PDA

View Full Version : Standard deviation



Quick2kill
08-18-2009, 02:33 PM
I can't find this explained in the manual.

Obviously this is the standard deviation on the win rate, but more than this is not clear.

Firstly what are the units it is in? Looks like it's just in bb/hand, but I'm a bit suspicious from my numbers that it's actually in bb/100 hands (bb = big blinds) or it could even be big bets.

Secondly the standard deviation on win rate varies depending on sample size. You obviously calculate it by dividing the total number of hands into blocks containg y hands then calculate the average variance over the whole sample for a block of that size. What is y? I assume 100 hands? Or is it ten or even just 1 hand? Hope this is clear.

In case the above is not clear I am talking about sample variance (obv take the square root to get standard deviation) as described in e.g. http://mathworld.wolfram.com/Variance.html, which I assume is what HEM is calculating.

netsrak
08-18-2009, 03:47 PM
I'll forward this to Fabio for an answer.

Rvg72
08-18-2009, 04:26 PM
We have two different stats, one for standard deviation of big blinds and one for big bets. This is how it is calculated:

10*stddev(ph.NetAmountWon/GT.BigBlind)

This is done for every hand and then divided by the total number of hands.

Roy

Quick2kill
08-18-2009, 04:50 PM
We have two different stats, one for standard deviation of big blinds and one for big bets.
Ah so you do, missed that sorry.



This is how it is calculated:

10*stddev(ph.NetAmountWon/GT.BigBlind)

This is done for every hand and then divided by the total number of hands.

Roy
I'm not following here. what does the function (method? this looks like c++ to me :)) stddev actually do? the factor ten seems somewhat random.

Your last comment seems to imply you are calculating the standard deviation on win rate for 1 hand. So I presume you are doing the following: sigma^2_i = (x_i-\bar{x})^2 where x_i is the net number of bb won, i.e. the input you show for stddev, in hand i and \bar{x} is the mean number of big blind one per hand over the whole sample. Then you get the average sigma over the whole sample. Is this right?

Quick2kill
08-20-2009, 06:31 AM
bump. I was really hoping for a follow up answer on this. I really want to use the standard deviation from my stats, but I can't do that until I know what it actually means, i.e is it calculated from the average variance on win rate for 1 hand or the average variance over 100 hands. Both can be expressed in bb per hand, which appears to be the answer to the first half of my question, but the variance, and hence standard deviation will be different so it really matters.

Rvg72
08-20-2009, 07:09 PM
We're actually using postgresql's stddev function and running it across the set of hands. The 10* is done to make it a standard deviation for bb/100