Stryker wrote:Polls also predicted that Gore would win by at least 10% of the vote 4 years ago. 'Nuff said.
dead wrong. i worked for the National Gallup Poll during the last election. i understand the polling process fairly well. I know of no polling agency who predicted anything near a 10 percent margin either way in 2000. In fact, for the first time ever, Gallup called the election "too close to call". How right they were.
Lothar has it right. The respondent is first asked
if they voted in the last election, then
if they plan to vote in this one, then
who they will vote for in the upcoming election. There's no verbal trickery here. When the voter goes in the booth they pick one candidate or another. same thing in the polls. If the polls say bush is leading i believe them.
as far as polarization...
i don't think its really relevant. i think the polls will reflect the polarization. for instance, important demographics like 18-25 year olds likely to chose a liberal candidate since the respondent is asked if they intend to vote. In the 1992 election, more 18-25 year olds voted than in any previous election, so clearly when these demographics become polarized it is important. that was a big reason why Clinton won. the polls should detect the intentions of those in that demographic. hence, polarization won't fool the pollsters.
basically, the polls will detect, in each demographic, who intends to vote and for whom. The demographics are crucial because thousands of surveys may be completed and then only so many results used to reflect demos such as age, race, gender, income, etc. once a demographic is filled, those respondents screen out of the survey. each demographic is represented based on comparison to census info.
some may have heard me say this before, but the polls are accurate on the issue of the presidency, especially Gallup. if the polls weren't accurate then the trends would fluctuate wildly. they dont. there are consistent trends showing voter preference over time. also, gallup's ability to predict who would win is very impressive. afterall, they not only said that 2000 was "too close to call", they predicted truman would defeat dewey.
Re: goober's comments...
As far as looking at historical poll results...these are used to show the trend, but have no bearing on the results of any individual study. past voter turnout is not something used to derive sample (the phone numbers called). past voting may be used to predict if the respondent will vote in the upcoming election, but this is based on self reporting of past behavior, not data from the actual voting sites.
the sample is taken from a simple factor of knowing area codes and the socio-economics of where the phone prefix is located geographically. Then the last four digits of the phone number are randomly generated.
In sum, when you say "past data is meaningless" this is both true and false. Past data was NEVER meaningful in the sense that it changes the predictions made by the current study. In other words, no past data is used to "extrapolate" current data. Past data only shows the trend. The useful part of looking at past data is to view the trend over time.
