TT vs Fish

Why would we consider folding to this silly line? They can just show up here with 3s3c or some random AJ.

Mmm snapping

Also calling. We beat both bluffs and his merging mistakes.

Ready to make a note of both his donking strength and halfpot rvr strenght

So one thing I found (over a limited sample) is that D-XC-D can be underbluffed, so I don't think the turn check necessarily caps their range.

But with his river size I'm also calling.

I do have good data for Donk-Bet-Bet/XC-Donk-Bet and XC-XC-Donk and the only one that is underbluffed is XC-XC-Donk100+

EVERY other line is overbluffed (by a lot) even with overbets.

In these mass data analysis on over and underbluffed lines, has anyone ever analyzed whether individuals (in this case donk betters) are consistent or if they are generally marginally balanced?

I mean, the guy in this instance is a single data point and may be bluffing or valuebetting in this current example. How much more likely is it, that he is also bluffing/valuebetting the next time he takes this line? How valuable exactly is our note.

Not sure what you mean exactly but if they overbluff or underbluff then they aren't balanced by definition.

You won't get into this spot vs this exact line vs this specific person enough to make taking a note very valuable. You just defer to population data as a default.

Let's say MDA says this is a bluff 50% of the time.

We don't know if that means that players who take this line have 50% bluffs on an individual level or if half the players are always bluffing and the other half not. Likely it's somewhere in between, but how much I wonder.

I always note down what people donk with and they seem pretty consistent

Yeah you won't know that information, you just take the GTO number and compare it to the population so if they B100 we know they need to have 33% bluffs so if it's higher than 33weak we call.

I'd also caution against using always/never or even thinking like that since there's too many unknowns in poker to ever be 100% about a specific individuals tendencies.

Maybe I am not making my thoughts super clear. Maybe I am hindered by not being a native speaker.

Say we look at a flop donk bet 33% analysis and say MDA says like 10% strong, 40% mergy and 50% weak.

We get to showdown and villain tables a flopped set. In this case, I speculate that this player's overall db33 range is likely way stronger than populations', since the data will include donkbets from people who never donk 33% with nuts. It is like an extra data point similar to a CO flat, that might be useful in the future. The question is just how useful.

Oh okay so I think that would just be a statistics question using a binomial calculator. I'm not great at statistics but let's say he was 10% strong and you saw a set 5 times in a row. You would use the binomial calculator to find out the odds of him being truly 10% strong if he show's up with a set 5 times in a row.

So I think it would go like this:

Number of trials: 5
Success probability: .10 (10%)
End Point: 0

So X or 10% strong would be closer to 41% after seeing a low probability event happen 5 times in a row.

If you just saw the set once in 1 trial, you shouldn't change your strategy but if you see it twice in 2 trials you can update your confidence levels to this.

Now the 10% changes to 19%.

Cool question and please someone double check my math here.

I think your methodology is correct if we want to scientifically assess the likelyhood of villain falling within the theoretical parameters (10/90) based on our observations. For your listed numbers, I THINK you want to use 0.9 for the success rate. It seems to me like you have calculated the chance of not seeing a set within 5 hands vs the chance of seeing at least 1 set.

I meant something different however, kind of like marking a player as a possible fish based on a CO/SB flat or an open limp. Allow me to try and make a simplified example. Let's look at a random non-GTO action (like donking the above board), and not care about mergyness. Our MDA-pool has 5 players (simplified obviously), and they all play donkbet 33% as follows:

1: 100weak, frequency 10/10 (relative frequency ; arbitrary scale)
2: 95 weak, frequency 2/10
3: 90 weak, freq 4/10
4: 0 weak, freq 1/10
5: 70 weak, freq 3/10

See, if we have observed a donkbet from a player with a set in this dataset, player 1 should be completly excluded because he never donks strong. In this dataset, the chance of the player being strong on a donkbet in the furture would increase from 12% to 24% because player 1 can be excluded. Furthermore, there is a 42% chance that this player is player 4, and an 80% chance that this player is either player 4 or 5. This gives cause to even further caution dealing with donks from this player in the future.

I am aware that this is way simplified, and that such strong effect will not be evident in a large dataset - however, if some players NEVER donk strong and other NEVER donk weak, it should have some affects even with a single data point.

I think I did it correctly but waiting for people better at math than me to confirm.

As for your question, I don't know enough about statistics to confirm or refute what you are saying but I don't think it has a ton of real world applicability beyond general curiousity.

But if you are determined to get a satisfactory answer you should PM someone like Tombos21 as he is an expert in questions like this.

doc MDA is no different than any other exploit in poker in a sense; it's just a general guide to how humans play. I think what you're referrring to is the inherent complexity of fish lines where they can be unpredictable to the point of uncategorization.

I think what mitigates that is their lines are usually so extreme/off piste we just make up for that margin of error by all the times we're right. Fish in general don't know how to play range v range poker and most exploit play (MDA or otherwise) just takes advantage of that generality. The margin of error is precalculated and 'smoothed out' in the exploit. Just like opening trash from SB.

That's not to say we never hit outliers. It's just that by and large humans are predictable en masse, and particularly if they're mainly playing 'feels' poker.

Let's say drivers use their turn signals on average 98% of the time they make a turn. You observe a driver, who doesn't use his turn signal on a turn. What is the chance, that this driver won't use his turn signal on the next turn?

I argue that 98% is wrong. I think, that the likelyhood of them not using it next time has increased, even though it is based on a single data point. How much it has increased depends on how many people extremely rarely foreget to use it and how many people very often forgets. I argue that it basically depends on the shape of the normal distribution, and that this normal distribution would be interesting to know in rarely used poker lines.

At a certain point MDA for rare lines falls apart though.

1) they don't happen often enough with the same people to ever find out what they would do consistently on certain extremely rare runouts against an ever-changing roster of different opponents and 2) strategies are always in flux anyway. This is evern more true for fish. Mr 2% turn signal guy is going to read a book at some point and realise people keep crashing into him unless he tells them he's turning. Suddenly all our data and calcs are useless in a heartbeat for that specific outlier.

I'm only just starting using MDA, I'm a noob. I'm looking through my DB and the patterns are def interesting and useful. But even the broadest takeaways (eg. fish overfold turn OB) are usually fragile and context dependent. There are showdown biases, game biases, individual biases. Drunk guy biases. All these variables spoil the broth. To quantify all that you'd need computers the size of this:

Also AFAICT you can actually have too much data for MDA. E.g. if you surveyed the entire planet for their favourite colour and you took that answer and applied it to every continent. The game (and sapiens) are too dynamic to solve at the extreme ends, and any algo to account for those wilder fluctuations is just god mode.