Vorlon: I realize Rasmussen's typical thing is to poll very small amounts over a number of days. So what does this say about the New Hampshire poll that was just apparently polled over one day (Sept. 15)?
I've always heard bad things said about one-day polls and really long polls too (and Rasmussen in general).
I actually like to have a number of days "in the field" for a poll. You can certainly argue the "ideal" number but I would say 3 days or so it likely the safe minimum.
The problem with doing a one day poll is that you by definition can't do any "call backs" where numbers you could not contact are tried again.
lets take an admittedly extreme example to illistrate this.
We are doing a poll in Wisconsin on if to build the Green Bay packers an New Stadium. We (showing very bad judgement) decide to do this poll on a Sunday afternoon when 70,000 Packer fans are watching the game in the seats of
Lambert Lambeau field.
No matter what we do, no matter how we structure our sample, we will NEVER be able to call the people who are not home, but are at the game - hence the population that is at home on Sunday afternoon is not a representative population, hence our poll is likely badly skewed.
The way a reputable pollster gets around this is that on Sunday afternoon they try all (for example) 1000 phone numbers in their sample, and among those they do not reach, they DO NOT just simply replace these numbers, but rather wait till MOnday, and "call back" the numbers on Monday, Tuesday, etc...
That way on Monday as representative portion of thge Packer Fans who were at the game get included in the poll.
In general, Republicans tend to be home a bit less than Democrats, For example the very old, say 70+ are quite strongly Dem leaning and also stay at home a great deal more, where as the 44-65 Demographic is more strongly GOP and tends to trael more, go out of the house more, etc. If you do not do callbacks you will thus get a skewed Demographic politically.
This is why doing a poll over 3-6 days is generally considered the optimum time frame.
Regarding Rasmussen:
Rasmussen utterly blew it in 2000. His polls were junk.
The problem is he uses computers with recorded voices, not actual humans, to do his polls.
The "response rate" the % of people and more importantly the pattern of people who will do an automated poll is rather different than the general population.
("If you support Candidate X press 7 on you phone, if you Support Y, press *, etc...)
Think Grandma with her bankcard at the ATM if you want a good mental picture..
AS a result in 2000 Rasmussesn final poll had Bush up 9% or so If I recall.
To "fix" his poll in 2004, Rasmussen has done a "hard weight" where he ASSUMES that Democrats make up 39% of the voting population, GOPers 35%, 26% are Independants and he imposes these ratios on his results regardless of the actual results hismachines bring back.
39/35/26 is not an insane assuption, and is similar to the last few Presidential elections, however going back a bit further, party ID can and has changed radically between elections.
For example in 1980 there were 15% more Democrats at the polls than GOPers, in 1984 that ratio had shrunk to just 3% - If you had weighted to 1980 party ID in 1984, you poll whould have been structurally out by 12% (See New York Times Poll - 1984)
If Democrats do out number GOPers at the polls by 4% in 2004, the Bot poll might be quuite close. If the ratio changes, Rasmussen could have a lot of egg on his face.
Zogby also does a similar hard weight.
Historically, Zogby is either very close, or utterly out to lunch.
In 2000 he was pretty close, in 2002 he was a total joke.
In 2000 he guessed right on party ID, in 2002 he guessed wrong.
In 2004....?
We will see
