Can you learn anything from 5,000 rounds?

[Bigdaddy]

This post is for the technical analysts -

From the beginning of this year, I've been meticulously tracking my results, including number of rounds in the session, win/loss, EV, estimated average bet and standard deviation based on the game rules and spread employed (thanks to CVCX).

I have around 5,000 rounds in so far and I'm running way above EV, but substantially all of the overage appears to be attributable to one or two crazy sessions where everything went right; i.e., sessions that were greater than 1.5 standard deviations.

So it occurred to me that maybe I could make something more meaningful out of my 5,000-round results if I were to look at the results only for sessions that had a z-score of between -1.3 to +1.3. This covers about 80% of expected results. The thinking is that the distribution of these results should tend to accumulate closer to the true EV number. And if I was way over or way under EV based on those sessions, then that might be an indication that I was using an incorrect EV (either over or underestimated).

What do you think?

[Norm]

5,000 rounds really has no meaning at all IMO.

[stopgambling]

If the 5000 rounds of play all have an edge of 5% or greater then it has a lot of meaning.

[DSchles]

It is virtually impossible to conclude anything from some players' entire lifetimes of play. 50 hours isn't even worth discussing. I used to play 50 hours every time I went to Las Vegas for 10-day trips. Drawing conclusions from any one of those trip results would be an utter waste of time.

Don

[Stealth]

I believe the tracking and analysis of these results will begin to provide valuable information both on the cumulative progress and on the incidents of abnormal variance that may have occurred. I do not believe such analysis provides answers but more likely trends, flags and perspectives that can be very valuable to your journey.

Needed sampling for statistical analysis would require the collection of more than 5,000 rounds to begin to be meaningful.

I have very similar data on over 4,000 sessions of (both team play and individual play) and have found a lot of meaningful data from analyzing those results with a wide variety of statistical and graphical tools.

For example here is a screen shot and some graphs showing various metrics on 498 double deck sessions over about a two year period:

SCOREBOARD of Cumulative results
Scoreboard.JPG

Cumulative performance showing EV/Actual and Standard Deviation
Cumulative.JPG

Session ZScores and a cumulative ZSCORE trend line
Zscores.JPG

Distribution of session Zscores for the 498 sessions
ZscoresDistribution.JPG

Range of results depicting standard deviation of normalized sessions (bars) and the actual results (line) that occurred. A visual to illustrate both the range of volatility of the play and the actual results.
Range of Results.JPG

These charts represent some 498 session and over 50,000 rounds. How "meaningful" they are is an interpretation you will need to make but for me they are highly instructive.

[BJGenius007]

Even one million rounds mean nothing. I would say ten million rounds is minimal. I have my simulation program run ten million rounds each time. The program is accumulative. After running twenty times or two hundred million rounds, I feel the data are much better in the sense that the result is consistent. In other words, the result of two hundred million rounds will be consistent but the result of ten million rounds may not.

[Bodarc]

It is virtually impossible to conclude anything from some players' entire lifetimes of play.

Don, if this is true isn't everything in BJA meaningless because your entire lifetime of play means nothing? It would all boil down to pure luck. You could forget bet ramps, index numbers, spread, penetration and everything else you've studied. A non AP would have just as much chance of being lucky as you have.

I have played enough on CVBJ to have seen the downswings you can have and if those downswings happened when you were starting out it could be disastrous and I agree that would be bad luck. How about trading stocks etc. Wouldn't the same principle apply there? A lifetime of trading is meaningless so why bother with any advantage calculations etc. You will either be lucky or unlucky.

The same could be said for a trip bankroll. Why compute one because a trip is meaningless.

The way I see it, you do everything you can to gain an advantage and determine what should happen statistically speaking and then compare your actual results with that. Even though you may be unlucky and be behind the curve, it is still necessary to track your progress.

I agree that no statistical projections can be made from a small sample size but I think tracking your progress from day one and comparing your results against the norm is an important step that should not be overlooked by a player.

Please don't think I am being flippant because I respect you too much to do that. It is just that I see that comment often and it bothers me.

[Bodarc]

What do you think?

