it's fair to suggest he is an elite goal scorer. That is irrespective of how many OHL goals one may hypothesize for him in a season that will never happen. What I did was provide a short list of assumptions you may be overlooking because you're the one that theorized his point totals would go down after I claimed your "software" was tin foil hat territory:
Many assumptions which you believe I am overlooking are wrong assumptions about how I function/account for or just plain disingenuous, like pointing out that it is flawed to assume that a player will have fewer points in a harder league, if that is the case one simply can't make any prediction, might as well say that from the start.
I understand why you want to move the goal posts now, but I never "clearly suggest[ed]" that his projected OHL goal totals "should be higher." I don't know how I could make it any clearer for you than in the opening two sentences of my response to your aforementioned post, which was the first inkling of any methodology to your point projections.
I never said that you suggested that.
Not unless you've successfully built regression models to unpack the relevant variables necessary to accurately predict future production of a player as they move through different leagues (i.e. NTDP -> NCAA -> NHL) and done so successfully for a multitude of possible development paths. Hence why I asked if you could predict his points next year at Wisconsin. And you said you can't. Because if you did that scale, you could reasonably and accurately predict that 72 goals in the NTDP -> X goals in the NCAA -> Y goals in the NHL, at which point you could work backwards from Y goals in the NHL -> X goals in the OHL based on your ability to accurately predict X goals in the OHL -> Y goals in the NHL. Thus, you'd be able to predict what 72 goals in the NTDP would have been in the OHL had the player been in the OHL instead. And if you've done that, congrats and well done, you just reinvented the NHL and you're going to be very rich. But if you haven't, you'll never be able to verify the accuracy of your conversions. A conversion algorithm is meaningless without a way to validate the accuracy of the conversions. Which means you either need to:
- reverse engineer those conversions by predicting the separate development paths to a shared league so you can work backwards to convert a player's totals from league a to league b and vice versa, i.e. see above.
- start getting a lot of players switching leagues (i.e. NTDP -> OHL) in the middle of the season - where the number of games played in each league in the same season is a large enough sample size for appropriate weighting - in order to accumulate data that is verifiable based on player data from both leagues in the same season.
- realize if you've already cracked another potential verification model that you don't need to wait years to accumulate data to validate your conversion algorithm; you have plenty of historical data to use to calibrate your model and you don't need to wait for future data that will be just as volatile as historical data.
I can't because the model isn't made for 19 years old players. I don't exactly know how much better a 19 years old player is than is 18 years old version of itself on average. I do know how to make the calibrated adjustment for 17 years old vs 18 years old because I have a large enough pool of data to know whether or not they arrive within a reasonable range at the same total combined. How my lists will perform will be very telling of his validity. To be of value it doesn't have to be perfect, it only has to be more precise than the imprecise way we naturally evaluate prospects, anyway it can't be perfect, like they aren't enough players switching from USDP to OHL during a season to have a significant enough sample size to do what you suggest.
- realize if you've already cracked another potential verification model that you don't need to wait years to accumulate data to validate your conversion algorithm; you have plenty of historical data to use to calibrate your model and you don't need to wait for future data that will be just as volatile as historical data.
That is valid and I do realize that, would be something I like doing if I had the time, that is how I started programming it actually. The bigger that age gap with the draft-eligibles the larger the leagues difficulty might have been altered over time though.
