JustinHorton wrote:Given the difficulty in assigning an accurate Elo equivalent to an ECF grade (I know of no calculation that will do it satisfactorily) I suspect what you'd be looking at is this week's controversy all over again - but bigger.
I actually think the attempted regrading shows us a possible template for a future Elo conversion. What you do in outline is
(a) determine the particular subset of possible Elo parameters that you are going to follow. These include but are not limited to the frequency of recalculation/publication, the rules for K factors, the rules for introducing new players, minimum ratings, 350/400 point rules, the underlying probability distribution etc.
(b) You then go back several years and do a quick and dirty conversion on the ECFs at that past date. You might alternatively use the past international Elo for players that had them.
(c) You then roll forwards using your proposed formulae in (a) applied to the actual past results. New players automatically pick up new ratings using your preferred new player routine. You keep a note of the shadow Elos at each recalculation point.
(d) Then you publish the shadow Elos complete with their history and invite review. They may or may not have the same shape as the ECF grades they would replace. You might wish to calibrate them with a fudge factor to bring them in line (on average) with the internationals.
The point of a wind back/ roll forward method is that you're relying on the inherent self-correcting nature of the Elo formulae to iron out any anomalies you might inherit from the conversion. It also demonstrates how the ratings evolve from one year to the next. Rapidly improving players would remain a problem of course. Presumably you have to tune the K-factor to try and keep up or just accept in the interests of stability that there will always be an element of ratings lag.
Everyone has a shadow rating and history thereof long before you go live on the new method.
Elo ratings aren't great result predictors either. I ran some tests on the last 3 British Championships basically summing the absolute value of the (W-We) calculation. This is the measure of the (actual minus expected). What I got was about 0.1 per player per game. Thus over 10 games it's 1 per player. Expressed in ECF terms this is the expected 6/10 scored by a +10 on the ECF scale being just as likely to have been 5/10 or 7/10 in practice.