Those are accurate as far as I know. 250 buy-ins???????????? That has to be the biggest I've ever heard!!! You run baaaaaaaaaaaaaaaaad

I'm trying to get my head around the variance in SNGs and using the red line as some sort of indicator. I understand what the red line means, but what factors could cause the red line to go down?

Getting it in bad I know is one reason, but I am trying to work out what else.

I haven't been playing 6 and 9 man SNGs for long. I played MTTs successfully previously and I played micro 18 mans really successfully before that.

I have adjusted my game considerably since moving to SNGs, but (although I am running like ****) I am curious as to why my red line has been trending slightly downwards.

How big is your sample.

Reasons for redline can be getting it in bad, losing a lot of chips wo a showdown, being carddead, running into tops of villains ranges, or if youre a losing player. The redline isnt completely luck neutral, just moreso than the greenline,

I'm pretty much running similarly at the moment. And it's not your playing style. It's called "variance".

About 700 (which I know is not a lot). It's just that I've never taken on a game before and not beaten it; very frustrating.

I disagree with this. Just think of how much worse your expectation would be in the games if you made all plays based on 65/35 structure(especially bubble calling, which is way different in a lot of situations).

Played a lot (prob 100k or somthing) games on this site, and the red and green lines converge as they seem to for most of the other regs also.

running very good this month, what are the chances of this % wise?

[IMG]http://imageshack.us/photo/my-images/829/augustsng******[/IMG]

thanks in advance

cant make imageshack work, but im 5.7% above ev over 2126 games in 6max sng with 70/30 payout

Over a sample of 20.000 SNG's as well? One would think that variance plays a much smaller role over a sample like that, hence my question.

In terms of buyins you are more likely to be a given number away from expectation after a large sample than a small sample. The luck adjustment should only be expected to converge to zero (and so the red line approach the green line) in terms of profit per game, or ROI. 250 buyins over 20k games is only a 1.25% difference in ROI which is slightly unfortunate over than many games, but it can be worse.

Yes, your expectation will be worse, but that's not the same as using it for this (ie: luck will still be centred around zero and the estimates of the $equity you are up/down will just be off by a small fraction - the fact that ICM doesn't account for skill could mean that it's off by much more yet it still works reasonably well).

Juk

I'm not sure I understand that response at all. Basically, I'm saying that if you are playing these correctly, your expectation is actually going to be substantially higher than the red line currently has it when its calculated incorrectly for a 65/35 payout(because the games are actually 70/30 payout). Your green line would be the same, because its calculated correctly based on the real results(70/30 payout).

Your play should have no effect on the calculation.

You can ignore the green line completely as the red line is simply the green line plus the EV_diff values (ie: the distance between the red and green line = EV_diff) .

If you were to plot the EV_diff lines for the correct 70/30 payout and the 65/35 then you'd see there isn't that much difference: both would still be centred around zero and have approximately the same shape and values (yes they would be different/wrong but not by a huge amount). In the long run it should still converge to your "true" ROI, but it'll just take a bit longer due to the scaling errors the slightly erroneous payout structure adds...

Even if you used a totally erroneous payout structure and/or decided to add random offsets to the EV_diff values, your red-line ROI would still converge eventually.

Juk

I think my understanding of what exactly is graphed in HEM results is off perhaps. The graphed green line SEEMS to not be the aggregate of the EV results from the improper 65/35 calcs, but rather the REAL results of your sit n gos(70/30 payout structure). The red line seems to be the net aggregate of the improper $ev calcs(65/35). If all of the ev diff values are calculated wrong, then obviously the green line + the ev_diff lines is going to be off as well, and not necessarily ever converge. What am I understanding wrong?

A player playing appropriately for the 70/30 structure will have an artificially low red line because he is being punished rather than rewarded by the ev calc in the long run every time he appropriately deviates from 65/35 strategy.

NP and I'm not trying to cause problems here - just trying to explain

If you think about the effect it'll have on the EV_diff calculation then it really won't be that big: it'll just have a small scaling effect on each value calculated for each all-in (ie: the value's sign will still always be correct but it'll be a bit more or less than they should be). The bigger the error in the payout structure, the bigger the scaling effect will be.

