To sashay to 3NT or not
Bob Richardson
looking at results where the opening 1NT bid was invited. The major purpose
is to determine whether or not it makes sense to bash into 3NT via 1N-3N, or
to get there gradually. More importantly, how harmful is it to give the
defenders that additional information.
Included in these "invitational" hands were auctions that started
1N-2C-2X-2N as well as 1N-2N. I excluded all hands where the final contract
was something other than 2NT or 3NT, thus avoiding the use of 1N-2N to show
a long minor and other hand patterns.
<<< Stopped at 2NT >>>
Combined Number Pct. Pct. Making Actual Play Success/
HCP of hands Making 2NT with DD Play DD play Success
20-22 21 62% 67% 93%
23 67 66% 57% 116%
24 34 71% 74% 96%
25 7 86% 86% 100%
26 1 100% 100% 100%
Of the above 130 hands that stopped at 2NT:
35 (27%) were set
56 (43%) made exactly 2NT
39 (30%) made 3NT, or more
Stopping at 2NT is not all that safe. The 71% making 2NT with 24 combined
HCP compares with the 55% making 3NT with 24 HCP. Of course, all 24 HCP
aren't the same. Shape and many other factors need to be considered.
<<< Continued to 3NT >>>
Combined Number Pct. Pct. Making Actual Play Success/
HCP of hands Making 3NT with DD Play DD play Success
20-22 13 15% 15% 100%
23 21 24% 24% 100%
24 49 37% 35% 106%
25 50 56% 60% 89%
26 27 59% 70% 84%
27-32 29 97% 100% 97%
The above shows the folly of continuing to 3NT with less than 24 combined
HCP, but also shows a low success rate when the combined assets are exactly
24. The prior thread showed a 55% success rate for the bashers while the
above shows only 37% for the sashayers. This implies that the info given to
the defenders is significant.
I'll next try to show the impact of responder's shape on these figures.
rhm
"There's a lot more to hand evaluation than just HCP"
In thomas Andrews article "card values for 3NT" at
http://bridge.thomasoandrews.com/valuations/cardvaluesfor3nt.html
he showed already years ago that that the 5th evaluator, which
subtracts one fifth of a point from a queen and king and gives this two
fifth of a point to the ten improves 3NT bidding markedly and suggests
to bid 3NT when only 24 points (roughly) are present.
This might also partially explain why the experts may have succeeded in
some contracts and not in others. (presence of body)
It would be interesting to see how this evaluator would affect your
data analysis though your analysis would now have a granularity of 0.2
HCP..
Rainer Herrmann
paul...@infi.net
study? Personal experience (from basketball and baseball simulations)
suggests 10-20%. In my basketball simulations I typically pit each team
against each other team for 500 games, or 5500 games PER TEAM in a 12
team league, and STILL I cannot say whether a team which wins 62% is
better than one which wins 59%...and baseball is even worse. I think
bridge hands are more like baseball in that events cascade much more
than in a basketball game where every floor trip is virtually an
independent event.
I realize how long it takes for simulations to run, or to track down
real-life results, but the only conclusions I'd accept from this sample
of less than 200 hands is that it's a bad idea to bid 3NT with less
than 23 points and a good idea to bid it with 27+.
Alex Martelli
> Would someone please estimate the margin for error on this sort of
I tried in my original Bridge World articles, but made an interestingly
subtle mistake and ended up as a bad example in "American Statistician"
(fortunately I mis-estimated the error as _higher_ than it was: so, I
ran many more simulations than necessary, but at least my published
results are definitely established to very high confidence!-).
So, I'm reluctant to risk other such mistakes, at least until I should
happen to get some refresher course in elementary statistics!-)
> I realize how long it takes for simulations to run, or to track down
With Bo Haglund's DD DDL as ported to my laptop, about 0.65 seconds per
deal (using just one core, at about 100%; using both cores I can run two
processes each churning at about 0.67 seconds per deal, again just on
this one laptop) -- that's with flat hands of middling strength at NT,
but those tend to be pretty tough cases for DD analysis, so I wouldn't
expect higher runtimes for more unbalanced hands.
> real-life results, but the only conclusions I'd accept from this sample
Ay, there's the rub -- unless OKbridge or BridgeBaseOnline publish some
anonymized version of their huge databases of actually-played hands,
it's going to be very difficult to collect substantially large corpora
for any reasonably specific real-life situation.
It IS interesting to reflect that the first crucial statistical results
on contract bridge (by Jean Rene Vernes, including the Law ot Total
Tricks, in the mid-60's) were based on much smaller corpora... just the
Bermuda Bowl matches up to the early '60s which were published in their
entirety, many fewer hands than Bob is using!
