Lies, damn lies, & statistics.

[pokerfan]

As I said earlier the only use I can think of is that it might show that you play badly on long sessions (if you had sufficient samples).  ie your average win rate in long sessions is lower than in short sessions.  Probably quite important for an old boy like Red.

Oi!

As it happens, my ave session length is, apparently, 8.14 hours, followed by 4 hours of tantric sex, followed by an hour's drive home to Mrs Red.

Haha.

[skolsuper]

Sample size has nothing to do with it, the measure is just mathematically wrong and therefore totally unreliable. It gives undue weight to results from shorter sessions, a breakeven or even slightly losing player who quits early when they get ahead would have a positive average win rate by this measure even after a million samples. If that player tried to garner anything from it, they'd play shorter and shorter sessions for absolutely no good reason. Just ignore it, it tells you literally nothing.

The overall stat is useless, but if you filtered session length (don't know if you can) it might have some use.

It would then exactly match the overall average winrate for those sessions. I pm-ed Jamie a proof of this, can repost here if you're interested.

[doubleup]

Sample size has nothing to do with it, the measure is just mathematically wrong and therefore totally unreliable. It gives undue weight to results from shorter sessions, a breakeven or even slightly losing player who quits early when they get ahead would have a positive average win rate by this measure even after a million samples. If that player tried to garner anything from it, they'd play shorter and shorter sessions for absolutely no good reason. Just ignore it, it tells you literally nothing.

The overall stat is useless, but if you filtered session length (don't know if you can) it might have some use.

It would then exactly match the overall average winrate for those sessions. I pm-ed Jamie a proof of this, can repost here if you're interested.

If we assume that we have sufficient samples, the fact that the overall stat and the session stat are significantly different might alert you to look deeper into your stats?

I agree though that it does look like someone just put it in without thinking it through, just trying to suggest some reason for it being there.

[Jon MW]

Sample size has nothing to do with it, the measure is just mathematically wrong and therefore totally unreliable. It gives undue weight to results from shorter sessions, a breakeven or even slightly losing player who quits early when they get ahead would have a positive average win rate by this measure even after a million samples. If that player tried to garner anything from it, they'd play shorter and shorter sessions for absolutely no good reason. Just ignore it, it tells you literally nothing.

The overall stat is useless, but if you filtered session length (don't know if you can) it might have some use.

It would then exactly match the overall average winrate for those sessions. I pm-ed Jamie a proof of this, can repost here if you're interested.

If we assume that we have sufficient samples, the fact that the overall stat and the session stat are significantly different might alert you to look deeper into your stats?

I agree though that it does look like someone just put it in without thinking it through, just trying to suggest some reason for it being there.

Given that it's just looking at averages and averages of averages the statement about it being 'mathematically wrong' is a bit odd - 'statistically meaningless' might be more appropriate.

Given a sufficient sample size you can make any kind of statistic and analyse it - whether they were in any way useful would be a different question.

[RED-DOG]

So does it matter whether I use average or overall win rate per hour?

[skolsuper]

Sample size has nothing to do with it, the measure is just mathematically wrong and therefore totally unreliable. It gives undue weight to results from shorter sessions, a breakeven or even slightly losing player who quits early when they get ahead would have a positive average win rate by this measure even after a million samples. If that player tried to garner anything from it, they'd play shorter and shorter sessions for absolutely no good reason. Just ignore it, it tells you literally nothing.

The overall stat is useless, but if you filtered session length (don't know if you can) it might have some use.

It would then exactly match the overall average winrate for those sessions. I pm-ed Jamie a proof of this, can repost here if you're interested.

If we assume that we have sufficient samples, the fact that the overall stat and the session stat are significantly different might alert you to look deeper into your stats?

I agree though that it does look like someone just put it in without thinking it through, just trying to suggest some reason for it being there.

Given that it's just looking at averages and averages of averages the statement about it being 'mathematically wrong' is a bit odd - 'statistically meaningless' might be more appropriate.

Given a sufficient sample size you can make any kind of statistic and analyse it - whether they were in any way useful would be a different question.

I reiterate, sample size is not going to help. It is mathematically wrong because it is treating numbers that are not equivalent as if they were, comparing apples and oranges. It is as useful as a one-number average of your opponents' heights and weights. It's a meaningless number anyway, and if it goes up there's no way to tell (from just the number) if it's because your opponents are getting taller or heavier.

@RED-DOG, use overall winrate per hour. Ignore the other one.