Page 60 of 61
Re: 4NCL Online
Posted: Mon Aug 17, 2020 8:16 am
by Matthew Turner
Roger,
I think it is pretty well established that average online performance is lower than average OTB performance. This is reflected in the 4NCL graphs. We could easily incorporate this into the Regan results by lowering everyone's rating by 100 points and increasing their z scores by circa 0.5
All the evidence suggests that would indeed make the Statistical analysis more accurate and we could have more players banned for cheating, but this is not a change I am in favour of.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 9:01 am
by Roger Lancaster
Matthew Turner wrote: ↑Mon Aug 17, 2020 8:16 am
Roger,
I think it is pretty well established that average online performance is lower than average OTB performance. This is reflected in the 4NCL graphs. We could easily incorporate this into the Regan results by lowering everyone's rating by 100 points and increasing their z scores by circa 0.5
All the evidence suggests that would indeed make the Statistical analysis more accurate and we could have more players banned for cheating, but this is not a change I am in favour of.
Matt, I think you're very likely correct [in thinking that average online performance is lower than average OTB performance] but the fact we might both believe it doesn't make it true. And, even if it is, now that you've accepted the principle that the means might be [and likely are] different, why are you so resistant to the idea that the standard deviations might also be different?
Re: 4NCL Online
Posted: Mon Aug 17, 2020 9:22 am
by MartinCarpenter
Roger Lancaster wrote: ↑Mon Aug 17, 2020 9:01 am
Matthew Turner wrote: ↑Mon Aug 17, 2020 8:16 am
Roger,
I think it is pretty well established that average online performance is lower than average OTB performance. This is reflected in the 4NCL graphs. We could easily incorporate this into the Regan results by lowering everyone's rating by 100 points and increasing their z scores by circa 0.5
All the evidence suggests that would indeed make the Statistical analysis more accurate and we could have more players banned for cheating, but this is not a change I am in favour of.
Matt, I think you're very likely correct [in thinking that average online performance is lower than average OTB performance] but the fact we might both
believe it doesn't make it true.
There's some fairly solid emperical evidence in favour of it being true as well, so I genuinely don't see why anyone would worry about it?
Roger Lancaster wrote: ↑Mon Aug 17, 2020 9:01 am
And, even if it is, now that you've accepted the principle that the means might be [and likely are] different, why are you so resistant to the idea that the standard deviations might also be different?
They will be, yes. There's a significant extra factor - how well people react to being moved to playing online - in the second set of data. That basically 'has' to increase the SD.
The globally reduced playing strength gives you a pretty big safety factor to compensate though.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 9:33 am
by Matthew Turner
Roger and Martin,
It really doesn't matter what the overall standard deviation is because we only care about the part of the distribution where players are massively outperforming their rating. So we are back to where I started, Is it more or less likely that a player will objectively outperform their rating by 600 points online as compared to OTB.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 1:19 pm
by Roger Lancaster
Matthew Turner wrote: ↑Mon Aug 17, 2020 9:33 am
It really doesn't matter what the overall standard deviation is because we only care about the part of the distribution where players are massively outperforming their rating.
I'm sorry but this indicates a failure to grasp the subject. It really does matter. It matters because, if there are irregularities in other parts of the distribution which can't be accounted for by one's chosen explanation [in this case, cheating] then one has to identify the cause[s] because the same cause[s] may, and very likely will, impact on the part of the distribution in which one
is interested.
My hypothesis, and I'd prefer to state it in my own words, is that the excess figures at the 'high' end of the online graph are only partly due to cheats and are partly due to the way the evidence is produced.
To support this, let's look at the opposite end of the graph. With a Normal distribution, some 92.8% of data should fall in the central 1.8>z>-1.8 range. So 3.6% should occur at values lower than -1.8. In fact, as closely as I can calculate, the figure for OTB data is 2.9% [within acceptable margin of error] but the figure for online data is 7.4% which decidedly isn't. Of course, it's possible to argue that this is the result of three dozen players indulging in cheating through rating manipulation [that is, deliberately playing badly] but this doesn't seem overly likely. My explanation would be that the excess online values in the -1.8>z range occur as a result of the online data having a higher dispersion [or scatter or spread or whatever you like to call it] than the OTB data and that this is a contributory factor in explaining the larger excess for high positive values of z.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 1:55 pm
by Joseph Conlon
Roger:
you seem to be assuming that the online distribution ought to be normal. The normal distribution has the feature that the number of 'outliers' - i.e. the chance of being 'far' from the mean, here z>3 - depends on the standard deviation, and increases drastically if the s.d. increases.