Alex
Chris Ryall
>In thomas Andrews article "card values for 3NT" at
>http://bridge.thomasoandrews.com/valuations/cardvaluesfor3nt.html
>he showed already years ago that that the 5th evaluator, which
>subtracts one fifth of a point from a queen and king and gives this two
>fifth of a point to the ten improves 3NT bidding markedly and suggests
>to bid 3NT when only 24 points (roughly) are present. This might also
>partially explain why the experts may have succeeded in some contracts
>and not in others. (presence of body) It would be interesting to see
>how this evaluator would affect your data analysis though your analysis
>would now have a granularity of 0.2 HCP.. Rainer Herrmann
I'd already come across http://rfrick.info/bridge/halfpoints.htm which
is a little more anecdotal in its analysis, and shaves fractional points
in a slightly different way off the poor old queens and knaves, again
boosting tens.
I found Thomas's version more convincing - especially after finding the
raw data on his 'mother' page
<http://bridge.thomasoandrews.com/valuations/>
I take it we are all agreed on how wonderful aces are?
--
Chris Ryall Wirral UK <cjr...@my.domain>
Chris Ryall
>It IS interesting to reflect that the first crucial statistical results
>on contract bridge (by Jean Rene Vernes, including the Law ot Total
>Tricks, in the mid-60's) were based on much smaller corpora... just the
>Bermuda Bowl matches up to the early '60s which were published in their
>entirety, many fewer hands than Bob is using!
I don't feel these are a fair or reliable source of data. Although the
standard of play 'should' be good, funny things happen in bridge as a
knock-out or tournament cut situation approaches. I watched the UK gold
cup final end stanzas online about a year ago and with one side trailing
some of the bridge was barely worthy of my mother's local club. In
particular people were punting 3NT on just about anything.
Alex Martelli
> Alex Martelli wrote on "To sashay to 3NT or not"
> >It IS interesting to reflect that the first crucial statistical results
> >on contract bridge (by Jean Rene Vernes, including the Law ot Total
> >Tricks, in the mid-60's) were based on much smaller corpora... just the
> >Bermuda Bowl matches up to the early '60s which were published in their
> >entirety, many fewer hands than Bob is using!
>
> I don't feel these are a fair or reliable source of data. Although the
About as much as any other -- they're just too small!
> standard of play 'should' be good, funny things happen in bridge as a
> knock-out or tournament cut situation approaches. I watched the UK gold
> cup final end stanzas online about a year ago and with one side trailing
> some of the bridge was barely worthy of my mother's local club. In
> particular people were punting 3NT on just about anything.
So the statistics may record some crazy bridge as well as some sound
one... that's a fair sample. Just check the result sheets of any 100
OKBridge or BOL hands and tell me how many crazy bridge you see;-).
_IF_ one had a truly huge corpus one might attempt winnowing -- for
example, in a large-ish field, cutting off the best and wost results of
each board, Butler-like; or, ignore the last segment of a knockout if
the distance per deal betwen the teams is larger than some threshold.
I'd bet, intuitively, that this will make no significant difference to
the results. But it's at most a 2nd-order effect compared with the
issue of collecting such a corpus in the first place. Consider for
example that results based on selections from Ginsberg's library of DD
analyses (to identify specific situations of interest) are routinely
challenged as being based on too small a sample -- and that library has
750,000 deals, if I remember correctly... over a century's worth of play
at a regular 20 deals a day...
Alex
Alex Martelli
> I'd already come across http://rfrick.info/bridge/halfpoints.htm which
> is a little more anecdotal in its analysis, and shaves fractional points
> in a slightly different way off the poor old queens and knaves, again
> boosting tens.
That page focuses on boosting the evaluation of KQx/xxx above that of
Kxx/Qxx -- the old fallacy of concentrated honors. I thought that
Vernes/Charles, L'évaluation des mains au bridge [Brive, 1995] had put
that one to rest, and that my articles in the Jan/Feb 2000 issues of the
Bridge World had nailed the coffin down firmly, but apparently this is
not the case.
So, I wonder if another study might help -- consider two hands differing
only on that detail, deal N hands for partner and opponents (with 3-way
rotation as in my BW articles), pick the most lucrative contract the
hands can declare. For example,
K62 - Q53 - AJ87 - KQT4
vs
KQ2 - 653 - AJ87 - KQT4
Robert Frick claims the second hand is fractionally but measurably
stronger, I (and Vernes and Charles, etc) claim this is not so (they're
equivalent). Just wondering: is there anybody who agrees with Frick,
and would be ready to be convinced otherwise, if results show no
statistically significant difference for a large N? How big an N
(considering each deal is played N*3 times, each time in 5 contracts,
one per strain) would it take for the study to be convincing to you?
Alex