HUD Reliability: Number of Hands and Sample Sizes | #226

In this episode, I discuss the number of hands it takes for HUD reliability, as well as individual stat sample sizes for trusting the percentages.

In episode 225, I discussed how to learn the stats in your HUD and how to practice exploiting your opponents with it.

HUD Reliability

The HUD is a tool that when used properly helps you exploit your opponents. But it can easily be misused and lead to loss and ruin.

For example, you are facing a flop cbet and all you have is a Jack high hand. You look at Villain’s Flop Cbet stat and you see that it is 75%. Because they cbet so frequently, you decide to make a bluff raise. You raise it 2.5 X, then they come back over the top and shove all-in.

You ask yourself, “What went wrong?” You glance at their stats again and see that you only have 60 hands against them, and you notice their cbet of 75% is only three out of four opportunities. You done screwed up royally, so you fold your hand in disgust of your failed bluff.

You made the ultimate HUD–related mistake of not paying attention to the numbers behind the stat percentages.

I’ll dive into both the number of hands and sample sizes right after today’s Challenge:

Challenge (3:20)

Here’s my challenge to you for this episode:  Over your next 5 sessions, before you make a decision based on a HUD statistic, look at the number of hands you have on your opponent, and the sample size that the statistic is based on. The larger the sample size, the more reliable the statistic is.

Put this question on a sticky note on your computer, and answer it before every stat-related decision you make:

“Is this a reliable statistic?”

Tag and review all important hands.

Now it’s your turn to pull the trigger and do something positive for your poker game.

Number of Hands Played (4:50)

The first sign of a statistic’s reliability is the total number of hands you have on your opponent.

The numbers I’m about to give you are not set in stone, but they will help you be more confident in a player’s HUD stats.

0 – 100: this is simply a start. Most statistics are not that reliable yet. But there are some preflop stats that accumulate quickly and can be reliable at this point. These are VPIP, PFR, Limp, RFI and Call 2bet.

100 – 500: things are getting better now, and more of your stats are reliable especially the 5 I just mentioned. Any steal-related stat is also reliable at this point, and you’re flop-related stats like Flop Cbet and Fold to Flop Cbet are reliable as well. Beyond that, the turn and river stats are probably not reliable just yet.

500 – 1,000 hands: now were talking. This is a pretty good number of hands to be relatively confident in your pre-flop and most of your post-flop statistics.

1,000+ hands: this is great. With 1,000 hands, your opponent has played 166 rounds of 6-max poker or 111 rounds of full-ring poker. Most of your stats are reliable at this point and you can make lots of exploitative decisions based on them.

But the number of hands played is just your initial sense of reliability on a given stat. What is more crucial is the sample size for each statistic.

Statistical Sample Sizes (6:40)

Sample sizes are the number of opportunities that each statistic is taking into its calculation. That 75% cbet example from earlier, might seem like a lot. But, if it’s only 3 out of 4 opportunities, that is not as reliable as being 75/100 opportunities.

Anything can happen over 4 opportunities, but with 100 opportunities their frequencies and natural play style becomes evident.

When do sample sizes become reliable?

For the most part I think it is at 10 or more opportunities. Some stats require more, but let’s think about the 10 opportunities in terms of VPIP.

At one out of 10, or only 10% VPIP, this player could have been dealt nine poor hands out of the first 10 played.

We’ve all gone through streaks of 20, 30 or even 40 hands of folding every single hand because we kept getting dealt J4o, Q7s or 64o. So, seeing 1/10 doesn’t really tell us that this person is super Nitty. But, what if it’s 90% or 9/10 opportunities to VPIP? Well, they could’ve been dealt nine decent hands in a row, but it’s more likely that they are flop-loving player and found reason to put money in the pot with 9/10 hands.

VPIP happens with every single hand dealt, but some opportunities happen only occasionally. Let’s take Cbet Flop as an example. If you have the opportunity to cbet on the flop, that means you were the pre-flop raiser. If your Cbet Flop stat is only at 10%, or 1/10, then that means 9/10 times, after you showed pre-flop aggression, you chose NOT to continuation bet.

This is pretty indicative of a “flop honest” player.

If on the other hand, if your Cbet Flop stat is at 90%, or 9/10, then you are not so flop honest and you love cbetting on the flop.

But we must also consider the situation that led to your opportunity to cbet. If your Pre-flop Raise stat was at maybe 40%, this means you love to raise pre-flop so you get into many cbetting opportunities. If you couple this with the idea that you like to cbet bluff a lot, then we should give your cbets zero respect.

On the other hand, if your Preflop Raise is infrequent, like 10%, then your 90% Flop Cbet could be full of value bets.

But 10 opportunities aren’t enough to base decisions on (9:35)

I have to disagree with this. The extremes are where you can find good exploits, even with smaller sample sizes.

Here’s an example I love:

You are married and before you go golfing on Sunday, you have to ask your wife. Your friend Jimmy has to ask his wife as well. Over the last 10 Sundays, your wife said “yes” 9 times. So even though 10 is a rather small sample size, she has said yes 90% of those opportunities. So, when you ask her next Sunday if you can go golfing, you’re reasonably certain that she is going to say “yes”.

But your friend Jimmy’s wife told him “yes” only 1 out of 10 times. So, because she has said “yes” only 10% of the time, the next time he asks her, you and Jimmy can expect her to say “no” with reasonable certainty.