But there is no reason for this online distribution to be normal, not least because it certainly has a definite population at z>3 arising from engine assistance.
A statement that (a) the online distribution is normal AND (b) the s.d. is increased by 50% for online play carries the direct implication that the chance of someone online playing above their rating by 600 points is something like 30 times higher than OTB, and then also that the chance of outperforming their rating by 800 points is 500 (? - typing fast so haven't time to check the SD tables) times more likely online than OTB.
To me these conclusions suggest that either (a) or (b) is wrong.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 1:59 pm
by MartinCarpenter
Also that its symmetrical!
My hypothesis to explain the observed OTB vs Online graph is that the mean of the distribution has moved left of what is being used as the mean for testing purposes (about 1 standard deviation) and that there's then a basically normal distribution - with similar SD to over the board play.
It might well also slightly 'naturally' skew left due to some people really not adjusting to online play at all well.
There is then a markedly large excess of extreme results on right hand side of the distribution to explain, a large percentage of those very likely to be cheating.
Re: 4NCL Online
Posted: Mon Aug 17, 2020 6:19 pm
by LawrenceCooper
LawrenceCooper wrote: ↑Mon Aug 17, 2020 7:23 am
A further increase to 221 teams.
Yet another update, now 223 teams. One more and they could have seven divisions of 32
Re: 4NCL Online
Posted: Tue Aug 18, 2020 9:18 pm
by Neil Graham
LawrenceCooper wrote: ↑Mon Aug 17, 2020 6:19 pm
LawrenceCooper wrote: ↑Mon Aug 17, 2020 7:23 am
A further increase to 221 teams.
Yet another update, now 223 teams. One more and they could have seven divisions of 32
Or Divisions 5A, 5B and 5C of equal strength with play-offs to decide promotion.
Re: 4NCL Online
Posted: Tue Aug 18, 2020 9:48 pm
by LawrenceCooper
Neil Graham wrote: ↑Tue Aug 18, 2020 9:18 pm
LawrenceCooper wrote: ↑Mon Aug 17, 2020 6:19 pm
LawrenceCooper wrote: ↑Mon Aug 17, 2020 7:23 am
A further increase to 221 teams.
Yet another update, now 223 teams. One more and they could have seven divisions of 32
Or Divisions 5A, 5B and 5C of equal strength with play-offs to decide promotion.
I believe that 5 x 32 and a swiss for the rest is the most likely.
Re: 4NCL Online
Posted: Wed Aug 19, 2020 4:42 am
by Richard Bates
They should just have a massive Swiss for everybody and play until such time as the online competition is no longer required!
Re: 4NCL Online
Posted: Wed Aug 19, 2020 7:13 am
by LawrenceCooper
LawrenceCooper wrote: ↑Mon Aug 17, 2020 6:19 pm
LawrenceCooper wrote: ↑Mon Aug 17, 2020 7:23 am
A further increase to 221 teams.
Yet another update, now 223 teams. One more and they could have seven divisions of 32
226 teams now.
Re: 4NCL Online
Posted: Sat Aug 22, 2020 7:22 pm
by LawrenceCooper
Re: 4NCL Online
Posted: Sat Aug 22, 2020 7:27 pm
by Richard Bates
Can you explain why some round 1 pairings seem to involve teams not in the same division? Or am i missing something obvious?
Re: 4NCL Online
Posted: Sat Aug 22, 2020 7:45 pm
by LawrenceCooper
Richard Bates wrote: ↑Sat Aug 22, 2020 7:27 pm
Can you explain why some round 1 pairings seem to involve teams not in the same division? Or am i missing something obvious?
There are some errors in the tables that need correcting. White Rose 1 & 2 need swapping as do Wood Green & Wood Green Monarchs. If there's anything other than that I can pass on a message. I suspect that the 4NCL have had enough of me querying things today though
