commentator lead problem

[han]

*it's a side issue, but based on 3 samples of 100 simulated hands a club lead costs a trick in the suit 25% of the time compared to declarer's normal line.

This is a wonderful comment. Could you elaborate?

Without looking at whether it was a 'good lead' in terms of beating the contract (as I said, it was hopeless) I was interested in looking to see how passive a club lead actually is. By 'passive' I mean that leading one will make no difference in the number of tricks that declarer would make were he left to play on the suit himself. So on 75% of hands whether or not you lead a club doesn't affect the play in the suit, because either

i) Declarer is solid (AQxx opposite KJxx or whatever), or

ii) Declarer's normal line picks up partner's holding (Kxx in dummy AJ10xx in hand, or AQxx opposite K10xx etc)

 

On 25% of hands you cost a trick by leading the suit, because you pick up partner's holding, which is either

i) One declarer would never play for (Kx opposite AQJ9x)

ii) One where declarer would not usually play for that layout (e.g. AJ10xx in dummy opposite Kxx in hand, or Qxx opposite AK98x)

 

There are a few hands which are fuzzy e.g. A98x opposite Q1076 in hand, and I counted about half of these as costing (e.g. if declarer has to play on the suit himself and knows you have led a 4-card suit his choice of who to play for club length will depend on this combined holdings in all the suits).

 

The club lead actually turned out to be more passive than I had expected because of the combination of our diamond length and responder's lack of major suit length - if we were, say, 5521 or if responder had shown a 4-card major (e.g. 1NT - 2C - 2D - 3NT) I believe - although I haven't checked - that a club lead will cost more often.

 

I am vaguely interested to know why a club does so well DD, when on my simulation it virtually never beat the contract. Can you extract any of the hands where a club lead was successful and other leads weren't?

Did you go through the hands manually or were you able to program this?

 

I didn't keep the hands for my simulation. I don't know if there were any hands where a club was the only lead that would beat the contract. I think that almost always one of the other three suits would just as well. The double dummy analysis assumes that your side finds the right switch later so when the club lead doesn't give up a tempo it is almost always one of the best double dummy leads.

[Finch]

Did you go through the hands manually or were you able to program this?

Manually

[nige1]

Watching the women trials I heard the theory that AJxx, K10xx or Q9xx are the worst holdings to lead from against NT.

 

What would you lead with ♠ AJxx ♥ K10xx ♦ Q9xx ♣ x against 1NT-3NT at IMPs?

 

My double dummy analysis not surprisingly suggest that the small club is best but this seems very suspicious for real life. Apparently the second best double dummy lead is the spade ace followed by a low spade. Hearts is worst (well, the diamond queen is worst :(). What would you lead at the table?

Did you go through the hands manually or were you able to program this?

Manually

We are grateful to both Han and Frances for taking the trouble to perform their simulations. Assuming representative samples, it is an eye-opener to learn how significantly different the analysis results are when comparing double-dummy with single-dummy.

 

In the case of a ♣ lead, perhaps, on reflection, the difference is not quite so surprising. Like most BBOers, however, I would lead a ♥; never a minor.

 

Unless opponents are stretching, Frances' explanation of the case for a ♠ lead is pretty convincing.

 

Han's conclusions seem counter-intuitive so Frances' post came as a relief; but each simulation is instructive in its own way. Thanks again, both of you.

[han]

I didn't post conclusions.

[nige1]

I didn't post conclusions.

Sorry Han, I meant the result of your simulation.

[han]

Ah ok, no problem of course.

 

I won't be able to run simulations for the next month or so but I hope to get back to this topic and compare for example AJxx with Axxx and K10xx with Kxxx.

[benlessard]

Ive only done manually 52 deal so far wich is not enough to post any numbers and barely enought to make speculations. What im getting is similar to Frances. My specualtions are a club and diamond seldom work. H and S are about the same. M and D often blow a trick. In My 50 deal sample a C lead blow a trick only once wich is why lower then i would expect. The distribution of clubs look like

2=2

3= 8

4= 17

5=11

6 =12

7=2

 

So my gut feeling is that in MP leading the stiff clubs is the best lead by a mile.

 

The condition ive are about the same for declarer excepted that ive accept some of the 17 with 5 card suit hands.

For dummy ive put 7pts & + with 4M possible and rejected a lot of hand where 7-8 hcp isnt enough (but keept all the AKxxxx) and discarded everyhand where i would go throught stayman. I was Imps agressive the worst hand ive blasted to game was Q9x,QJx,xx, KTxxx. (and 3 nt has no play at all on any lead)

 

In restrospect ive should have put a limit to north hand because im getting too many hands where 3nt is cold. So my sample where a specific M lead is necessary to beat the contract is way to small to check wich M will turn out best.

 

Can we put lin files of many deals on the forum ? if yes what is the limit on the numbers of deals ?

[Cascade]

Some single dummy data of 4-card suit leads versus 1NT 3NT.

 

These data were obtained by GIB playing hands single dummy.

 

Each line of data represents 100* hands played in which the given honour combination occurred.

 

* Some lines show more than 100 because for example when I generated 100 hands with QJTx twice there were two suits in the same hand with the identical honour combination.

