I don’t think the Tribune’s headline was a result of polling. Rather it was because the election was close, the paper had a deadline and through rose colored glasses it looked to the editors and the publisher like Dewey was ahead at press time.
I never believe any poll I see, especially with regards to elections. They aren’t even double-blind, and the pollsters can subconsciously slant the questions. Besides that a sample size of 1000 is just not enough for me - the goverment samples 60,000 people to determine current unemployment and that’s almost a large enough sample for me to believe!
Sure the questioners can slant their manner of asking the questions. I’m not sure what double-blind means in the case of a simple opinion poll. Double-blinding is done when a comparison between two treatments is being made. That’s “treatment” in the general sense of two different operations or sets of conditions being used on a test group and a control group.
Just trying to picture a “double blind” political poll:
Pollster: Good evening, Mr. Smith. May I ask for a few minutes of your time to complete our survey?
Voter: Sure.
P: Will you be voting for Candidate A or Candidate B this November?
V: Well… who’s Candidate A?
P: I can’t tell you that.
V: What do you mean?
P: Well, we’re not given that information.
V: click
:smack: Of course.
That is a problem, but it is not sampling error in the statistical sense, it is measurement error, which is something else entirely. Sampling theory assumes that the measurements are exact - actually they have error bars on them independent of sampling errors.
That’s exactly why there are calls to do the census by sampling, not by actually having everyone fill out the form. With bias from those who don’t want to talk to a government man, relatively untrained census takers, and other sources of error, I would guess the error bar could be greater than sampling error, and the sample could be taken with more precision, since more money can be spent on getting it right.
There was a thread about using a sample in place of general electtions. It would never be accepted from a political and sociological standpoint but I’ll bet it would be more accurate that the present method of counting thousands of ballots marked by people who don’t know what they are doing and counted by ad hoc counters who are in the same boat.
As to the census, no one can count 300 million items accurately. Well, maybe we could put a barcode on everybody and have them walk by a scanner.
a) It doesn’t matter whether you believe them; the fact is that political polls work to a very high degree of accuracy.
b) The wording of questions can affect opinion polling to a great extent, but is an insigificant factor in a poll featuring just a list of candidates.
c) Large samples are used so that breakdowns to subgroups are large enough to be statistically meaningful. These subgroups have even fewer than the 1000 or so members that you object to.
Think of it this way. Lets say you had a jar of 1 million marbles that either say Kerry or Bush on them (you can allow for undecided and other if you wish, but let’s stick to just K and B for now). You don’t know what percentage for the marbels are K and what are B and you can’t look in to see. However, you can draw them out and you also know that each marble has an equal chance of being drawn (ie, one in a million). The sampling size question is this: how many do I need to draw to know with a certain degree of confidence, let’s say 95% confidence, what the percentage of K and B marbles are? I think you can see that just drawing a 1,000 might be more than enough to give you an idea.
In political polling, phones are used to try to randomly pick out people. Of course, that ain’t perfect, but the bias this creates can be dealt with.
Hope that makes sense.
I think you guys are forgetting a lot of polling bias, like who they call, what time of day they call (what type of person is more likely to be at home during the day/evening), and some people use cell phones now which pollsters probably won’t call - what type of people are they? If a pollster is asking random people on the street what type of person is likely to be walking around a city? Who has caller ID? I wouldn’t pickup a call from someone I don’t know, so it is unlikely I will ever be included in a poll, which just sounds biased to me.
With regards to double blind - I’ve heard of tests between 2 groups that were read the exact same instructions but the people reading them were told two different outcomes were expected - guess what, two outcomes were obtained. There is unconscious and conscious bias everywhere, and I’m sure you couldn’t prevent it for polls, but you have to realize it’s there.
I’d really like some proof that shows polls are accurate. A good example is the current presidential race - Kerry was leading Bush by a good amount a while back, and now Bush is up 8 points. People are pretty set on who they’re voting for from day 1, I can’t imagine 25 million people in the US decided in the last month to switch their vote from Kerry to Bush.
I don’t really understand why that would surprise you. An eight point swing is only one in twelve people. Do you honestly NOT realize how easily swayed and gullible that many people are? As I pointed out in another thread, you are living in a country in which 20% of all adults actually believe that Iraq used weapons of mass destruction against the United States last year. You don’t think a sudden flood of publicity can’t sway the votes of people like that?
The last polls held just before an election are almost invariably accurate within their stated confidence interval.