Also, a good amount of the largest "swings" that make up the red line happen when you are heads up (where the payout isn't important).

How somebody plays has no effect on the way the EV_diff values are calculated: they could be the biggest nit in the world or go all-in every hand and it would still work out fine... The calculation just needs to know about the parameters of the all-in (eg: the %s and the stacks). For the same reason their play has no effect on whether the sum of EV_diff values for their games is positive or negative (which equates to the red line being above the green line or vice versa).

This is true for any payout structure you choose; however wrong it is compared to what they are actually playing.


If it still doesn't make sense then hopefully somebody else will chime in here as I'm probably not explaining my thoughts on this very well

Juk

I havn't been trying to say the way people play affects how the ev_diff values are calculated. I'm just saying that the resulting calcs based on an improper payout structure would on average result in a net loss over time for the player playing the appropriate strategy for the correct structure(versus a proper calculation). I still don't see quite why this wouldn't be the case. *shrug* Hopefully HEM bothers fixing it soon so we can see what the difference ends up being.

.

These two sentences are a contradiction though?

The EV of an action (ie: what SNGWiz tells you), profitable plays, "playing the appropriate strategy", etc, all have no place at all in the discussion about all-in EV... But I don't know how else to explain it other than repeat what I've said already in the previous posts?

Yeah, but there is an equal chance you will be running over EV HU so all this is doing is adding some noise to the EV_diff value (like I said in the post above) and thus meaning your ROI converges slightly slower... All of the other non-all-in luck factors work like this.

Juk

Here is an example of how slightly wrong payouts effect the results:



These are just my last 500 or so games (don't have any more handy atm), but as you can see the use of a slightly wrong payout structure makes very very little difference.

Hopefully somebody else will better explain the reason why your play has no effect on all-in EV calculations. Although not in the context of tournaments, this thread might help.

Juk

Again I think you are misunderstanding what we're saying in terms of the strategy to HEM red-line relationship, but perhaps we're just confused. Again, nowhere are we saying that how you play changes the how the ev is calced in any way. The change in payout structure DOES change it obviously, as your graph shows, and I strongly believe the RESULT of that change is typically going to RESULT in a deflated red line on average for an optimal player. The post you linked to doesn't seem to have anything that even remotely addresses our logic behind that. The fact that you linked that leads me to believe you don't understand the logic of my argument, but I unfortunately can't think of a way to make it clearer. While how you play doesn't affect an ev_diff calc, an improperly calculated and biased ev_diff calc could theoretically impact different play strategies differently is all we're saying(if one leads to a different finish distribution than the other). Ultimately, I also think we just disagree on what is "very very little difference". These are super turbos and people play 5000-10000 of them in a single month. The little differences compound into big differences in a hurry.

The difference should be very slight, as suggested in Juks graph. We are talking about very slight dialations and constrictions of ranges going from 65/35 to 70/30.

Well what are you saying?

Going back to the original two posts that sparked discussion:
I've tried to explain why I thought what I did, posted a graph that shows what I expected and tried to explain that "making plays EV" has nothing to do with all-in EV, so really I'm at a loss now...

Juk

EDIT: You edited your the post as I was quoting it (and keep changing it)

Or in JD4WG91's case, 15000-20000

I think what Juk is trying to explain is that the redline calculates your expected winnings after the fact, meaning that even though there are slight differences in terms of what hands you would push and fold based on the payouts, the redline only takes results into account, and is therefore accurate.

Amirite?

Yes, agreed.

I just don't get where you are getting this idea from? It makes no sense if you understand that your play cannot effect the calculations.

and because of this: either it will make a deflated red line for all players or none, and I have tried to explain why it's none: because basically it's just introducing small scaling errors (which you can actually see an example of if you look at the graphs).

But these differences are dwarfed by the other luck factors and the fact you are using the ICM model to predict your future equity gains.

The ratio of the difference between the red line and the invalid payout lines (and thus the error in ROI prediction) will actually converge towards zero as you get a bigger sample (the absolute difference won't but that's not important here). All you are doing is adding some extra bit of randomness, but what you are adding is really really very tiny compared to the other noise introduced from other areas.

I've tried to justify why I think what I do so feel free to do the same

Juk