[BUG] Training form probabilities

[Brian Clough]

I have been keeping careful track of training results for over a week now, and there seems to be a bug in the training system because my observed frequencies of training form results on my team differ significantly from those posted on the training page.

To remind everyone, current system purportedly works this way
    Light : The chance of getting bad and good training result are 50% and 0% respectively.
    Medium : The chance of getting bad and good training result are 5% and 3% respectively.
    Heavy : The chance of getting bad and good training result are 0% and 12% respectively.
    Very Heavy : The chance of getting bad and good training result are 0% and 30% respectively.

My results (updated daily):
Setting               Number             Bad                   Okay                  Good
Light                     40                  19 (47.5%)         21 (52.5%)            0 (--%)
Medium                247                 21 (8.5%)        226  (91.5%)       0 (0%)
Heavy                  113                   0  (0%)             107  (94.7%)         6 (5.3%)
Very Heavy            22                    0  (--%)             16 (--%)               6 (--%)
emboldened figures differ significantly from those above
Flipping a coin 5 or 10 times won't necessarily result in 50% for each side. But flipping a coin 100 or 1000 times should produce results that are increasingly close to actual values.

I will continue to update my results, and encourage others to share theirs.
Observed probabilities will be posted only after I have around 30 "samples" for statistical power.

so far I can conclude that there are some significant differences in results above from those figures outlined on training page, which constitutes evidence of a bug in the system...more conclusions to come as adequate data becomes available

Edit: At current rates, the probability of my 201 training results at medium setting without a single good form result is 0.97^224 = 0.1%, a very unlikely outcome excluding a bug

[Mr.Es]

I feel like you too but not sure that is a bug .

[gabrielis]

yeah, continue updating. i should post myself in gko record post-no good training form for a month

[Brian Clough]

I feel like you too but not sure that is a bug .

It is not a feeling...my conclusions are based on binomial statistical analysis using a significance (α) level of 95%. This means that for each highlighted result, there is at least a 95% chance that highlighted differences between my observed results and the true probabilities are not attributable to chance. With more data this may rise to 99% for both medium and hard training settings (already above 99% for good form result from medium setting).

Thank goodness my college education didn't go to waste *lol* 

[SoLid_frOg]

Maybe you should show it all ie. null hypothesis, alternative hypothesis, t-statistic, t-critical etc. lol

But i think the results skew to the "bad" too.

[Brian Clough]

Maybe you should show it all ie. null hypothesis, alternative hypothesis, t-statistic, t-critical etc. lol

But i think the results skew to the "bad" too.

pmsl

I just might if admin doesn't address this pretty soon...


Meanwhile, if anyone has ever gotten good training form result from medium setting I'd like to know...I know at least two of us haven't yet

[panchchou]

I checked my training report everyday. and never found the "good" report.

[gabrielis]

this is ridiculous, today 4/22 players with medium workload had bad form. thats almost 1/5

[SoLid_frOg]

I have a bad feeling about this.

And i'm not a jedi lol.

[Brian Clough]

[gabrielis]

it looks like admin is ignoring

[gabrielis]

you dont have any problems in saving new youth trainings?i am not allowed to save it

[Brian Clough]

Whether admin wants to address this or not, it is really looking like the ACTUAL probabilities are something very close to this:
  Light : The chance of getting bad and good training result are 50% and 0% respectively.
  Medium : The chance of getting bad and good training result are 8% and 0% respectively.
  Heavy : The chance of getting bad and good training result are 0% and 5% respectively.
  Very Heavy : The chance of getting bad and good training result are 0% and 30% respectively.
 

[Kr10s]

So... i wasn't crazy after all 

[gabrielis]

you dont have any problems in saving new youth trainings?i am not allowed to save it

[Brian Clough]

Admin, if you can't or don't want to fix this bug (anytime soon) then at least change the text on the training window to reflect accurate values of training form probabilities. No one in GKO has ever gotten good form with medium training, so this really should be declared! If you are working on it (or plan to soon), please let us know!

[gabrielis]

6 bad training forms with medium workload, anyone had more? :-) no good training for lots of months

[gabrielis]

admin has posted ban list but no response here so far

[paul_gazza]

yesterday 6 bad, now 5 bad... GOOD....

[MathewsPateras]

Yesterday 5 bad, today 8 bad....GOOD 0!!
All with medium trainning

[Brian Clough]

@buncombe: Do you have updates stats of training results for this period. Im beginning to sense that its actually even worse than the presentet data suggests.

I got tired of doing it after admin refused to even address this or say they are working on it. Basically, these are my conclusions:

Whether admin wants to address this or not, it is really looking like the ACTUAL probabilities are something very close to this:
  Light : The chance of getting bad and good training result are 50% and 0% respectively.
  Medium : The chance of getting bad and good training result are 8% and 0% respectively.
  Heavy : The chance of getting bad and good training result are 0% and 5% respectively.
  Very Heavy : The chance of getting bad and good training result are 0% and 30% respectively.
 

...and I stick by these.
While it may seem worse when 5 or 8 bad form results come up. On average, one should expect 2 bad form results out of 24 on medium training. With these odds, the probability of getting 8 bad form is not that extraordinarily unlucky...

[admin]

This feature has been tested and it works properly.
The small amount of data that you guys collected can't be used as a statistical data.

GKO

[gabrielis]

This feature has been tested and it works properly.
The small amount of data that you guys collected can't be used as a statistical data.

GKO

so not getting good training form for 2 months is normal?for god sake, nobody has posted a fact about getting good training form and you say its normal?cmon, thats not even funny

[Brian Clough]

This feature has been tested and it works properly.
The small amount of data that you guys collected can't be used as a statistical data.

GKO

Really? Let's find out!
I hypothesize that observed good form results from medium training are statistically different from those reported.
 
Test 1: z-test using normal curve
A z-proportions test was used to to test whether significantly fewer "good" form results from medium training intensity were observed than are reported (3%). Of the 247 observations, 0 (0%) were good and 247 (100%) were not. This difference is statistically significant (z = -2.7639) at α = 0.05, indicating that fewer good form results were observed than would be expected by chance. Thus the probability that the actual probability of getting good form result from medium training is less than the reported 3% is > 99.71%.

Test 2: chi-square test
A chi-square test was also used to test whether significantly fewer "good" form results from medium training intensity were observed than are reported (3%). Again the difference is statistically significant (X² = 7.6392, df=1) yielding the precise p-value of 0.006092.

n = 247 for both tests.

Thus we should reject the null hypothesis (there is no difference between observed and reported probabilities) in favor of the alternative (stated above). These results provide very strong evidence that there is at least one bug in this feature.

The z-test is easily done using a calculator and normal probability table.
The chi square test was done using two separate programs found online here and  here so anyone can run the test for themself.
However, as Gabriellis pointed out, the fact that there have not been any reported good form results from medium training really speak for themself!
 

If someone else can record and post 50-100 results from hard training we can test whether there is also a bug in good form results from hard training. Due to the relatively low number of observations I have made there, the resulting p-value is a little bit high to draw conclusions with real confidence (p <0.08 based on the data in my original post), but with some more, much stronger evidence can be reported. I just don't have the time tbf.
 

[Ruta]

This feature has been tested and it works properly.
The small amount of data that you guys collected can't be used as a statistical data.

GKO

I never have good form from medium training either. I have more then 15 players in that regime every day!
So your test must be wrong Or you has been tested only about bad form which woks properly.