Re: Stats-n-such

Date view Thread view Subject view Author view

From: Rahul Dave (
Date: Tue Nov 28 2000 - 13:28:18 PST

Changing the methodology midstream is bad.

However, you have to ask what the measurement really is.

The state totals are what matters, but the measurements are county wide.
Some counties punch, some have levers, some others may hand count.
Actually then, we ought to probably say methodwise.
Thus if two counties use the same method, they should have the same standards.
But its ok for different standards in different methods. The mechanism
for combining the error statewise should take that into accounts. For
example, on the false assumption of only statistical gaussian errors we need
to find the distribution of a sum of gaussian variables, each with a different
sigma. I say false assumtion becase undercounts I would think have a
systematic component..but I an not sure how to account for that.

So if Broward is counting dimpled chads, so should Palm :-)

No county should only recount partially in one way, thats plain nuts.
> Ahhh, very good question! Perhaps it has something to do with changing the
> methodology, in some cases repeatedly, midstream? Imagine measuring a set
> lines on a piece of paper several times, and in some cases switching from
> english units to metric, using different rulers, and trying to convert and
> normalize it all via a calculator with only a single decimal place of
> accuracy... and imagine this being done by someone who *very badly* wants
> to see a particular result.
> That's basically what we're doing, here...
> jb

Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Tue Nov 28 2000 - 13:32:32 PST