 

IMPs won is calculated based on a cross-IMPs. For each hand I made the "standard" lead from each of the four suits and cross-IMPed the results - no average (I think they use the same sort of scoring in the Cavendish but on BBO they average the cross-IMPs).

 

100 is really quite a small sample but you get a feel for the ranking of various leads.

 

I did not distinguish 9s so I have no explicit data on Q9xx etc they are lost within Qxxx - actually if I could be bothered I could extract the data for Q9xx compared with Q8xx etc but these would be even smaller samples.

 

I also have data for all 1, 2, 3 and 5-card honour combinations. e.g. a small singleton in the simulation lost on average 1.22 IMPs per deal.

 

Also not recorded here the simulation showed that there was a large premium on leading a major rather than a minor on the given auction. Possibly this was over estimated as I excluded some 4-3-3-3 hands with responder where 'standard' practice might be to not look for the major fit. Nevertheless I am confident as I am sure that you will be that there is a premium on leading a major.

 

 

Combination Frequency IMPs Won Mean IMPs SD      rank
AKQx         100       920      9.20     13.74    1
AKQJ         100       616      6.16      7.67    2
KQJT         100       515      5.15     11.67    3
AKQT         100       363      3.63     11.87    4
AKJx         100       337      3.37     14.05    5
KQJx         100       296      2.96      9.47    6
JTxx         106       222      2.09     10.87    7
AKJT         100       197      1.97     12.00    8
QJTx         102       168      1.65      9.76    9
xxxx         105       160      1.52     11.05   10
KQTx         100       130      1.30     12.04   11
QJxx         103       130      1.26     11.49   12
Qxxx         102        98      0.96     11.26   13
Txxx         102        46      0.45      8.10   14
ATxx         100        42      0.42     11.82   15
QTxx         101        41      0.41     11.35   16
Kxxx         103        31      0.30     11.16   17
Jxxx         105        30      0.29      9.36   18
AJxx         100        -9     -0.09     13.44   19
AQTx         101       -49     -0.49     11.65   20
KTxx         101       -59     -0.58     10.57   21
AKTx         100       -66     -0.66     10.69   22
AQJT         100       -71     -0.71     10.77   23
KJxx         100       -96     -0.96     10.26   24
AJTx         100      -120     -1.20     12.00   25
KQxx         103      -124     -1.20     10.64   26
Axxx         100      -121     -1.21     10.13   27
KJTx         100      -144     -1.44     13.12   28
AKxx         100      -144     -1.44     11.98   28
AQxx         100      -176     -1.76     11.38   30
AQJx         100      -288     -2.88     13.59   31

[jtfanclub]

KQxx         103      -124     -1.20     10.64   26

Axxx         100      -121     -1.21     10.13   27

KJTx         100      -144     -1.44     13.12   28

AKxx         100      -144     -1.44     11.98   28

AQxx         100      -176     -1.76     11.38   30

AQJx         100      -288     -2.88     13.59   31

I'm curious, if you don't mind, how the bottom 6 did with the opposite-from-standard lead. For example, if the standard lead from AKxx is the king, how did it do if you led low?

[jdonn]

Interesting that AKQx fares so much better than AKQJ. Partner has five cards in the suit that often??

 

Now that I look more, how could AKQT fare so much worse than both AKQx and AKQJ? That makes no sense at all, I think something here is flawed.

[jtfanclub]

Interesting that AKQx fares so much better than AKQJ. Partner has five cards in the suit that often??

 

Now that I look more, how could AKQT fare so much worse than both AKQx and AKQJ? That makes no sense at all, I think something here is flawed.

It's just a small sample size....

 

Do keep in mind, though, that this is cross IMPs vs. ALL other suits. So if the whole hands were....

 

AKQx

Jxx

xx

xxxx

 

vs.

 

AKQJ

xxx

xx

xxxx

 

you would expect a spade lead to score better on the first hand not because the spades do better, but because underleading the Jxx is more likely to give up a trick than xxx.

[Cascade]

KQxx         103      -124     -1.20     10.64   26

Axxx         100      -121     -1.21     10.13   27

KJTx         100      -144     -1.44     13.12   28

AKxx         100      -144     -1.44     11.98   28

AQxx         100      -176     -1.76     11.38   30

AQJx         100      -288     -2.88     13.59   31

I'm curious, if you don't mind, how the bottom 6 did with the opposite-from-standard lead. For example, if the standard lead from AKxx is the king, how did it do if you led low?

Good question.

 

For a number of honour combinations I have wondered how a different 'standard' lead might do.

 

One day I will do this.

[Cascade]

Interesting that AKQx fares so much better than AKQJ. Partner has five cards in the suit that often??

 

Now that I look more, how could AKQT fare so much worse than both AKQx and AKQJ? That makes no sense at all, I think something here is flawed.

I think almost all of the discrepancy is based on the size of the sample.

 

100 hands i quite a small sample.

 

However given in the simulation I did all one up to five card suits - I have only shown the result for four-card suits above - which is 111 one hundred hand simulations from memory. Each hand is played four times. So we are talking about 40000 hands. These are played single dummy. Depending on the settings GIB plays a hand a little faster than the average face to face human game but a comparable amount of time. This means that we are talking weeks of computer effort to do this sort of simulation.