12 Jul 2012 15:53

## Kristofer: Small correction

```When you said that the optimal value for the "p" in Warren's formula
could be found by trial and error, using an empirical bias measure
(such as correlation between q and s/q), I said that it isn't
necessary to do it by trial and error, because, since a certain kind
of probability distribution function is assumed (exponential), to get
the conclusion that p is a constant, and because, in any case, there
are ways to estimate a good approximation to that distribution
function.

But of course, it could still be worthwhile checking the correlation
between q and s/q, for various p values, because of course, as I've
said, the approximating function is only a guess or an assumption. So
yes, the trial and error optimization of p makes sense, if p really is
constant with an exponential distribution function, and if that's a
good estimate for the distribution function.

But that trial and error optimization of p would only be valid over
many allocations, because Weighted-Webster most definitely does not
claim to minimize, for each allocation, the correlation between q and
s/q. So really, it might be better to just try to make a good estimate
of the best function to approximate the distribution function.

Suggestions:

1. Interpolation in small regions, using several
cumulative-seat-number(population) data points in and near each N to
N+1 interval

2. Least squares, using more data points

```
(Continue reading)

Gmane