I don't have the training in statistics to tell you how accurate these numbers are, per se. What I can tell you is how much data the numbers are based on.
Here is how much data we've got:
I'll give you three numbers for each training type. Each entry represents a week where a player got at least 1 minute of training in the given regimen.
Total entries, then entries where player's height is ≤6'7", and then ≥6'8"
Passing (PG) - 3358, 2926, 432
1v1 (F) - 7500, 4328, 3172
1v1 (G) - 3443, 3153, 290
JS (F) - 2406, 1319, 1087
JS (G/W) - 5086, 4529, 657
[JS(W) and JS(G) are treated together since their results were nearly identical]
Handling (PG) - 243, 208, 35
Handling (PG/SG) - 128, 121, 7
Re: Passing – So, your comment prompted me to look into this a little more. And I'll show you what we're dealing with. But I ought to say a one thing first. The Training Simulator, like the "training speed analysis" page on the BB Training Project, models the height effect linearly. But linear models may not be the whole story.
When I look at training speed data for 19yos (the age where we have the most data) for PA(PG) and group it according to height I get this (the # in parenthesis is the number of data points included):
19yo
5'9"-6'3" : .58 (284)
6'4"-6'9" : .65 (178)
6'10"-7'6" : .48 (21)
This suggests the height effect for passing PG might better be represented as a curve (with middle height players receiving the benefit) than as a line. But when we look at 18yo's and 20yo's it looks like this:
18yo
5'9"-6'3" : .62 (235)
6'4"-6'9" : .60 (127)
6'10"-7'6" : .70 (23)
20yo
5'9"-6'3" : .57 (369)
6'4"-6'9" : .54 (153)
6'10"-7'6" : .60 (25)
With 18yo's and 20yo's it looks like there might be a curve that favors the extremes instead of the middle. When the CPU weighs everything in, and assumes a linear model, the TS is what you get. If the line is actually a curve, my guess is that this model matches the short end well and overestimates bigs a little. But that is based on instinct and not data. The data supports exactly what the TS simulates.
As you can see, most people don't train bigs in Passing. So this won't effect most users. But it also gives you a sense of where our data gives us more or less certainty.
When we tested the TS against players we'd trained in the past, it was accurate to within half a pop per skill over 6.5 seasons of training for a 6'4" SF and the same for a 6'2" over 3 seasons.