Donk Bet Raising Success Rate...

[Talking Poker]

This topic came up in , but I think it's worthy of it's own thread. You can go read the details there, but here is a quick summary in a nutshell:

A "donk bet" is when someone leads into the preflop raiser on the flop. Generally, this person either limps in preflop and then calls a raise, or they call a raise from the blinds, but they are always out of position against the preflop raiser. Then they lead out at the flop.

In my experience as the preflop raiser, I've had great (perceived?) success in raising in this situation. In the other thread, I guessed that with a decent raise, I take the pot down right there more than 50% of the time, making this a +EV play, assuming I'm not overbetting the pot with my raise (and I'm not).

Other people have had very different experiences with raising donk bets:

-meanness says he wins the pot with that raise about 5% of the time.
-sjay guesses it's 15-20% for him.
-Reel Deal guesses it's close to 50% for him.
-Robbr says he wins the pot "most of the time" - so > 50%.

So, what are the real numbers? Well, that tricky little de-coder came up with a way for us to find out. I have yet to do so on my own data, but I will shortly. Here were de-coder's results from analyzing his 560k hand, full ring, cash game observed database - mostly $50 and $100 NL (I'm editing what he wrote to what I believe we are now agreeing the formula is doing):

So, it seems that random $50NL and $100NL players are having about a 40% success rate with raising donk bets. But what kind of table image do they have? Are they raising the correct amounts, on flops with the right texture, against the right opponents? We don't know. So I think it would be better to analyze our own data.

So let's do that. Here's how:

1. First, in order to run some SQL against your db, you need to do this:

Next, figure out what your Player_ID is by pasting the following in the top window and hit F5 - The results will appear in the bottom window:

select player_id, screen_name from players where screen_name like 'YourScreenNameHere%'

Now that you have your player ID, copy and paste this into the top window and hit F5 again. This will give you the results for the hands when YOU were the preflop raiser:

select count(x.game_id) from game_players x join game_players y on x.game_id = y.game_id and y.fbet = 1 and y.praise != 1 where x.raised_first_pf = 1 and x.fraise = 1 and x.fcheck != 1 and x.real_player_id = YourPlayerIDGoesHere

Lastly, this will give you the number of times the donk bettor folded to your flop raise:

select count(x.game_id) from game_players x join game_players y on x.game_id = y.game_id and y.fbet = 1 and y.praise != 1 and y.ffold = 1 where x.raised_first_pf = 1 and x.fraise = 1 and x.fcheck != 1 and x.real_player_id = YourPlayerIDGoesHere

If you'd like to see the general numbers for everyone in your database (not just the numbers where you were the preflop raiser) or if you want to run this against your Observer Hands database, just remove this part from the query and run it again:

and x.real_player_id = YourPlayerIDGoesHere

Be aware - these queries might take a while to run, so be patient.

OK, I'm going to screw with this for a while and post my results. I think it would be very cool if everyone reading this right now did the same. Hint, hint.

[Talking Poker]

de-coder: For my main database, there are a lot of limit hands in there from when I used to play limit. Is there a way to filter those out and only look at the NL hands?

My observed db is entirely NL, so I'm checking that one first.

[Talking Poker]

OK, this is taking FOREVER on my computer. Guess I'll have to play some Mahjong while I wait.

OK... From my 2007 Observed Hands database (630k hands) - 99%+ is 1/2 NL through 5/10 NL, and it's all 6 max:

3275 times this situation occured.
1561 times the raise took down the pot right there.

So, that's almost 48% of the time. This number is for higher limits on average than the one from de-coder's database, so I'm not at all surprised to see this play being more successful (many bad players can't fold draws, even when they should). The 6 max vs Full ring could also contribute to the difference.

FYI - it took me 10 minutes and 45 seconds to get the first number and almost 6 minutes to get the second.

I'll run this against my db next, but without a way to filter off the old limit hands, I'm afraid my results will be skewed lower than they really are.

Hmmm... I'm having some issues with my player ID (I get multiple results for Rogue23 and when I use Troll Pie's player ID (that's the one I want to use), I'm being told this situation came up 0 times, which I know is a lie. Maybe it's because of the way "aliases" are set up in PT? I'm not sure.

de-coder - Little help?

[Robbr25]

NOTE: I think this happens more than below. And I am not sure if limit is included. I will do more analysis later.

Based on these queries here are my numbers:
Times occurred: 148
Times villain folded after the raise of his/her donk bet: 33

Now I wanted to see how many of the 148 I actually won the hand, that number is 85.

So based on this it looks like I win the hand when I raise those situations more than I lose. It just may take the turn and the river before I take the hand down.

[Wes]

So you are telling me if you auto raise less than the pot sized bet it is automatically going to be a +EV situation since hey over your 3500 sample it worked out 48% of the time. Yay thanks! Now if you could lay out some other black and white rules I might win at poker!

[Talking Poker]

Actually, no... I believe what I said was this:

Note no mentoin of a 3500 hand sample... just some speculation about a certain situation in poker we were intelligently discussing.

But sure, if you want another black and white rule (feel free to write this down): "AA is never behind preflop, so folding this hand preflop is -EV."

But thanks for your wonderful contribution to this thread anyway. Don't you have anywhere else on the internet to be?

[Wes]

Next can we figure out how often someone calls a river check/raise? That way I can figure out the % of the time I should be bluffing with a river check/raise.

[Talking Poker]

Why are you being retarded?

Oh, right.

