

So I've been trying out cash games lately, and so there's that statistic in my HUD ... kind of a strange little thing, eh?
Look at my graph for the Zoom Poker games I've been playing this morning: It makes it seem like I've really been having a rough go of the odds, but the first big dip at around hand #32 was AJs<QQ ... like I wasn't the favorite. Then the big step up to EV parity at around hand #163 was a flip AJs>77. The next big drop at hand #192 was again a flip TT<AQo. And then my QQ's ran into AA. Wouldn't you expect the graph to show me running right on EV? I guess I'll go post a question over at the PT website again ... the forum customer service is excellent! Will come back and share what they say 




I'm glad I took a minute to stop and put some time into trying to answer my own questions before asking somebody else for their time ... I guess I wasn't confused so much as I was surprised by the results, and the fact that they represent something different from what I thought they did.
So ... with flips, the expected value will always be around $0, because both people have about equal equity, right? Except since only one person can win, the winner will show as being 'above EV' for half the size of the pot (which would be about the size of the stack they put in), and the person who doesn't win will show as being 'below EV' for half the size of the pot. For some reason I thought the allin EV graph would show parity because the equity was equal And I guess the big jags look so huge because most of the hands I was playing were tiny, but get all excited and go nuts every time I get a top 10 hand and shove The really interesting hand is the AJs vs QQ, because again I guess I expected the EV graph to only show a disparity when a person takes a bad beat? But really it's a reflection of the fact that I had 32% pot equity, and so the actual $2.12 that I lost, in EV terms, is offset by my 32% share of the $4.32 total pot? ie EVsam(AJs vs QQ) = $2.12  0.32($4.32) = $0.73 expected loss or whatever (where my actual loss was $2.12) Hmmm ... will have to wait and see whether this 'allin EV' is all that meaningful a reflection of one's performance. Like because even a 2:1 has positive pot equity, doesn't that mean there's always going to be a tendency for the graphs to run EV even when one sort of 'deserves' to lose? Like look at had #120 ... it was AJo>ATs, where instead I was the 2:1 favorite ... but because I won, the differential isn't nearly so extreme. Just thought that was kind of interesting that it shows me running so under EV, when in fact (1) I haven't taken a bad beat, and (2) I've won 50% of flips. Like I thought it was a reflection of somebody winning longshots (if they're running above EV), or losing to bad beats (if they're running below EV). I guess it could potentially be a reflection of somebody losing more than their expected share of flips, like say 70% of flips when you'd expect to win at least half. But maybe you'd need to filter out hands or something to be sure? Hmmm ... will have to take a look around the net to see what other people have said about this statistic ... 




212 hands Sam. Like everything else in this game of ours, you need a FAR bigger sample size before you can start reading much into this stat.





Well, I know I'm right ... but if that's how it's come across to people, then I haven't explained my reasoning in a way that's making sense. So I failed in that regard
Will have to think of a way to better explain ... let me ponder it 




Revised





Oh, okay wait ... so I checked the site, and allin EV is calculated based on pot equity on the street in which the allin occurs. Even if the majority of the chips went in on a prior street.
Hence, with the AJo>ATs hand, the remainder of the money went in postflop, when I was already a 93% favorite, so that's the number that was used to calculate allin EV, not the 67% preflop. And so if two people have the exact same hands, but each holding the opposite's hole cards, and the favorite wins both times, then it should in fact wind up a wash. Sorry, just thinking out loud, trying figure out what specifically this statistic represents. I mean like it represents allin EV, but what specifically it represents in a meaningful way ... it's not necessarily what people might think if people are thinking that it means they're necessarily running bad if their graph shows them to be running under EV. Because every win runs above EV. And every loss runs below EV. So when you're losing a lot of hands, even when you're not getting any of them in when you're ahead (like say you keep getting/making monsters, but keep running into better), you're not technically getting it in when you're ahead, so you're not it's not like you're the favorite, and you're losing to a bad beat. But the graph'll still say you're running EV. Like it might just be a case of getting coolered, in which case the EV graph'll be a reflection of that. But with people who are learning, somebody who's making calls when they're behind ... that'll also show on the graph as EV over time, merely because a person keeps losing. And some of those losses might be the result of less than ideal choices. Like somebody who's a losing player could look at consistently EV graphs and just assume that they're constantly running bad. But the graph isn't necessarily differentiating between coolers and bad play. Like the assumption I think for most people in just looking at a EV graph is that it means a bad beat. But when you look more closely at the actual hand of AJo<QQ that wasn't the case at all ... it's really a reflection of me making a questionable choice, and then losing because I wasn't the favorite. Maybe this graph is more meaningful for pros, who know it's not them? For beginners ... maybe it's better to take this graph with a grain of salt? Still pondering ... 




