Multi-tabling, which recording method gives a more accurate SD?

[JTG51]

I usually play three tables at the same time when I play online. That gives me two options for recording my results. I can combine the results from each table during the session and record it as one session, or I can record each table as a seperate session.

It never occured to me before, but the two methods should give a very different result for my SD and ROR, right? So which set of numbers is closer to my true SD?

[MaxPower]

I think it depends on your unit of measurement, whether it be per hour or per hand (or per 100/hands). The SD doesn't exist independently of the measurement.

[JTG51]

I calculate everything on a per hand basis, but I don't see how that makes a difference. No matter what unit I chose, my SD (and ROR) is going to be higher if I combine multiple tables into one session because each session is going to be longer and have bigger wins and losses, right?

[Im Just A Bill]

If you are talking about the SD of $ won/lost, then I would think that combining tables would lower your SD (decrease your variance. Bad beat on one table makes less of a difference. This is basically equivalent to playing 3 sessions on a single table and combining the results. Combining tables increases your sample size.

[JTG51]

If you are talking about the SD of $ won/lost, then I would think that combining tables would lower your SD...

Yeah, I think you are right.

That still leaves the question though, which is a more accurate measure?

[daryn]

you should really treat it as 2 or 3 different sessions, as if you split yourself into 3 clones and you each went and played poker.

you'd just have to keep in mind that your bb/hr would actually be bb/hr/table. your bb/100 would be the same.

[Im Just A Bill]

I'm not sure that "more accurate" necessarily has meaning, it just kinda depends on the data you are interested in. For me, I would prefer to track each table independantly so that I would have summary stats based on a single table session. If you always play 3 tables, then you might be more interested in combined summary stats for the session. I don't think there is necessarily a right answer here.

[JTG51]

you should really treat it as 2 or 3 different sessions, as if you split yourself into 3 clones and you each went and played poker.

I do, but I don't see how that's any better that treating it as one session. If I played a 12 hour session at Foxwoods and changed tables twice in the middle, would you say I should split that into three sessions as if my two clones and I each wanted to play?

you'd just have to keep in mind that your bb/hr would actually be bb/hr/table. your bb/100 would be the same.

Yeah, I'm not worried about winrate here. Just SD and ROR. I keep all of my online stats on a per hand basis so that I don't have to worry about the number of tables I'm playing when I track my winrate.


Maybe it's a dumb question that doesn't have a good answer, or maybe I'm not phrasing it well. Let me try again.

Lets say every day I play poker online for 2 hours at 3 tables at a time. I can recod that as a single 6 hour session or as 3 seperate 2 hour sessions. If I always use the first method I'll have an SD of $x. If I always use the second method I'll have an SD of $y. Will the ROR using $x or the ROR using $y be closer to my true ROR?

Thanks again, and I'm sorry if this is a pointless question or if I'm just confusing a simple concept.

[MaxPower]

I see. The problem is that we record our results as sessions rather than in the actual units (hrs, hands).

If you wanted to get your hourly SD you should record your results each hour, but since that isn't practical we record sessions. It seems that lumping your 3 tables into one session would lower your SD, but I'd guess that whether you lump them of keep them seperate, the SD for the two methods should converge if you have enough sessions.

My assumption is that once your sample size gets large enough it shouldn't really matter whether you lump or not.

Where is Bruce Z? I'm sure he could answer this.

[Ben]

No idea what is correct, but as an additional data point:

I play 4 tables and I track my results as one session with 4x the time.

-Ben

[JTG51]

My assumption is that once your sample size gets large enough it shouldn't really matter whether you lump or not.

I was wondering about that. I have a zillion sessions recorded but I'm not curious enough to take the time to combine every two or three into one. I'm not very good with Excel but maybe I'll see if I can find a quick way to do that later.

[Jimbo]

[ QUOTE ]
I have a zillion sessions recorded

[/ QUOTE ]

If you have played a "zillion" sessions and have not added to nor subtracted from your bankroll I believe it is safe to assume that your ROR is near zero. /images/graemlins/smile.gif

Jimbo

[Big Trog]

If you play enough hands at one table to give you reasonable statistics, and keeping track of three separate tables doesn't overburden you, then I would keep the stats separate simply because the competition at each table may be different.

If you're playing against nine chumps at Table 1, nine average players at Table 2, and nine seasoned pros at Table 3, you RORs at each table would be very different, even if your SDs appear the same. The presence of one or more maniacs at a table would force the SD up since the pots (and resulting swings) would be larger.

OTOH, if you have no reason to believe that the quality or style of play differs substantially from table to table, then combining the stats from all tables is reasonable.

[Bozeman]

This is the beauty of SD: it doesn't matter how you do it. The more sessions you have the more accurate your SD will be, so breaking it by table will help slightly; HOWEVER, the numbers computed each way will converge for a reasonable (~20-100) number of sessions.

Craig

[Jerrod Ankenman]

[ QUOTE ]
This is the beauty of SD: it doesn't matter how you do it. The more sessions you have the more accurate your SD will be, so breaking it by table will help slightly; HOWEVER, the numbers computed each way will converge for a reasonable (~20-100) number of sessions.

Craig

[/ QUOTE ]

Unless you have some kind of significant self-selection for sessions, because you, say, don't like to get up stuck or have some strong aversion to giving back a significant win. Then your session results can actually contain a pattern of bias that suppresses standard deviation that would be recorded by per-hand data.

Jerrod

[Bozeman]

True, but even this non-normal distribution would have its average converge after a reasonable number of sessions (central limit theorem).

Craig