I feel like you're jumping back and forth on samples a bit. My contention is simply that a player-season is a player-season, whether it happened in 1924-25 or 2024-25. It seems like I'm promoting single season comparisons, but that's mainly because I'm trying to show that patterns in scoring show up again and again. The considerable differences in point totals and PPGs is mainly due to an increasing number of games played and the seasonal scoring levels. When you acknowledge that league average scoring of 210 in a 70 game season is the same scoring level at 246 in an 82 game season, any point total differences are completely down to the 12 extra games played (plus a good dose of randomness).
Like, look at the Bill Cook season of 29-30 - 29+30=59 points in 44 games, 21.3 G%, 43.4 P%. You have the whole offside rule change helping to double league scoring, they didn't even track even-strength and power play goals, and the Rangers ran 6 forwards and 4 defensemen. Seventy six years later, Vincent Lecavalier puts up 52+56=108 points in 82 games, 21.4 G%, 44.4 P%. It's the exact same season. In 24-25 scoring, Cook's year is 55+57=112, Lecavalier's year is 54+59=113, minor variations easily attributable to randomness. [In 29-30 scoring, Lecavalier's year rounds to 29+31=60, Cook's year rounds to 53+54=107 in 06-07 scoring.]
That's really all I'm doing when I convert seasons. I'm assuming that the Bill Cook that put up a 21.3 G% and 43.4 P% in 29-30 is talented enough to put up a 21.3 G% and 43.4 P% in 06-07, and vice versa for Lecavalier. Because scoring doesn't change, similar seasons from widely disparate eras in the NHL have the same underlying numbers.
I think it depends on what you mean by volatility. Does volatility go up in a 24 game or 44 game season? On one hand, smaller sample sizes are subject to more random variation. On the other hand, it is really hard to develop a substantial gap in that few games. Also, because you're not rolling 4 lines and 3 defensive pairs in 1920, and the same players are on ice for 50 minutes of the game, there's no other players to have that randomness occur to, so the same players were getting points no matter what.
Another way to measure volatility is the gap between VsX and Average VsX for a season, and whether it is above or below (or equal). If that number is above 1, that means VsX was set too low for the scoring level, if it is below 1, that means VsX was set too high for the scoring level, and I considered 1.00/1.01/0.99 as correct for the scoring level.
In the first 25 years of the NHL, it was above 1 22 times, and below 1 3 times. VsX was correct twice, within 5% 7 times, within 10% 13 times, within 20% 18 times, and was above 20% the other 7 seasons, peaking at 29% off in 40-41, where VsX was set at 44, but should have been 56.55 (league average was 129 in a 48 game season - compare that to 12-13 where league average was 127 in a 48 game season, and VsX was 57 - should have been 55.67).
The second 25 years of the NHL covers the O6 era prior to expansion from 42-43 to 66-67. The only time it was below 1 was the artificially set season of 43-44, and it was only within 5% twice. It was within 10% 11 times, 20% 15 times, and actually above 30% 3 times - 46-47 VsX of 63, should've been 83.29, 48-49 VsX 54, should've been 71.45, 66-67 VsX 70, should've been 91.61.
The next section is the 12 years post-expansion prior to the WHA merger. It was 9-3 in favor of being higher than 1. Also, VsX was correct twice, within 5% 3 times total, within 10% 8 times, and never above 20%, though it did peak at that in 73-74 (VsX set at 91, should've been 109.15).
Beyond that, you have the 25 years between the WHA merger and the 04-05 lockout. It was all the way down to 14-11 in terms of being above or below 1, though you had 1 season at 1.01, 2 at 1.00 and 3 at 0.99, so it is more like 11-8-6. Along with those 6 correct seasons, you had 15 total years within 5%, and 22 years within 10%, with the outliers being 83-84 at 14% (VsX of 121, should've been 138.52), 86-87 at 19% (VsX of 108, should've been 128.88), and 98-99 at 12% (VsX of 107, should've been 94.68).
Finally, you have the 20 years post-lockout. We've had 8 years above 1, and 12 below, though we have 3 exact years, 1 at 1.00 and 2 at 0.99, so 7-10-3 as a final tally. To go with those exact years, we have 8 total within 5%, and 19 of the 20 within 10%, with the peak year being 11% in 14-15 (VsX of 86, should've been 95.56).
The entire premise behind Average VsX is that the 05-06 through 18-19 seasons is a representative sample of scoring talent. I've toyed with adding more seasons, but I'm afraid it would be a bunch of work for less accuracy.
I don't know if that was a great exploration of volatility, but it was an explanation. Basically, I'd agree with you that volatility has decreased with time, but that the limited rosters and fewer games played mitigated it to an extent in the earliest years of the NHL. The O6 era lost the limited rosters and added more games, plus you had the disruption of WW2. The expansion era generated volatility from the gap between the haves and have-nots, and the lack of parity. The results after the WHA merger show that 20 teams and 80 games pretty much mitigates volatility, and 30 teams even more so.
