Question for Don: Tracking EV and Variance

[DSchles]

Thanks for weighing in. The reason I try and track EV and variance on a per shoe basis is because every shoe is unique. By that, I mean, one shoe might be heads up, 1.5 decks cut off, with 2 hands per round. Another shoe might have 1.75 decks cut off, 4 players, with 1 hand per round. Wouldn't the EV and variance for each of these shoes differ? Isn't it important to properly track EV and variance to locate AV on the bell curve? This is why I'm considering recording these variables for every shoe that I play. Does that make sense?

No, it makes no sense at all. First, your personal individual results are totally meaningless in determining EV and variance,. I have stated this over and over again ad nauseam. You CANNOT determine how good or bad a game is by actually playing and looking at your results.

Do you need a sim to tell you that the first game you described is much better than the second? But once you've answered no, and you want to know the specifics, you run a couple of billion-hand sims for each one to determine the correct EVs and variances for each. You don't look at your personal results to determine such numbers.

Do you understand??

Don

[MJ1]

No, it makes no sense at all. First, your personal individual results are totally meaningless in determining EV and variance,. I have stated this over and over again ad nauseam. You CANNOT determine how good or bad a game is by actually playing and looking at your results.

I agree completely but that is not what I'm asking. I'm not suggesting to look at personal results to determine EV and variance. That is why we have simulations! Personal results are statistically insignificant and indeed meaningless.

What I'm saying is that I record how many hands I play per round, penetration, # players, and whether it was play all or Wong out, for each shoe that I play. I do NOT record the win/loss from each shoe. So, suppose I play 24 shoes in an 8-hour session. I win $800. Now I would like to know where this result falls on the bell curve. So, I look back at my records for the session and run a simulation for each type of shoe that I played to determine theoretical EV and variance for each shoe. I then use the sim output and calculate aggregate EV and variance for the 24 shoes. I then use that info to calculate where my result falls on the bell curve.

I would do the same thing for subsequent sessions and determine cumulative EV and variance (via simulation) over however many sessions I play and see where my results fall on the bell curve. So, after 20 sessions, I might play through 480 shoes and using my system, I would know precisely what the cumulative EV and variance is through simulation.

Now does what I'm saying sound reasonable? I would like an accurate method to determine my EV and variance over the course of several hundred shoes. I believe you estimate rounds per hour to be 100. I suppose you also use a single number for # players and penetration. I'm not comfortable with this 'one size fits all' approach. There has got to be a better way.

Thanks,
MJ

[DSchles]

The one size fits all that you mention is uniquely so that SCORE can make apples-to-apples comparisons. If you prefer to generate hourly win rates for different scenarios, by all means be my guest. But, for the life of me, when you write, yet again, "I would like an accurate method to determine my EV and variance over the course of several hundred shoes," I'm trying to explain to you that a result over several hundred shoes is utterly meaningless. Can you understand that?

I truly suggest a rereading of the least commented-upon chapter in BJA3, chapter 12, which has always disappointed me. A great deal of work went into writing it and doing the simulations, but it has never gained the recognition I feel it deserves. And, I think it is mostly because so many people concentrate and rely on short-term results in order to make (mostly erroneous) statements about win rates that they don't understand the error of their ways.

Finally, I understand your question, even though I'm not sure it's very important. If you want to aggregate results from all different scenarios of shoe play, just add and average-weight the EVs and the variances. But make sure you do it for billions of hands for the true EV and variance and not from your personal results from "hundreds" of shoes, which have little or no meaning at all. Unless you're talking about 4-N0, if not nine-N0, results, which could be thousands of shoes, you're probably just wasting your time, even for the part where you determine your personal EV.

Clear?

Don

[artoozi]

Can you just do a normal hourly EV to AV comparison? Trying to do it on a per shoe basis seems like it's adding an extra layer of complexity with no benefit.

[DSchles]

+1.

Don

[MJ1]

Thanks, Don. I think I see what you are saying. The results from a couple hundred shoes are not statistically significant, so why bother comparing EV to AV so early? Well, it might provide some useful information. What if after a couple hundred shoes my AV is 3 standard deviations to the left of EV? This will indicate that I'm either being cheated or making errors at the table. In any case, a couple hundred shoes will eventually turn into a couple thousand shoes. I would like to track results from the very beginning.

MJ

[aceside]

Just a suggestion. When you post a question, it’s better not to direct it specifically to Don the blackjack expert. I haven’t read this long stair, but I am sure many experts here can answer your questions.

[MJ1]

Can you just do a normal hourly EV to AV comparison? Trying to do it on a per shoe basis seems like it's adding an extra layer of complexity with no benefit.

Ok, maybe I'm missing something. Let's talk specifics. Over the course of an hour, I play through 3 shoes. The particulars are as follows:

Shoe #1
4 players, 1.5 decks cut off, 1 hand/round, Wong out

Shoe #2
Heads up, 1.75 decks cut off, 2 hands/round, play-all

Shoe #3
5 players, 1.25 decks cut off, 1 hand/round, play-all.

Now, how would you go about estimating your EV and variance for the hour of play? Better yet, several hundred hours of play? Are you going to use a 'one size fits all' approach and come up with some gross approximation of say 2 players, 1.5 decks cut off, 1 hand/round, play all, 100 rounds played per hour? This would not yield a reliable estimate. The only reasonable methodology seems to be to track the aforementioned variables for each shoe and then run simulations to find the theoretical EV and variance. Yes, it will take 1000s of shoes to reach N0.

MJ

[artoozi]

Ok, maybe I'm missing something. Let's talk specifics. Over the course of an hour, I play through 3 shoes. The particulars are as follows:

Shoe #1
4 players, 1.5 decks cut off, 1 hand/round, Wong out

Shoe #2
Heads up, 1.75 decks cut off, 2 hands/round, play-all

Shoe #3
5 players, 1.25 decks cut off, 1 hand/round, play-all.

Now, how would you go about estimating your EV and variance for the hour of play? Better yet, several hundred hours of play? Are you going to use a 'one size fits all' approach and come up with some gross approximation of say 2 players, 1.5 decks cut off, 1 hand/round, play all, 100 rounds played per hour? This would not yield a reliable estimate. The only reasonable methodology seems to be to track the aforementioned variables for each shoe and then run simulations to find the theoretical EV and variance. Yes, it will take 1000s of shoes to reach N0.

MJ

You could take each shoe scenario and figure the hourly EV, let's say $90/hour. Then figure the actual time you played, lets say, 20 min for 1 shoe, which is 1/3 of an hour. So the EV for that shoe would be 90 / 3 = $30.

[MJ1]

You could take each shoe scenario and figure the hourly EV, let's say $90/hour. Then figure the actual time you played, lets say, 20 min for 1 shoe, which is 1/3 of an hour. So the EV for that shoe would be 90 / 3 = $30.

Hourly EV is a function of how many rounds you play per hour. This of course will vary based upon # players at the table and dealer speed. So how do you estimate rounds per hour? Wouldn't it be easier to just use a sim to determine, on average, how many rounds there are per shoe, given penetration and # players at the table?

MJ