Join Bridge Winners
2018 USBC analysis & Levin - Weinstein profile
(Page of 9)

Introduction

Having done the 54 EBTC analysis, I decided to move on and try to make the same analysis for the 2018 USBC. It turned out to be easier said than done – the USBC website doesn't appear to have scores and cardplay data like EBTC. Basically, the only data regarding the competition on the USBF website is team members, CCs, and the IMP result of each KO match segment. I don't know the reason for this medieval approach, but I guess it could be the subject of a whole new article.

However, all matches starting from the quarterfinals were on Vugraph, so for that part of competition, data is plentiful. I would like to thank Tomislav Ahri Gracin for making the .lin files readable and adjusting them to the desired format. Without his help neither this analysis, nor the one made for EBTC would be possible. I would also like to thank Al Hollander and Jan Martel for providing me with a couple of missing .lin files. Being done with intro and credits, it's time to move on to the deals.  Click on the link below to see the full analysis of the 2018 USBC.

https://public.tableau.com/profile/ante.miji.#!/vizhome/2018USBC/2018USBC

When examining it, make sure that you enter the full-screen mode (button is located at the bottom-right corner). If you have any difficulties understanding the dashboards, you are likely to find your questions answered on the Introduction dashboard, or in "dashboard notes" on the right side.

There are two chapters of the analysis: statistical overview and pair profile. In this article I will try to give a short explanation of each of these chapters and dashboards within.

Understanding the data

Before we get started, I would like to point out certain facts that you may have already read in the Introduction dashboard of the link given on the previous page. However, it feels convenient to write it here once more:

Data refers to the QF, SF, and F of the 2018 USBC. This means that all the data is from the KO matches. In the quarterfinals (QF) there are 8 tables, in the semis (SF) there are 4, while in the finals (F) there are only 2 playing tables. This means that Cross-IMP scoring is more or less representative in the QF, questionable in SF, and highly misleading in the finals.

For example, let's say that in the finals, open room NS was in 4, which is a cold make. They made it, but in the closed room NS was also in 4 and for some reason went down. Cross IMPs scoring at open room will be +10 NS and -10 EW, while in the closed room it will be vice-versa. And if the same event occurred at RR type of competition, with 20 or 30 tables, an unexpected score of closed room would have a large effect only on that table, and a substantially smaller one on other tables. 

This makes Cross IMPing unreliable. If we add to it the small data set (1,650 different scoring inputs), it's obvious that any statistical analysis will not be meaningful in its conclusions. So, due to the type of the competition and sample size, the whole point of the article is not to make some statistical observations and conclusions - it's more about trying to explain the possibilities of this type of analysis. 

So why did you make an analysis of a KO type of competition at all? Wouldn't it be better to do the analysis only for RR competitions?

It would. However, this data was retrieved from Vugraph archives, so in the .lin files there is bidding and cardplay data. There is no RR competition where all tables are covered by Vugraph, meaning there is no RR competition (that I know of) where there is bidding data from all tables. While I had no idea how to do any real cardplay analysis, I wanted to examine the bidding data and see what types of measures can be made with it. I focused on 1NT openings, preemptive openings, and 10 or 11 HCP actions. There is more bidding analysis to be done, of course, but this is what I've done so far.

I understand that by requiring players to use tablets for input(an idea presented in Ostende), bidding and cardplay data from all tables will be available, and so will the full bidding analysis, with Cross IMP scoring being a valid measure. Having said that, I still think that playing bridge on a tablet is a lot like playing football with VR glasses - just not the same! However, it's only my opinion, and there is a lot more to be said on this subject. 

STATISTICAL OVERVIEW

Cross IMPs types of views

On most of the dashboards, you can switch Cross IMP views between Total Cross IMPs, Average Cross IMPs, and Relative Cross IMPs. Total Cross IMPs will return the sum of the Cross IMPs scored in the metric that you are viewing. For example, if you are viewing 1NT openings with 5m(332) and set your view at Total Cross IMPs, the bar will return the sum of Cross IMPs scored by the opening side when opening 1NT. Average Cross IMPs will return the average by board, and Relative Cross IMPs will display the Cross IMPs scored using the strength of the opponents it was scored against.