[de-coder]

The good news is - yes there is a way - you can here the but coming though right? But - the level has an integer code ID. I know that 1 = $50 NL, full ring - as for the the rest (I have a few others I could track down, but certainly not all of them). If you have any pull with the developer you could ask him what the code values are for game.game_level_id.

[de-coder]

I'm pretty sure aliases would be the problem there. I don't have any aliases so I don't know how they work. I assume it's something like this:

rogue23 has player_id 5
troll pie has player_id 10

there's a alias_id field in the player table that I would assume would have the number 10 in it for rogue23's record. Then it looks like all the hands are imported under player_id 5.

I suggest unaliasing yourself and then running the queries, but that's going to take a while too. Unless I mess around with this alias thing to figure it out that's all I've got.

[melioris]

TP- in your analysis how many times did you fold to a donk bet and how many times did you raise?

Good work guys, I find this thread interesting and useful.

[Talking Poker]

Really? You don't have any annoying comments to make about black and white poker rules that no one is suggesting? Good, I'm glad someone finds this useful. I know I do.

As for your question, I haven't even been able to narrow this down to the hand where I was the PFRer yet, lewt alone dig deeper to see what percent of the time I'm raising the donk bet vs. folding. I definitely don't do it 100% of the time though - besides some of the reasons I listed above somewhere (like table images, flop texture, donk's skill level and so on), I think the donk's stack is a big consideration. The play works better with deep stacks.

I'd like to see the results specific to my player ID, but I'm not sure how, since I keep getting 0 for the results when I run this against the ID that I think is correct... And then there are the limit hands in my personal db that skew things too. My observed numbers are solid though.

[helmuths henchman]

Here is a question, how often do you "donk bet?" Is it a reccomended play?

[Talking Poker]

I usually donk bet strong - flopped set, for example - and/or when I can squeeze a guy behind me. If the PFR is the one in between me and another cold caller, I definitely wouldn't donk bet (because I don't want him to raise and drive the other player out), but if the preflop raiser was the last to act, that's a good spot to squeeze the guy in the middle if you have a big hand. So, it really depends.

Generally though, I don't donk bet much. Heads Up, I'd prefer to check raise or check/call, depending how strong my hand is (or check/fold, of course). But on some flops against the right type of opponent, the donk bet will work too.

Sorry if this isn't a good explanation, but it's very situation dependent.

[helmuths henchman]

I understand what you were saying. I usually "donk bet" when I flop top pair. The result is usually either they fold or go over the top. I hate flopping top pair with an Ace kicker and he over the tops me.

[Talking Poker]

And that's exactly why I like raising the donk bet. When you lead out with TP and I raise you, you're in a bad spot, especially knowing you might be facing two more bets before you get to showdown. Now that you've led out and helped build the pot, you have to decide if you're willing to put your whole stack in with TPTK, because it certainly might come down to that.

I think donk betting with TP is a very bad play.

[helmuths henchman]

With that situation would you call the eventual bet, over the top him, fold or use all 3 on a situation by situation approach? Assuming you didn't donk bet but checked?

[Robbr25]

I have tried donk betting when strong, but so far it backfires (they fold).
So I don't know how successful it can be for me based on my playing style. I think it is very dependent on your opponent.

[Wes]

Instead of just blankly saying that a raising a donk bet works 48% of the time, why don't you set some parameters to the player that is doing it (say if his vpip is under or over 30 for 6 handed)? Otherwise, this is completely useless.

[melioris]

Is this the type of thing that is easy to do in Holdem Manager?

[Talking Poker]

Situation dependent. I try not to cold call often from OOP in the first place though, especially with hands that will results in this TPTK mess of a situation situation. If I'm cold calling OOP like this, I'm not looking to flop TP with my hand, and generally, I'm willing to give it up if I do.

[lightfungus]

And shouldn't the donk bet situation be specified as the lead out bet OOP is less than a percentage of the pot? Like say 50%...or 40%? If someone leads out for >50% isn't this a completely different story than someone who leads out with less than 50%? Or did we take that into account already?

[Talking Poker]

Because:

1. That's not what we were talking about in the first thread. We were simply talking about the success rate of raising donk bets across the board. I'm not sure how you don't get this.

2. What you are asking for sounds great. Translate that into something posgresql friendly, and I'd be happy to push the button and share the results with you. In the mean time, I think looking at a sample of over 1 million hands and counting (total, not how many times this situation came up) is better than nothing at all, where people were guessing the success rate to be anything from 15% to over 50%.

Also, I never said that raisaing a donk bet works 48% of the time. I said, that through the 630,000 6 max NL hands in my observed database (1/2-5/10 NL), it has worked 48% of the time.

Any conclusions you drew from that and any changes you've decided to make to your play because of that are completely on you. I presented the (useless, accordsing to you) results, and that's it.

I think fine tuning this query to only show hands where I was the PFRer and to include varion stats about our opponent would be great, but I have no idea how to do that with Poker Tracker. This is a good starting point though.

[Talking Poker]

If a guy leads into the PFRer (obviously OOP), that's a donk bet. Whether he leads out for a min bet or the entire pot will obviously effect if the PFRer chooses to raise or not, but once he does, it's being lumped in to these results.

Narrowing down the results based on the size of the donk bet as a percentage of the pot would definitely be interesting, but again, I don't know how to do that.

[de-coder]

I've figured out the alias issue and it's pretty simple actually.

change x.player_id to x.real_player_id and you're good to go.