Swiss Pairs using VPs but not IMPs
Before I start, we have a good reason for running Swiss pairs but not using IPMs.
Can anyone point me towards conversion tables from Match Point percentage to VPs? I found one before, but
a) I now can't find it, and
b) It made no mention of boards per round, which surely it should do?
Thanks
Comments
I could be wrong but go to the event details in EBUScore Swiss pairs and select the "movement details" tab.
Enter the set up for the session (tables and boards per round) and then click the "print IMP> VP scale" button. You can vary the VP compare by selecting various scoring methods such as "cross imps to vps" but Swiss pairs will give you a table comparing VPs to Match points.
Hope this helps.
https://www.ebu.co.uk/documents/laws-and-ethics/white-book/white-book.pdf p51
Thanks Dibbler - whoever you are.
Worked perfectly.
Does anyone know where the numbers come from? The WBF only refers to IMPs to VPs.
I think the EBU scales are based on the principle of all match outcomes being equally likely: 20-0, 19-1, 18-2, ... 10-10, ... 0-20,
This was the basis for the IMP to VP scales used by the EBU before 2013.
I asked Max Bavin a similar question in 2009 and this was his reply:
As with our [EBU] IMP to VP Scales, the objective is that between two evenly matched pairs, every result from 20-0 to 0-20 should be equally likely. So, there's a 1/21 chance of any given pair getting any score from 0 to 20 in any given match (against opponents of equal strength).
This is why the band for a 10-10 draw is quite narrow (as a close match is quite probable), then the bands get ever wider as you go up/down the scale. At this point the mathematicians take over; the distribution of possible scores in terms of MPs/IMPs is what is known as a 'normal distribution' [the graph is bell-shaped], and the cut-off points which you pick to obtain the 1/21 probability are well known (e.g. they can be looked up in a book).
The only thing the mathematicians need to know is what the 'standard deviation' is, and this is trial and error; or, rather, it involves (involved) an analysis of literally thousands of results from Swiss tournaments (where it is a reasonable assumption to make that the opposing pairs are evenly matched after the first few rounds). I think we ended up with s.d. = 20/3, though I could be wrong about this (maybe 6.5; for sure not as high as 7). The mathematicians could tell you very quickly by simply looking at the scale; they could also explain this rather better than I am doing!
The research was originally conducted by Mike Pomfrey well over 30 years ago when Swiss Pairs was first introduced. Since then the likes of John Manning and John Armstrong have had an input, and I still have the original notes from all three of these in my office. But it's actually a fairly straight-forward process once the basic concept has been grasped; the current scales (having re-evaluated the s.d. some while ago) were initially calculated by David Martin, and then 'tweaked' by myself and John Probst (20-0 was coming up slightly too frequently on the original Martin scales).
I bet you wished you hadn't asked! The teams (IMP) scales are calculated the same way.
Tkanks Gordon.
I did wonder if it was based on the Normal distribution (most things are - and I can prove it!)
That does seem to be an assumption though - has anyone done a chi-squared test to validate this?
I'm also wondering about the " the objective is that between two evenly matched pairs, every result from 20-0 to 0-20 should be equally likely".
To my mind, you would expect equally matched pairs to generally have results near the median. I would have thought that a better goal would be that every result from 20-0 to 0-20 should be equally likely between any random pair of pairs. If evenly matched pairs can get 20-0 1/20th of the time, then unmatched pairs would have 20-0 results almost all the time. (I'm aware that they don't.)
In the end I suppose a good test is how often each possible score comes up in practice, and whteher they are evenly distributed. I feel an analysis coming on!
John Probst certainly did a lot of work on VP scales, much of which was described on his web site, which I believe is now long gone. I did save a copy of his paper on VP scales, which he described as "public domain subject to acknowledgement". I can e-mail a copy if anyone wishes to contact me at Barrie@bridgeclublive.com
John was CTD for Bridge Club Live from 2005 till he had a serious stroke in 2010. During that time he developed VP scales for Butler IMP Pairs matches of 18 boards that later led to the ability to include IMP-scored events such as Multiple Teams, Swiss Teams and IMPed Pairs into the NGS.
Section 10 of his paper shows how John developed for me a VP scale for a Sheffield and District League with matches of 24 boards and Aggregate scoring after I sent him data of 418 previously played matches. Again, we started with the principle that each VP outcome should be as equally likely as possible .
Digressing away from VP scales, John created a formula to enable a sensible ranking list for Match Pointed Pairs events in Bridge Club Live in which players can play any number between 16 and all 96 boards that are available to be played in a fixed 24 hour period.
It is known as the Square Root Formula and it enhances the score of a player who plays more than 16 boards and scores more than 50%.
A players adjusted score is 50% + ((raw score – 50%) x sqrt(number boards played)/4
(where 4 is the square root of 16, the minimum number of boards that one can play).
John put this to Max Bavin and it became the basis for Master Points for such an event.
Till he had his stroke, John was an experienced EBU TD and I learned much from him.
Barrie Partridge - CTD for Bridge Club Live