Since there are a million posts a day on variance and since I don't contribute all that much to society, scratch that I mean this forum, here is a post on variance. I know there are quite a few very good posts on variance out there but none of them give any actual experienced data.

Hopefully this will help shed some light on all of the "how come I ran at 21% for 1000 games and now I cannot win anything."

These are all 27s, total roi 6%. If you have a better ROI you can expect the negative numbers to not be as negative and the positive numbers to be higher I would assume. The opposite is also obv true if your roi is lower.

I am also finding out that 12k is really not all that big of a sample, Ill make another post when I hit 25k.

I think the main thing to take away from this is even at 1k game samples the variance can still be quite drastic and it is not until 2k game samples that the rois are closer to the cumalitive roi.

Feel free to add in your samples.

Breakdown of 500 game samples

1-500 5.8 38.4
501 -1000 10.5 39.6
1001 - 1500 7.3 36.8
1501 - 2000 -5.7 34.4
2001 - 2500 6.0 37.8
2501 - 3000 3.3 29.2
3001 - 3500 27.2 45.8
3501 - 4000 18.0 42.8
4001 - 4500 -5.0 36.8
4501 - 5000 5.5 37.8
5001 - 5500 3.0 37.4
5501 - 6000 4.5 38.4
6001 - 6500 -2.3 32.8
6501 - 7000 13.7 41.2
7001 - 7500 -7.8 32.2
7501 - 8000 11.0 40.2
8001 - 8500 8.2 39.2
8501 - 9000 -1.0 37.2
9001 - 9500 7.7 37.8
9501 - 10000 2.0 37.4
10001 - 10500 1.3 36.2
10501 - 11000 7.2 38.2
11001 - 11500 12.2 40.4
11501 - 11855 10.8 41.4

Breakdown of 1k game samples

1-1000 8.2 39.0
1001 - 2000 0.8 35.6
2001 - 3000 4.7 38.5
3001 - 4000 22.6 44.3
4001 - 5000 0.3 37.3
5001 - 6000 3.8 37.9
6001 - 7000 5.7 38.7
7001 - 8000 1.6 36.2
8001 - 9000 3.6 38.2
9001 - 10000 4.8 37.6
10001 - 11000 4.3 37.2
11001 - 12000 11.6 40.8

Breakdown per 2k game samples

1 - 2000 4.5 37.3
2001 - 4000 13.6 41.4
4001 - 6000 2.0 37.6
6001 - 8000 3.6 37.5
8001 - 10000 4.2 37.9
10001 - 12000 7.6 38.9

Edit: sorry about the spacing but every time I try to make more space it just bunches it back together.

Cool post, thanks. Do you still think 7% is max?

always thought the mainly suggested 1k sample was a bit too small to judge...

So, to put all this numbers into perspective:
How many tables do you play?

I think a really good player can hit 10% over the long run but I think thats max at least right now unless you are playing only a few tables. Most likely a player that can hit 10% though is moving up anyways.

15-22 depending on various factors like time of day, how much interwebs browsing I want to do during etc.

Thanks.

How do the number of tables correlate to anything given the data provided?

The more tables you play the lower ROI I would assume.

like... err... alot.

There are guys sustaining rois above 15%, but they play 2 tables max. and some have 1.5% roi, but they 40-table.

So, knowing if he only plays 8 tables or 40 is kinda important.

Very nice post, which has made me feel weirdly sad.

jurrasstoil, the sample size needed is correlated to ROI somewhat.

what's your biggest downswing during these 12k 27s?
more than 40 buyins?

Pifhluk, thanks for this. Very nice, this is much better than any theory about variance.

Couple of questions:

What do you think about your game? Are you getting better? Or is it basically still the same? Do you tilt often?

Then, how long did it take you to play those 12k games?

Do you think that players are getting better? More regs per table than it used to be?

Thanks.

+1

Do you think programs like sng luck-analyzer, which take into account all situations of all-in results, can give a better approximation of our expected ROI on a same sample? I mean, of course it should, because all-ins are a part of the randomness (english?) implied in SNG results, but what's your idea of by how much using the luck analyzer expected ROI could reduce variance in the numbers?

