[HPFGU-OTChatter] I voted!!! was Re: Obama and daughter HP fans

[Kai Wen Lee]

CJ earlier:
Polls just aren't wrong.

Geoff:
That's a potentially dangerous statement.

CJ now:
While I can't speak to UK pollsters, w.r.t. US polls, my statement above is historically accurate. Only four times in the history of presidential polls has the aggregate average of the major pollsters been off by more than 3 percent; since 1956 polling aggregates have missed the mark by an average of less than 1.5 percent -- well within the polls' margins of error.

[You can see the numbers here: http://www.ncpp.org/files/1936-2000.pdf]

Geoff:
And wasn't there an (in)famous US occasion involving - was it Truman and Dewey about 1945-ish? - when the pollsters made a dog's breakfast of their predictions?

CJ now:
It was 1948 and it wasn't "the pollsters". It was just Gallup, doing his third election. 

Gallup's error in '48 from the modern pollsters' perspective was that he stopped polling some two weeks before the election. There were two factors that made the '48 election anomalous: the first was that there were two strong third-party candidates in play whose support collapsed during the final weeks, with most of their supporters breaking for Truman. Gallup missed the last-minute swing. The second was that the race was close enough for the late break to turn the election on its head.

And yet even in the event, Gallup was only off by five points -- a pretty respectable outcome even by modern standards. (He was off by more in '36, which is the year that cemented his reputation as a pollster when he called the election for Roosevelt contrary to national expectations. And he was off by nearly as much in '52, but nobody remembers that.) Since 1956 pre-election polls have missed the marked by an average of
less than 1.5 percent -- a rather remarkable achievement. For those
interested, the historical numbers are available at
http://www.ncpp.org/?q=node/101.

But all of that notwithstanding, browbeating modern pollsters with a 64-year-old polling error is a bit like questioning Steven Hawking's scientific credentials because he failed his second-grade science test. The '48 election remains the only blown call in polling history, for reasons which aren't likely to repeat themselves, and polling results have only gotten more accurate over time, not less. (There's an interesting graph on that at http://www.newsu.org/angel/content/aapor_polling07/1a_why/accuracy.php. Scroll down to the bottom of the page.)

Alla:
... Reagan trailed in polls too and he won nevertheless

CJ now:
In 1980, polls taken a week before the election gave Carter the edge over Reagan. It is not true, however, that the polls missed Reagan's late surge. RR's dramatic turn-around played out in all the polls and on front pages across the nation. The big lesson learned from 1948 is that polling data more than a week out is an unreliable prognosticator, and no reputable pollster would dare call an election based on week-old data. That's why there's always such a flurry of polling activate in the last week before elections. Any numbers more than three or four days old are generally regarded as good snapshots, but lousy predictors.

Unfortunately for McCain, this is one surge that ain't happening.

CJ

[Non-text portions of this message have been removed]