[exim] Estimating spam deliveries

Hi,

Does anyone know a good way of estimating how much spam is being delivered
to mailboxes?

I suppose the most reliable way would be to perform a random sampling of
delivered messages, and inspect them personally.

I've thought about measuring the correlation between delivery stats, and
rejection stats on the assumption that all our rejections are true
rejection. I suppose such a technique would assume that our detection rate
was relatively stable, and independent of spam source.

I suppose that a high correlation between the number of messages rejected
and the number accepted would mean that a high proportion of delivered
messages were spam, and a low correlation would mean that a small
proportion of delivered messages were spam. But, would knowing the
correlation coefficient and sample size enable me to estimate the actual
number of spam messages delivered?

--
Ian Eiloart
IT Services, University of Sussex
x3148