westafrica

Written, not computed

Why there is no order inside a band

We compute a weighted score. Here is the measurement that stops us printing it as a list.

Every publication covering this market prints an order, and the order is the product. It is worth saying plainly what an order requires before it means anything: the distance between two neighbours has to be larger than the noise in how they were measured. In this market it is not, and the arithmetic that shows it is our own.

Run the network scorer over the thirteen books on the register, on the five weights this domain is registered with, and the whole result covers a span of 5 on a scale of ten. The middle gap between one book and the book below it is 0.3. The widest gap anywhere on the list is 1.3. seven of them sit inside a single step of one another, and two pairs finish exactly level at the decimal the score is published to.

Now consider what a gap of a tenth is made of. A book that publishes a withdrawal floor scores fractionally above one that does not, because the floor is one of the things the model can read. That is a real difference in disclosure and this site reports it, in a band. It is not a difference in whether the odds are good, whether the money arrives, or whether the terms are survivable, and printing it as a first place and a second invites a reader to believe it is.

The counter-argument is that a reader wants to be told what to pick. That is true and it is not a reason to invent the answer. What can honestly be given instead is the shape of the evidence: which books will tell you what you are agreeing to before you agree to it, which will not, and precisely what each one is quiet about. That is a smaller promise, and it is one the data can keep.

The score is not hidden. It is run on every build, it produces the numbers in the second paragraph above, and those numbers are recomputed from the fact base every time this page is generated. What is refused is the last step, the one where a spread of tenths becomes a ladder.

Questions this raises

Are you hiding the ranking?
The figures that describe it are on this page and they are the ones that matter: the span, the middle gap, the widest gap and the ties. What is not published is the order, because publishing an order is asserting that the gaps in it are meaningful, and the gaps are the size of a rounding decision.
Would a different set of weights produce a clearer ranking?
It would produce a different one, which is the problem rather than the solution. Ten publications reading this same fact base with ten different weightings produce ten different leaders, and none of them is more correct than the others. A reader who saw all ten would conclude, rightly, that the ordering was a property of the weighting rather than of the bookmakers.