Look for the extremes

When somebody tells you that you can’t trust a Flop Cbet at 10% because it’s only 1/10 opportunities, they’re wrong. They could be correct if the percentage is somewhere in the middle like 4, 5 or 6 out of 10 opportunities. But, the closer the statistic lies to one of the extremes; 0, 1 or 2 out of 10, or 8, 9 or 10 out of 10, then you can use this number to exploit your opponent.

Beyond this, the bigger the sample size, the more reliable the statistic. Things start at 10 opportunities, between 10 and 30 is good, and anything 30+ is perfect.

You’re still looking to exploit stats at the extremes, though.

If your opponent has Folded to Cbets at 4/30 times, or 13%, it’s pretty safe to say they are NOT going to fold to your next cbet. You can use this information to cbet for value only and size your cbet bigger. And if you’re considering a bluff, maybe use only your best drawing hands and size it smaller if a bigger sizing won’t increase their folding frequency.

Let’s look at the other extreme: What if they’re folding 26/30 times, or 87%? Feel free to cbet bluff almost all the time, and if you’ve got a crazy value hand, you might be better off checking instead to allow them to bluff later or hit something worthy of giving you value.

Start your audiobook learning by picking up ‘Preflop Online Poker’ through Audible.com. Click the pic above to begin your free 30-day trial or to purchase the audiobook version if you’re already a member.

Start exploiting as soon as you get a read (15:45)

The thing about your HUD and statistics is you’re using it for reason: to exploit your opponents. If you don’t think 30 opportunities is enough to exploit somebody with, then go ahead and wait for a bigger sample size.

But, you’re not effectively using your HUD this way.

As soon as you get a read on an opponent, you should start exploiting them. You don’t have to wait until your HUD has 1,000 hands or 100 opportunities to know they fold to cbets a lot or love to 3bet.

Think about tournament players. They sit at a table and generally don’t have more than 30 to 50 on the other players. What, are they not supposed to make decisions based on HUD stats yet? If that’s the case, HUD’s would be useless for tournament players. All they would do instead is just play against the average opponent at the stakes.

You must utilize your HUD as soon as possible for exploitative decisions

For example, you only have 20 hands on Bob123. He’s limped in twice, called 4 open raises, hasn’t open-raised himself nor has he 3bet. It’s pretty easy to see that Bob123 is a passive player, and you should target them for value.

But what about Suzy789? In your first 20 hands with her, she’s VPIP’d 60%, raised 6 times and 3bet twice. Suzy789 loves to play aggressive poker, and you can assume she’s a LAG and play against her this way.

A lot of players don’t want to make assumptions because they don’t want to be wrong. But, poker is a game of incomplete information. Because you have incomplete information, you have to make assumptions. So, go ahead and start making your assumptions and plays based on them, and use your statistics (even with small sample sizes) to backup your decisions.

In Position versus Out of Position (18:00)

Most of these statistics in your HUD are the stat totals. But in your pop-ups, like I have within my Smart HUD, you might have the various statistics there for in position (IP) and out of position (OOP) play. These are useful even in “small” sample sizes.

Maybe you’re not comfortable exploiting a Total Fold to Flop Cbet of 16/20 hands, or 80%. I don’t know why you wouldn’t, but maybe you don’t think that’s enough folding to know they fold a lot versus cbets.

But, what if their Fold to Flop Cbet is 12/14 times when OOP, and 4/6 times when IP? Now, this tells me that they’re quick to fold when OOP, so if I’m IP and considering a bluff, you can be sure I’m pulling the trigger. But, if I’m OOP and they’ve only folded 4/6 times, I might only cbet bluff with the best drawing hands. Or, I might even make my OOP cbet bluff bigger to more likely get them to fold.

LIVE Players and Stats (19:10)

LIVE poker players can still use the idea of statistics to help make decisions.

When one of your opponents makes a play, ask yourself, “How often does this player do that?” (open-raise/call/bet/cbet/3bet/donk bet)

If you have any kind of a read on your opponent, then you might be able to answer this question. Your answer will guide you to the correct response.

I think practicing poker online while using a HUD is good for LIVE players because it gives you experience thinking about your opponent’s frequencies. The more thought you put into the frequencies of your opponents, and the more plays you make based on these frequencies, the more +EV decisions you’ll be making both online and LIVE.

Support the Show

There were quite a few more of ‘Profitable Bluff Cbets and Double-barrels Webinar’ purchasers who need a shout-out this week: Robert, Aleksey, John Sanford, Joel, Yury, Darren Layne, Jim Reid, Jerry, Jim Sterling, Wayne Altomare, Simon Grogan and Joris Schets.  Thanks for purchasing and for attending the LIVE webinar.  It was a ton of fun, and I hope you’re pulling off more successful bluff cbets and double-barrels than ever before.

Richard King, Jack Burkett purchased PokerTracker 4 through my affiliate link.  These two are soon to learn all the benefits that come with using a HUD (I sent them my Smart HUD) and having a database of hands to study.

Lance Abi Nader purchased the Smart HUD for PokerTracker 4.  The HUD comes with 16 stats and 6 custom popups along with specialized color coding for the most useful stats.  It’s perfect for getting the most from PokerTracker 4 and your online play.

Sky Matsuhashi