You have a large sample, so if you tried to use the luck analyzer on each of your 500 and 1000 games samples, we could see what the standard deviation from your global ROI is. This could give us precious data, and we could be able to have a more accurate idea of our true ROI with a smaller sample than it is the case right now, I guess.

Of course, it's just an idea, this is some work to do, but if someday you have time and motivation to do that, it would be very useful IMO (at least for me).

Thanks for the great post BTW

Variance, variance:

k4b4l 446 -$3.06 $27 -9% -$1,364 - PokerStars 9/1/2008 9/19/2008

It looks like 50 but I know I have had more than that. I can recall my account dropping down to ~700 and I almost always play the 27s with a 3k bankroll so that would be 85.

I think my game is pretty static. For 3 days of the week I play 150-200 tournies and when you play that much in a day it is hard to stay focused the entire time. I don't study all that much either mainly because there are just so many other fun things to do with free time.

I wouldn't say I tilt very often but I do find myself getting bored and not thinking as much as one should.

Those are from 9/18/07 - 7/31/08 I didn't include August through now because I have been trying out a bunch of different stuff and its not working out all that well. Also this is probably not all of the tournies played during that time, I know I lost some hhs once w/out loading them into tm and if you close out of a tourney too quickly sometimes tm wont get the full summary or something.

This year I have bounced around a few levels but I think I am going to stay at the 27s for a while so the next 12k should come alot quicker.

I think the overall skill level of regulars (ppl that play quite a bit) stays close to the same. Not too many ppl stay at this level for a long time, its either move up or back down. It comes and goes, sometimes you can't find a table without 3/4 regulars other times you can get tables with none.

I think the average player (donk, fish whatever) is getting slightly better at calling lighter / pushing lighter. I would attribute this to not as much new blood coming into the poker economy so the people that are playing have a little more experience of you shoving into them 100 times. Take this with a grain of salt however because the play is still really bad.

Someone better at math should probably answer this. Personally I don't use the luck analyzer, if you are running bad you know you are running bad because your good hands keep getting cracked, you keep getting 2/3 outered and you can't win a 50/50. I mean I don't really need software to tell me that and the software is not going to change the outcome. All it does is reaffirm what you already know.

Well I don't have PT anymore since my last computer quit and I think the luck analyzer needs an ms database. I do have HEM but I think that uses something else not ms access?

Hi there,

Thanks for the data. Can you explain the figures in the 2nd and 3rd column? I assume 1st column means no of sngs. 2nd column is cumulative ROI? What about 3rd?

professor76

1st is the sng sample. 2nd is not cumulative ROI but ROI for that particular sample. 3rd is ITM for that sample.

sorry about the spacing, I tried to go back and put some /s in but it said contact an admin to edit post and I was just too lazy.

This is a great post and great thread.

One thing that may be player dependent that I have discovered recently is that I find the more tables I play, the better I am able to play in a lot of situations. When I 9 table, I used to try to force things in certain situations. I shoved too wide in certain spots and just tried to make things happen. But since I've been 20 tabling (now granted, I don't have a nearly large enough sample size yet since I've started 20 tabling), I find that my play has improved, and I've kept things simple as opposed to falling victim to the dreaded FPS. Now maybe variance will catch up with me soon when I get a larger sample size, but this could also be something player dependent. I think it's possible to have a higher ROI playing more tables as opposed to fewer.

Also, what is your average ITM%?

Ok thanks! Now I got it. Thought 3rd column was profit or buy-ins earned, though the figures didn't quite tally. Now it makes sense. :-)

Thanks for sharing this data! This is really something that I've been looking for.

I especially liked this one.

No, luck analyzer only needs basic HH's to work properly...I understand your point, though, but still I think it would be interesting if you have the will to do it. Otherwise I will wait until I have a much bigger sample myself...

3001 - 4000 22.6 44.3

^^That musta felt great!