Hi TrustySam 
I got PT3 on trial last week and played 5001 hands of 6max 2NL over 4 days  now that I want to review those hands, and not being particularly maths minded, I am getting obsessed by what all the different stats it throws out mean. I saw Andy 'ahar010' put up some PT3 graphs in one of his training sessions with/without the +EV line. He seemed to be saying that when your actual results line is above/below the EV line it indicates the level of variance you have experienced, and over time the lines should get closer. As Darkman says, you need a much bigger sample for the stat to be meaningful enough to be able to interpret it. However, if I understand correctly you are not trying to interpret what it means for reviewing your small quantity of play, but just using the limited and individual instances you have experienced so far to work out how the stat is created and what it is designed to show, so that you can properly understand it when you have a big enough sample for it to become meaningful. This seems a sensible approach when faced with any new statistic. You have certainly helped me understand what is going on with this stat, taking the flip situations has clarified a lot. Good luck Ed from Edinburgh  EdinFreeMan 




I wish I could explain me as well as you're able to explain me Ed
That's so so spot on ... I'm in that 'how does this work' phase, so that when all these little individual stats get aggregated, I'll be able to tell if the group as a whole truly represents what it's designed to represent. Like, basically what you said already, but much more clearly 




Okay, so like ... some of the results from the 'allin EV' graph made me go, huh? So I was interested in exploring the formula, and the data a little more to see if the graph with a larger volume of data could nevertheless be reliable, meaningful, and truly represent what it's designed to represent.
Or is the margin of error so great, that it ony has the appearance of being meaningful and representative of what it claims to represent. And darkman brought up the fact that in statistics, you generally need large samples in order for the data to be reliable. And oftentimes that's true. With high variance data where there's a wide variety of outcomes, larger sample sizes are needed to approximate a standard normal distribution than for smaller ones. Other times, the sample sizes needed to generate an accurate prediction are shockingly small. Like for example the polls used to predict who's likely to be elected President in the US, for a country of over 300 million people, with tens of millions of people who actually cast votes, never needs more than 5,000 people polled to get a reliable prediction within the margin of error. Like in that case pollsters have found a way to profile and target specific individuals who are representative of the whole, such that it's not a truly random sample, and so far fewer responses are needed for them to get a reliable snapshot of the whole? The thing with this 'allin EV' statistic, is that it's not just my sample that's small  this 'allin EV' statistic only gets counted when at least one of the players goes all in, in a cash game, and it happens before the flop. For me, out of 500 hands played at 2nl zoom, that was only about 6 or 7 hands. For somebody who's played a hundred thousand hands, the number of hands that would fall into that category, while much larger in the aggregate than my 7, might still tend to be roughly proportional, and if so would represent a mere thousand hands out of a hundred thousand. And so isn't there then a question of whether that 1% of hands (at cash games, that are allin, prior to the river) can ever truly be representative of the greater whole? For anybody? Like some of the hands that aren't getting included in the allin EV statistic are hands like the one I put in HA, JTo>KK. Because the allin happened on the river after I was beat. Other hands that wouldn't be included in the statistic would be all hands won without showdown. If somebody with AKo gets called by somebody with JK, and the flop has a J and the caller leads out and the AK folds, that doesn't get documented as a bad beat. Like, even after 100,000 hands, that's always going to be the case? So there's that question of like, how accurately and reliably could this statistic truly represent EV, if it's missing so many hands? If people go to showdown more often ... like maybe in PLO that's more common, maybe there the graph would be relying on more data? Hmmm ... 




Hi again TrustySam,
here is an example and what I think I can take from this statistic. I took a couple of beats holding KK today, which had been my most profitable hand over the 6,800odd hands of 6max I played this week. I filtered the graph just to show today's KK hands  whcih included these two.. This shows up on my actual winnings versus EV on this graph What I take from this, in the same way we shouldn't be results oriented, is that I don't need to worry about the downswing here as it was not due to bad decisions, but the luck factor, and eventually this will even out. My overall graph for several thousand hands shows I have done better than ev  so I suppose I have to resign myself to the fact that some of that profit was not due to my skills and I am showing better results than I should really be, (so not to get too cocky). As you mention, this can only be measured when hands go to showdown or are all in (I'm not sure which  I think I have always been all in on these type of hands). btw  whilst typing this I got dealt KK again and won, so the luck is swinging back. Even with only 43 KK hands the PT3 stats are showing my win rate with KK as 88% so lets hope they keep dealing me those cards. Good luck Ed from Edinburgh  EdinFreeMan 