Hi Bigdaddy. I think the denominator will take care of everything.

http://beyondnumbers.lvablog.com/201...e-due-happens/

[RS]

Depends what you're looking for or what kind of precision you want after X number of rounds. Are we talking about being within 1% of the EV 98% of the time? Within 10% of the EV 75% of the time? Within 0.5% of the EV 99.99% of the time? Depends on what game and style you're playing (ie: variance), as well as how precise you want those results to be. OP might be looking for "being within 5% of the EV, 95% of the time" while Don might be thinking, "being within 0.01% of the EV, 99.999% of the time". However the problem being -- it's not so hard to go from, "I know what my EV is, how far away from it am I, and what's the probability I'm within / outside of range X"....but it's a bit more difficult to say, "My results are X, what is the probability my EV is between Y and Z"....which is what I think OP is asking.

[Bigdaddy]

I believe the tracking and analysis of these results will begin to provide valuable information both on the cumulative progress and on the incidents of abnormal variance that may have occurred. I do not believe such analysis provides answers but more likely trends, flags and perspectives that can be very valuable to your journey.

Needed sampling for statistical analysis would require the collection of more than 5,000 rounds to begin to be meaningful.

I have very similar data on over 4,000 sessions of (both team play and individual play) and have found a lot of meaningful data from analyzing those results with a wide variety of statistical and graphical tools.

For example here is a screen shot and some graphs showing various metrics on 498 double deck sessions over about a two year period:

Cumulative performance showing EV/Actual and Standard Deviation
Cumulative.JPG

These charts represent some 498 session and over 50,000 rounds. How "meaningful" they are is an interpretation you will need to make but for me they are highly instructive.

Thanks Stealth - I think the above chart is what I had in mind (i.e., a running/cumulative look at EV to actual results within the bounds of standard deviation). I also understand Don's and Norm's comments (ala CVData's warning: "any simulation run for under 200M rounds is meaningless"). So let me pose the question in a different way.

Let's say I'm one of your new team members and I've been assigned to play an outstanding DD game with a SCORE of 100 and N0 of 10,000 rounds. I have 50 sessions in totaling 5,000 rounds and I'm a net (cumulative) loser thus far. I map the results for the 50 sessions on your graph above and it shows that the line for the cumulative/actual results appears to be trending downwards and is running greater than 1.5 standard deviations from the EV line at all points.

Would you become concerned that I might be playing a losing game or that I'm a thief? Or would you wait for the results of the next 85,000 rounds (to reach N03) before assessing the situation and taking remedial action?

[Stealth]

It seems hypocritical to tell me to trust the law of large numbers and then dismiss any results I might see in my lifetime. I believe my results are meaningful in providing insight into the normal or abnormal nature of my results versus the expected range and results of the play at meaningful sample sizes.

I believe these points can be described with sound statistical procedures much like Dr. Richard Reid provided in his e-Book Dynamic Blackjack. All of our results are subject to probability of occurrence and those outliers make for difficult conclusions but their existence does not justify ignoring the entire body of data. Much can be learned regarding the strengths and weaknesses of individual and team play with such analysis.

Borrowing concepts from Six Sigma quality control used extensively in business can provide a statistical path to evaluating play.

So, while I am not sure the context of the statements "that they mean nothing", I am sure they mean something to me.

[Stealth]

Would you become concerned that I might be playing a losing game or that I'm a thief? Or would you wait for the results of the next 85,000 rounds (to reach N03) before assessing the situation and taking remedial action?

In those conditions sample size becomes critical in raising a flag (no conclusion drawn) about a session or number of sessions whose results are outside of normal variance. When such a situation does occur then a review of the results with the player using a concept of "cause and effect" diagrams might be conducted to determine if you can identify causes and then work with the player to improve. An example might be seeing a player log enough rounds to become a "sample size" and determine that their cumulative ZScore is 3.2. Then you should begin to search for over betting, under reporting or rounds played or other possibilities. It is not a witch hunt but a check and balance to early detect problems and to help rectify them.

I would refer you to Dr. Reid's e-Book and a non blackjack but related book "Lean Six Sigma Pocket Toolbook" in paperback for some procedural guidance. Norm's free blackjack book contains a chapter (18) from Dr. Reids book that you may find instructive.

These concepts and practices can be highly informative at a number of levels.

[ZenMaster_Flash]

Long ago and far away ... in Graduate School,
a member of my thesis committee went ape-shit
when I analyzed my data excluding outliers,
a la BigDaddy (see post #1) spawning a furious debate.
I still think that I was correct.