VP scales in Swiss Pairs

patricks

We run a local Pairs League (a round robin) using EBUScoreSwissPairs, and there are four tables in each division and each pair scores up the 10-board match with the three pairs from the other tables sitting in the other direction. We convert the resultant total imp score to VPs. We use the "cross imps divided by sqrt(rc/2)" option as this (imho) represents the fairest calculation.

We are seeing anomalies in the calculation of the VPs done by the scoring program. We have examples of

• score is 26.1 imps converted to 18-2 while the tables (WBF-D as displayed at setup) says that 17-3 goes with 23-27 imps,
• score is 0.4 imps converted to 11-9 while the tables says that 11-9 goes with 1-3 imps so we expect 10-10,
  and about 5 others that could be attributed to rounding errors but still look wrong.

It could be all about rounding errors, or it could be that the two parts of the program are working on different old/new VP scales.

It is not a great problem but it would avoid having to explain to people why scores don't match what they expect if we made these consistent.

Robin_BarkerTD

I have looked at the data (thanks)

There is a general problem: fractional IMP scores in the gaps in the discrete VP scale are being rounded away from average, whereas scores below n.5 should be rounded down to n and get the lower VP score - certainly match totals strictly between -0.5 and +0.5 should be a 10-10VP draw.

There is a more complicated problem relating to this particular event. The rounds are 9 boards but each sets starts with 1, 11, 21, etc. So it has been set up as 10-board rounds, with the last board being 'not played'. Although the program quotes a 10-board VP scale, it in fact is using the 9-board scale. The actual scale can be set on the 'Movement Details' tab as 'User Defined Board Set (for VP scale)'

patricks

Thank you - that does explain the anomalies which I have seen. I have fixed the movement to use the 10-board scale and will handle the small number of adjustments for the rounding method manually (until it changes!).

JeffreyS

I hadnt realised the rounding issue on the 0.5 boundary. I recall the White Book advised rounding scores away from average. Note also, there is a change with 'divide by 0.7' for this Ximps root2 type of scoring in the new WB, which I dont think has yet been incorporated in EBUScore..

Robin_BarkerTD

The WBF scales (even the "continuous" scales) assume the IMP scores have been rounded to a whole number - we found this buried deep in the WBF scales definition document. White Book 4.2.5.1 defines rounding as '..., with exact halves rounded away from average.'

Of course, none of the WBF stuff applies explicitly to pairs XIMP, so the White Book should be clear(er).

The 'XIMP/0.7r' in the White Book is based on text that has been around for a while. It is obviously meant as an approximation to XIMP/root(2rc). When I revised the White Book text from a suggestion to a recommendation, I did not know that EBUScore already did XIMP/root(2rc) as the right method for conversion to VP. Given that EBUScore does the right thing, the White Book should (I hope) revise the recommended method to XIMP/root(2rc).

patricks

Where cross-imping is used, the concept of dividing by root(2rc) is there simply to allow the use of the standard VP scale for the number of boards. It would be much easier for players if this wasn't done and the VP scale for NB*RC/2 was used instead. They could then check their own scores.