Relative Cross IMPs is calculated by adding your opponent's average Cross IMPs to the Cross IMPs that you scored against them. So, if you score +0.5 against opponents with average Butler of +0.3, your Relative Cross IMPs score is +0.8.

 

Played contracts

Select a side you want to examine – declaring or defending. Click on the bar or bars you want to view in more detail. For example, if you click on "NT", you will see all NT contracts played, no matter the level or whether it was doubled or redoubled. However, if you click on the NT game bar under "P" header, the right chart will only display 3NT, 4NT, and 5NT contracts that were not doubled or redoubled. Switching from declarer to defender will flip the axes. This is logical, because whatever declarer scores on a certain type of contract is the opposite of the defenders' score.

 

1NT opening

There are a few hand type bins that I made provisionally. Hovering over any of the bars will display the hand type name, number of boards it was opened at, and Cross IMPs. Cross IMPs are displayed based on the view you had selected at the top right corner. Clicking on any of the bars will display the Cross IMP score based on opening position and vulnerability. Note that you can filter both charts by 1NT opener HCP. So, if i click on "Balanced, no 5c suit", and 15&16 HCP, I will get only the (4333) and (4432) boards with 15 or 16 HCP, and their Cross IMPs scores based on positions and vulnerabilities. 

 

Preemptive opening

On the left chart there you can see the Average Cross IMPs based on preemptive suit length. Hovering over any bar will display additional information about it. You can filter these bars by level of preemptive opening, opening position, and opening vulnerability. Clicking on any bar will filter by the longest suit length and then split by the honors. For example, clicking on the "5" bar will display all the 5-card suit preempts sorted by their honor holdings. Filters apply on both charts of the dashboards.

 

10 and 11 HCP actions

These use the same dashboards, but one draws on all 10-HCP hands and the other on all 11-HCP hands. So, when having a chance to open, you can either open or pass. Holding 10 or 11 HCP makes this decision particularly interesting. On the left chart you can see the Cross IMPs that opening or passing results in, based on Type A hand categorization, while on the right chart you can see it based on Type B hand categorization. You can filter both charts based on vulnerability and opening position.

PAIR PROFILE

After I had done the entire analysis, I realized it's worthless if I don't get some quality feedback from the end users. And who is the better end user than the players who played in the competition itself? So I've sent some emails to the players, hoping that I will get at least one. It turned out I got a few, but the one that stood out the most, by its quality and players resume, was from Steve Weinstein. Steve and I talked for a while, going through the analysis and trying to make it more understandable, for him and now for you. 

Steve was gracious enough to make himself and his partner, Bobby Levin, test examples for what I am about to show. I call it Pair profile, because it dissects the pair's game by as many verticals (dimensions) as I could think of, providing you with detailed view of each vertical.

Due to simplicity, from now on I am going to refer to Levin - Weinstein as LW.

Dashboards that were named in statistical part of the analysis can also be viewed again, but this time with pair filter included. I am not going to go through them again, because this article is long enough even without a few extra pages. However, if you want to see how LW are doing with 1NT openings, preemptive openings, and close opening bid decisions, all you need to do is to select them in "Pair" filter on the corresponding dashboards in PAIR PROFILE chapter. This part is going to be more about other metrics that were not yet shown.

 

Note

Images may not be easy to read. If the image is too small, right-click on it and open it in new tab. The original size of the image is readable, but it just could not fit in the BW article.

Cross IMPs overview

I have separated boards into three bins, based on variance: HIGH, MID, and LOW. In bridge terms, variance is essentially the measure of board complexity: HIGH boards are swingy boards, MID boards are part score vs part score boards or white games that can be missed, while LOW boards are usually push boards. If you want to know more about board variance and the method of calculation, there is a Board variance explanation dashboard, and if you don't like the fixed variance categories, you can observe pairs score on Moving variance dashboard. 

Having said all of this, it turns out that LW trend very well on all three of these categories (as seen in the bottom right of the graphic below). They tend to do exceptionally well in HIGH category, where they score +1.19 IMPs/board over 54 boards.

