First of what will be a few posts on this. I ported the Mangame 21 DB into PCM22, I used the 21 season because I didn't want to be testing races that haven't happened yet. If anyone wants the DB let me know.
I played some with the US Pro Cycling Challenge first. The good news is the fourth best sprinter problem is gone. The bad news is I am a little worried that the problems with leadouts are worse. Going to look at this more.
But since it simpler I decided to focus on the impact of the new RDC variable. I used TTs to test this because they are easy to test and I am guessing that the impact of RDC changes in a TT are a decent proxy for the broader impact.
In the USA PCC I got early good signs. The correlation coefficient between my best guess at the stats impacting the prologue and the actual results increased from .80ish to .92 for the 2021 6k prologue from PCM20 to PCM22 (RDC = 2). When I tested the shorter prologue from this season, which is pretty much a pure PRL test, I got correlations in the .95 range with the prologue stat.
But wanted to formalize this and also isolate the impact of moving versions from the impact of limiting RDC = 2 in PCM22. Because the latter is a proxy for the impact across all races.
So I ran the Celtic Chrono 12 times (each test = 1 stage). 4 x in PCM20 (first 2 were actual race from last year), 4 x in PCM22, RDC = 5 and 4 x in PCM22 RDC =2.
I am not a statistician so hopefully there are no statical fallacies here. Table below is the correlations and the correlation squared (=r2) between my guess of the stats that drive the results and the actual results. I am using a blend of TT and RS for this, which generated the highest correlations - the share of RS is a little higher in PCM22. The first row is the average correlation across each test between stats and results. The second row applies the correlation to the average rank across two tests and the third calculates the correlation between stats and the average results across all 4 tests. So the first row represents how the game is played except for TT classics where we do 2 runs (which probably isn't needed in PCM22)
Version
PCM20
PCM22
PCM22
RDC
NA
5
2
Single Test Rank Ave Correlation
0.8242
0.8465
0.9295
Two Test Rank Ave Correlation
0.8971
0.9188
0.9534
4 Test Ave Rank Correlation
0.9348
0.9516
0.9721
Single Test Rank Ave R2
0.679
0.717
0.864
Two Test Rank Ave R2
0.805
0.844
0.909
4 Test Ave Rank R2
0.874
0.906
0.945
So I think this table is full of good news. By moving to PCM 20 to 22 we get small improvements in the correlation of the average single race .8242 to .8465 but narrowing RDC to two gets us another gain up to .9295. Also I would note that the average single run in PCM22 RDC=2 is more correlated than the average of 2 runs in PCM20 and almost as strong as if we ran 4 runs.
The r2 just puts this in terms of % of variation explained. The amount of increase in variation explained by stats from PCM22 RDC = 5 to PCM 22 RDC = 2 is about 15% (.864 - .717). And I would guess that is probably a conclusion we can extend across other race types (maybe not sprints).
Bringing this home in another way Zmorka's finishes in each test were:
I had been thinking that two is probably a bit low, cause +2 and -2 are negligible, the evidence supports that it removes a very large portion 'randomness' (its only random if your rider fails not the other way around tho). I was gonna run some +3s on my own because I feel like +3/-3 more accurately represent good and bad days which should exist in the game.
quadsas wrote:
I had been thinking that two is probably a bit low, cause +2 and -2 are negligible, the evidence supports that it removes a very large portion 'randomness' (its only random if your rider fails not the other way around tho). I was gonna run some +3s on my own because I feel like +3/-3 more accurately represent good and bad days which should exist in the game.
Think I agree with this but haven't played PCM22 or seen those testing results so would be interested!
Thanks for your great work Ulrich
24/02/21 - kandesbunzler said “I don't drink famous people."
15/08/22 - SotD said "Your [jandal's] humour is overrated"
11/06/24 - knockout said "Winning is fine I guess. Truth be told this felt completely unimportant." [ICL] Santos-Euskadi | [PT] Xero Racing
