[hrothgar]

inquiry, on 2012-November-24, 13:37, said:

well, these numbers even with stats are not going to help bud out, as these were not generated using "HIS" hands. And there is absolutely no reason to limit the test ot 20,000 deals. He should use 250,000 deals (giving a million unique hands)... or maybe even more. Still, the data presented above easily passes the Chi-Square goodness-of-fit test with unequal expected frequencies, he could even quote the p values (of course, not his deals).... The people at his club will not understand or accept that kind of validation either.

On the 20,000 versus 250,000 front: There is this handy little principal in stats known as asymptotic convergence. You don't need a quarter million hands. As I mentioned earlier in this thread, in the case of a chi square goodness of fit test, you want the expected number of observations to be at least 5.

http://www.talkstats...Chi-Square-test

(BTW, one of the limitations with Excel is that it still works very poorly if the number of rows increases beyond 65,536)

Personally, I refuse to accept that whether the people at the club with understand or accept the results of the analysis has anything to do with the choice to run the Chi square test.

1. If you genuinely believe that you're dealing with people to are immune to logic and reasoning then you should be blowing them off from day one. When they bitch about the deals do your best to ignore them.

2. However, if you are going to go to the time and bother of analyzing the data, then you should do so to the best of your abilities. To quote Hunter S. Thompson, "If a thing like this is worth doing at all, it's worth doing right".

FWIW, I wouldn't get all bent out of shape if someone went off and analyzed the data and didn't present me with a formal hypothesis test provided that they didn't know any better. As you folks have noted, not everyone has taken a real course in statistics. If someone doesn't know any better, so be it. Use it as a learning opportunity.

I might even prefer if the basic version of the presentation didn't consider these issues. (When I present to our senior management, I don't have confidence intervals in the PowerPoint deck. However, I make damn sure that I have performed a legitimate analysis and have back up slides available because if I don't and I get called on it I'm going to look like I don't know what I'm talking about)

I find your attitude baffling: You're willing to do 99% of the work, including the real heavy lifting. However, some perverted sense of stubbornness is preventing you from actually crossing the finishing line and performing a legitimate / useful analysis.

[mikestar13]

No point in doing any math if the goal is to convince skeptics among average club players. An example from my own experience as bridge director at a senior center. One Christmas season, we were having a special holiday game -- a two section Michell. As a treat for the players, I pre-duplicated the boards for both sections so that the players didn't have to do any dealing, and the two sections could play the same boards. The only response I got for my efforts was 24 boards of bitching about "these damned computer dealt hands", which I had lovingly prepared and hand dealt. Don't ever waste your chi-squares on club players!

[blackshoe]

If it were me, I'd just want to do one thing — and that's tell the skeptic to just STFU.

[BudH]

Thanks for the responses. I'm in San Francisco and just finished playing the finals of the Nail Life Master Pairs. (Unfortuntatly, a slightly below average evening session took me out of the running for a high finish.)

I might give the Excel method a try. Tim G is correct in that it will be difficult to convince the skeptics at my club that there isn't any funny business going on.

That being said, there were two days in the first few weeks where there were 12 and 14 voids in a session when there would normally be less than 7 voids expected. And looking over the hand records for those two days, I can seen why someone would think something wacky had occurred. But all the other days were somewhat close to normal.

So far, for the first 424 deals at our club,

Number Expected

Balanced 807 806.4
Voids 98 86.8
Singletons 534 542.4
7+ card suit 72 68.4

Although the voids are greater than expected, the number of splinters (voids plus singletons) is close to predicted.

After trying to defend "computer hands" recently, I was amused when the first hand in the qualifying sessions for the Nail Life Master Pairs had me bid 2S preemptive white vs. white with K876xxx and I found AQJ95 behind me with both black suits splitting 5-0!

Bud

[hrothgar]

BudH, on 2012-November-25, 03:34, said:

Tim G is correct in that it will be difficult to convince the skeptics at my club that there isn't any funny business going on.

That being said, there were two days in the first few weeks where there were 12 and 14 voids in a session when there would normally be less than 7 voids expected. And looking over the hand records for those two days, I can seen why someone would think something wacky had occurred. But all the other days were somewhat close to normal.

Here's the rub...

Folks won't ever remember the 3,048 different ways that the hands don't look fishy. They'll simply fixate on the fact that there happened to be too many voids.

If you look at enough sessions, you're going to find something that looks "weird".
If you don't, that's a near certain sign that something is wrong with your dealing program.

I'm going to attach a well known example from statistics. Which of the following two images looks more random?




Check out the following for an explanation http://blogs.discove...ing-randomness/

or jump to http://telescoper.wo...poisson-davril/ for a more detailed treatment

[nige1]

Published lists of random numbers or hands aren't random. For example, if a table of random numbers comprised ten million consecutive sevens, statisticians would reject it, although such a sequence is as likely as any other. Similarly a hand generator that dealt 13 card suits for half a dozen consecutive deals.

[BudH]

inquiry, on 2012-November-24, 13:37, said:

well, these numbers even with stats are not going to help bud out, as these were not generated using "HIS" hands. And there is absolutely no reason to limit the test ot 20,000 deals. He should use 250,000 deals (giving a million unique hands)... or maybe even more. Still, the data presented above easily passes the Chi-Square goodness-of-fit test with unequal expected frequencies, he could even quote the p values (of course, not his deals).... The people at his club will not understand or accept that kind of validation either.

It would be great to generate a few hundred thousand deals and check out even the void/singleton/7+ card suit statistics.

Unfortunately, Dealer4 is limited to 640 deals and BridgeComposer freezes up well shy of 10,000. Any other program out there that could endure a 100,000 deal PBN file to be loaded into it?

[helene_t]

As a good Bayesian I don't believe much in hypothesis testing. But is this case the null hypothesis a quite plausible so it makes a lot of sense. A chi-square test on Ben's numbers gives a p-value of 0.639665152.

[pigpenz]

if you wanted to study hands from BBO
Stephen Picketts bridge browser program would be great
but I dont think there is any new data from BBO or OKBridge since 2007
but the data before that is probably millions of hands

it just depends on what type of hand information you are looking for

[inquiry]

BudH, on 2012-December-16, 21:42, said:

It would be great to generate a few hundred thousand deals and check out even the void/singleton/7+ card suit statistics.

Unfortunately, Dealer4 is limited to 640 deals and BridgeComposer freezes up well shy of 10,000. Any other program out there that could endure a 100,000 deal PBN file to be loaded into it?

Again, you can use excel to do this for you. Send me 250,000 pbn hands via email (inquiry at bridgebase dot com) and I will get them into excel with the distributional and hcp information for you (thinks like how many hands have 22 hcp, or 15, as well as the singleton and voids and long suit things.... AND send you back the excel file so you can tweek it to get other information you want.

As for bridgebrowser, it is great, but it will never address YOUR hands... unless picket releases it into public domain and allows you to add hands to it.