Ratings Recalculation - Tested vs. Real Stats?

[Charlie Hough]

I'm wondering if any of the folks who do hardcore testing and data crunching can tell me if they've tested the ratings recalculations and their effects on statistical accuracy and realism.

For example, does using the one-year recalculation generally make player statistics more accurate from year-to-year compared to real life? And I'm talking about many simulations, each over multiple decades, with comparisons to see standard deviations.

Or does the three-year method produce better results when combined with adjustments and weakening for players with low totals for at-bats and innings pitched?

Here's my thought: I like the three-year method because it allows some flexibility and seems to make sense when you consider the varied circumstances that cause players to have an unusually good or bad year. Why unjustly punish a player's abilities or perhaps over-reward him because of one fluke season? By the same token, however, if you don't punish or reward those ratings, then how can the player actually have a year as good or bad as in real life? Doesn't it take away most of the possibility in these cases of unusually good or bad years?

Also, it seems to me that rookies and prospects become too good too soon because they often benefit from the third year in the ratings calculation, which is often their breakthrough season. So I would expect the three-year method to result in players exceeding their real life statistical output, especially in their early years. And this could drive veterans out of the lineup too early, thus leading to problems with older players as well.

If down years are smoothed out and young players develop good ratings early on, then I would expect statistics to be skewed over many simulations.

But I haven't tested it on that scale, so I'm interested to find out if anyone else has.

[PhillieFever]

Originally Posted by Charlie Hough

I'm wondering if any of the folks who do hardcore testing and data crunching can tell me if they've tested the ratings recalculations and their effects on statistical accuracy and realism.

For example, does using the one-year recalculation generally make player statistics more accurate from year-to-year compared to real life? And I'm talking about many simulations, each over multiple decades, with comparisons to see standard deviations.

Or does the three-year method produce better results when combined with adjustments and weakening for players with low totals for at-bats and innings pitched?

Here's my thought: I like the three-year method because it allows some flexibility and seems to make sense when you consider the varied circumstances that cause players to have an unusually good or bad year. Why unjustly punish a player's abilities or perhaps over-reward him because of one fluke season? By the same token, however, if you don't punish or reward those ratings, then how can the player actually have a year as good or bad as in real life? Doesn't it take away most of the possibility in these cases of unusually good or bad years?

Also, it seems to me that rookies and prospects become too good too soon because they often benefit from the third year in the ratings calculation, which is often their breakthrough season. So I would expect the three-year method to result in players exceeding their real life statistical output, especially in their early years. And this could drive veterans out of the lineup too early, thus leading to problems with older players as well.

If down years are smoothed out and young players develop good ratings early on, then I would expect statistics to be skewed over many simulations.

But I haven't tested it on that scale, so I'm interested to find out if anyone else has.

Personally,I've found that using a 3 year recalc with the current year's stats doubled gives you the best results in historical simulations.The reason I like to double the current year's stats is because of the issue you've brought up where first year players get too good too quickly,this really doesn't happen when you do this.Testing wise,I've run at least 50 or 60 100 year sims using various settings and I've found these to be the best when playing historically.

3 year recalc using real stats
current year's stats doubled
retire according to history on
played development off
injuries and trades on very low up until the 60's,and low from there on out.
Fatigue on high.
AI eval is set at 80 ratings 10 this year's stats,8 last year's,2 2 years ago
everything else default.