Text 143, 53 rader
Skriven 2005-09-08 09:52:02 av Miles Maxted (3:772/1.0)
Ärende: Swinging Polls
======================
G'morning all,

The local newspaper has sought to leaven the mish-mash of news 
about our upcoming Election by publishing an article answering 
`frequently asked questions' in an effort to explain how polls 
work and should be read.

As one of the pioneers of such polls in New Zealand, the detail is 
of professional interest - and considerable angst.

Right now,  we are being bombarded with polls, all of which claim 
to have "a margin of error of X.YZ%", depending on the size of 
their samples - 3.16% for 1000 people interviewed, 4.47% for 500 
and so on.

These claims are absolutely wrong and misleading. The facts may in 
fact completely explain the very different results that confuse 
and alarm some people today.

True,  the maximum error on 1000 respondents all chosen completely 
at random from the whole population is 3.16%.

But make enquiries as I do, and you soon establish that phone 
numbers, not voters, were chosen by computers to be rung. The poll 
becomes one of people answering landline calls,  and not of all 
possible voters.

Worse, as the NZ Herald  reports,  pollsters actually phone some 
2,000 people overall to get their 1000.   Mathematically (1),  the 
maximum error jumps instantly to plus or minus 26.94%,  a far, far 
cry from the 3.16% actually claimed.

My polls, conducted using face-to-face interviews on randomly 
selected homes, were decried by Norman Kirk (a former PM) as 
witchcraft - even when they got his electoral landslide exactly 
right for the NZ Herald.

By contrast, to-day's telephone polls depend more on Lady Luck's 
law of luminous lotteries rather than on any known science.  Very 
wisely, Colin James (newspaper guru) averages all their results to 
guess at the trends - even though I personally would refuse to bet 
on his averages.

Any comments ?

:-))

(1) The calculations for these margins are clearly set out in WG 
Cochran's "Sampling Techniques", 3rd Edition, Chap 13, pp359-396.   
___ MultiMail/MS-DOS v0.45

--- Maximus/2 3.01
 * Origin: === Maxie BBS.  Ak, NZ +64 9 444-0989 === (3:772/1)