Oooh ... I'm squinting my eyes and cringing at the backtoback KK<99 and KK<88 Ed ... ouch!!
On the other hand, thx for sharing those stats and the hands ... that graph is really cool, like in terms of trying to absorb this whole allin EV thing. So now we've seen almost all the kinds of situations where the 'allin EV' graph differs from the actual wins/losses graph: 1. Ed's bad beat  AIEV goes up, and the actual results graph goes down (Ed was favorited to win, but lost) 2. My coin flip at hand #192  AIEV stays constant, and the actual results graph went down (I had a 50/50 shot of winning, and lost) 3. My hand #32 where I was in bad shape against a monster  AIEV goes down, and the actual results graph goes down, but to a larger degree (I was expected to lose, and did) And then the opposite will hold true for wins: 1. My hand at #121 where I was ahead  AIEV goes up, and the actual results graph goes up, but to a larger degree (I was expect to win, and did) 2. My coin flip at hand #163  AIEV stays constant, and the actual results graph went up (I had a 50/50 shot of winning, and won) 3. 6th possible type of hand  (no graph for this yet) we suckout a win from somebody else who was the favorite when the chips when allin (We were expected to lose, but win) Like, most of the time for NL, probably for most people the two graphs ought to have similar shapes. Like with Ed's from hands #810, and mine from hands #130, #33120, #122162, #165190, #193208 ... those are hands that the graph doesn't have data for because it wasn't an allin before the river. But even though for most people these graphs rely on so few hands out of so many (even for larger sample sizes), maybe the could potentially be useful in the sense that the hands where they do count (#32, #121, #163, #192), they could possibly make the graph more reliable I guess if they represent a greater proportion of one's wins and losses (because they're big hands for big pots)? Like perhaps if somebody plays really tight, but then plays for stacks when they do play, because it's almost always a monster? Or else perhaps for a game like PLO where there's a lot of hands that get played for stacks? 




Well this might be fun ... I found a couple of pro graphs online, and knowing what we know now about the way the allin EV graph is calculated, maybe we can tell certain things from their graphs.
1. Phil Galfond: Step1. So the thing we want to look for is the spots where the two graphs turn different, and where the two graphs look the same. That tells us how much data was actually used to make up the graph. So for Phil, we see some noticeable differential at #140, #800900, #900950, #9501000, #1050, etc, etc.  at #140 the AIEV graph is showing at about constant, while Phil's graph is going down  that means he was in a 50/50 spot when the money went in and he lost  #800900, again the AIEV graph stays constant, while Phil's graph is going down  so that would seem to show that Phil's loosing a greater than expected amount of 50/50 spots, and so he's losing money when he should be breaking even  #950100  the AIEV graph goes down and his goes up  so there he wound up winning something when he wasn't the favorite  #1050  the graph goes up, and his goes up a lot  so he won a couple of big pot when he was the weak favorite in each So except for those 4 spots, all the other hands from #11000 didn't really factor into the graph in a noticeable way  so for those, like a person would have to check their hands manually to see whether the hands went to showdown and if they were a favorite and lost or what. (He took more bad beats (AIEV goes up, Phil's graph goes down), at around #1750, #1900, #2300, #2700. Went on a miniheater from #27002900. Then went back to getting bad beat in a big, big way from #29003100 (but not #3100 to 3800 even though the graph is still going down  because the allin EV graph is also going down  that means he was just playing big pots and losing ... like he might have been getting coolered, QQJJ running into AAKK or whatever ...). #38003900, he's losing more 50/50's again. 2. Mathew Didlick (chipstar) From mathew's graph  the fact that the two graphs so closely parallel one another seems to show that he doesn't like going allin until either he has the what looks to be the unasailable nuts, or else he plans his raises for them to be allin by the river (so that he can pull the plug if things take a turn, or after he's more sure he's got the best hand). Like obviously he's experienced as much variance as everybody else  but it'll never show on his graph if he plays cautiously and tries to not leave a lot up to chance? So that's kind of an interesting graph I thought 




Can't say I know too much about all this stuff. Never used tracking software unfortunately. Interesting though.
Just want to point out that 50/50 flips are EV because of rake. That said, folding instead of flipping is even more EV if you already put some money into the pot (like folding after someone 4bets your 3bet). Also, a couple hundred hands is not enough to come to any conclusions. The results at this point are meaningless as you can't distinguish luck and skill factors. At this point all you can do is review your plays and estimate to the best of your ability your performance so far. 




That's a great point about flips and the rake Rocker ... hadn't thought of that





I guess if I feel like this allin EV graph is less reliable than the productmarketers have claimed, then the onus is on me to make the case, since it's incredibly popular and has been accepted by most as meaningful.
So I'll have to keep trying because my points about statistical reliability and sample sizes (large isn't necessarily reliable if it's an aggregate of unreliable individual statistics, small isn't necessarily unreliable if the individual statistics are representative of the whole) don't seem to be connecting with people? Will keep trying 




I guess the only type of graph that runs the danger of getting misinterpreted might be this type:
I just picked it off the internet, not sure who the person is, but the things to note are that: 1. Both graphs are consistently trending downward, and 2. The actual results graph is consistently trending downwards to a larger degree. That shows that the person is playing a lot of allins, and that they're consistently getting it in with the worst of it. So if they're profitable, then they're just getting coolered and it was just a bad day. If they're not profitable, and if more often than not their graphs take this shape, that means they're not picking good spots. Like all losses that were allin before the river, even those where we're behind will show as running below EV, simply because we always have some equity in the pot. So that's why the actual graph is running below the EV ... but that doesn't necessarily mean that person's in a downswing, eh? I think that's the only person that runs the danger of getting the wrong message out of these types of graphs. If somebody's profitable, then they're doing great! And it's all good 