However, they don't perform as well on the first board of the segment (as seen in the bottom left of the graphic below). They have played 18 segments with an average first board score of -0.73 IMP. Beside board 1 and boards 8, 13, and 14, they average positive IMPs on every other board of the match, and tend to do very well in the late middle of the match (boards 9-10).

Note that these results are not to be taken as "truth", due to problems described on the previous pages -- I doubt that subsequent events will see them also losing IMPs on specifically boards 1, 8, 13, and 14 of a segment. These are examples for understanding what the data represents.

 LW - Cross IMPs overview

 

Bidding and detailed board view

Out of 240 boards played, LW had a Cross IMP score of -10 or worse on exactly 13 boards. How do I know this? Well, in Bidding and detailed board view, I had selected LW as a pair, in all rounds and segments, and set the Cross IMP range from minimum to -10. End result was shown in the photo below. If you click on any of the boards on the left, the bidding will be shown on the right side of the screen. Here we see LW falling at the merciless hands of Bathurst - Lall.

 LW - Bidding details

 

You can choose round, segment, and the Cross IMP range of your choice. For example, if you want to see LW boards where they score +10 or more, just set the range from 10 to maximum. Also, if you want to see all the boards from Segment 3 of QF, just select QF, Segment 3 of 8 and Cross IMPs range from minimum to maximum, so all the boards from that segment can enter the scope.

Cross IMPs by HCP and cards

After selecting LW as the pair to inspect, the following charts are displayed. You can see the Cross IMPs when opener was either Levin, Weinstein, or their opponents. Also, you can view the Cross IMPs by vulnerability split.

LW - All HCP and cards

 

Player 1 of LW is Levin, while player 2 is Weinstein. Let's say that we want to observe how does the pair do when Levin has 10-15 balanced hand, and Weinstein has 10-15 non balanced hand. The filtering is straightforward, so after entering the parameters the following graphs are displayed.

 

LW - 10-15 bal vs unbal

The charts remain the same as on the previous image, but the bars now only include the hands where Levin holds 10-15 balanced hand, and Weinstein holds 10-15 unbalanced hand. The hand types are taken provisionally, and I am sure they can be put to a better use. This was just a test example to show you what can be done.

Another feature is that you can set only the hand for one player, and leave the other player's hand type intact. For example, you want to see how LW do when Levin holds 15-17 balanced hand. You make that choice in Player 1 filters, but leave all the Player 2 filters intact.

Cross IMPs by other metrics

Selecting a pair (LW in this case) provides you with three charts - Opener, Dealer, and Type of bidding. Bars on each chart represents the Cross IMPs in the view you set, and hovering over them will show more data. 

 LW - Cross IMPs by other metrics full

Also, clicking on any bar of any chart, the other two charts are filtered. For example, clicking on Openers at Opener chart will filter the Dealer and Type of bidding charts by only showing the boards where LW were openers.

By clicking on the Competitive bar, the other two charts are filtered, so it's easy to notice that most of their competitive plus was earned when they were the opening pair. This is logical, because in competitive bidding it's usually the opening pair that has the advantage, or at least that should be the case.

LW - Cross IMPs by other metrics detail

Declaring and defending

All the LW stats are shown on the image below. Some interesting facts:

% Contracts made: 72.46%

% Partscores made: 76.79%

% Games made: 71.23%

% Slams made: 55.56%

% Contracts defeated: 31.06%

LW - declaring and defending

Conclusion

That's all folks!

Or not, I am not sure. I guess you could find many more details and interesting patterns in the data, but that was not the point. The point was to try to explain what can be viewed and examined, and the method to get the desired result. 

I know that this sort of analysis is not something that was made before, therefore it is not something that people are familiar with. So, if there is still a part of the analysis that is not understandable or not clear enough, please let me know and I will try to make it more user-friendly.

Also, if there is any unimplemented feature you would want to see, please let me know in the comments section.  Thanks!

24 Comments
Getting Comments... loading...
.

Bottom Home Top