Margin of Error? Does it exist? (vanity)

Theoretically, the margin of error does go down as you combine poles, but the problem with polls is not statistical uncertainty, it is a problem with the methods and motives of the pollsters.

Political opinion polls are just guesses with numbers thrown on for support, and all the guessers are looking at the same data and methods, and have the same motives (the same as MSM?).

Hope for the best, prepare for the worst.

You have issues of different pollsters making different assumptions about who to poll, when they are polling, where they are polling, and so forth. They may all poll the same basic spots but diverge in only a small fraction of their samples. The extra size then wouldn't give you greater confidence, it just would mean you have duplicated a lot of your polling. To add subsets together, they all must be done in exactly the same way at the same time and according to the same statistical plan that is designed to sample the entire diversity of the pool. That isn't the case when you average in, say, a Zogby poll and a Newsweek poll.

It's better to be ahead than behind, even within the margin of error, in all these polls. The 2000 election showed just how off the herd of pollsters can be, though.

Spot on. These idiotic snap polls featuring 548 or 702 or some similar number of ''likely voters'' are a statistical waste of time, considering the broad sampling population of millions (in the case of states, say, Ohio) or > 100 million nationwide.

Any statewide sample size < 0.001% of the available population has its MOE understated by assumption, said assumption being that the elements of the sample are normally distributed relative to the entire population. What rubbish.

My question is: If all the polls pretty much consistently show President Bush ahead of John Kerry, and they all pretty must show the same percentage points for each candidate, doesn't this diminish or eliminate the margin of error?

If the polls can be regarded as random samples from a uniform population, then yes, averaging multiple polls could reduce the MOE, although it would never be reduced to 0.0.

For example, if one poll shows B50-K44 while another shows B48-K46, and both polls have an MOE of 3.0, and the same number of voters are questioned for each poll, the average of the polls (B49-K45), should have a combined MOE of roughly 2.1. The average of four polls, each individually having an MOE of 3.0, would give a combined MOE of 1.5; the average of nine such polls would have a combined MOE of 1.0; the average of 16 such polls would have a combined MOE of 0.75...

IMO, given that the numbers that are released have been "adjusted" by criteria selected by the pollster, it is misleading to average polls and combine MOEs. Better numbers might be obtained by averaging the internals from each poll.

If you have a MOE of 2% that means that generally, if you were to preform the same measurement with the another randomly selected population of the same size, you would get the same result 98% of the time.

Actually, a MOE of 2% means that the probability is high (usually 95%, a value rarely stated, but a standard one) that the true mean of the population is within 2% of the result you saw.

This of course assumes that the population is well defined and that each item in the sample (voter) has an equal likelihood of being chosen. Neither of these are true, so we put a degree of precision on these polls that simply are not justified.

What bugs me is the so-called "undecideds". If they haven't decided by now, don't include them in the poll.

Say, for example a poll shows the results of "Bush - 47%, Kerry - 45% and Undecided - 8%" That makes it look close...doesn't it? But take out the undecideds in the same poll and you get the following results "Bush - 51.08% (which is 47/92) and Kerry - 48.92% (which is 45/92)"

Bush would clearly have a majority instead of a plurality...perception is everything and merely winning a plurality adds to the perception of illegitimacy the media would like us to believe.

I'm with you. I don't understand the MOE.

The margin of error for a poll is a statistical term that reflects ONE THING ONLY. Namely, it reflects the number of people who participated in the poll.

In essence, the margin of error says:

The margin of error is an estimation of the extent to which a poll's reported percentages would vary if the same poll were taken multiple times. The larger the margin of error, the less confidence one has that the poll's reported percentages are close to the "true" percentages, i.e. the percentages in the whole population.

Again: despite the misinterpretation that it reflects the actual accuracy of the poll, it is NOT!!!! that. As given in the media, it is ONLY a function of the number of people sampled for the poll. That's OK, if you understand the assumptions and limitations behind that.

The basic idea is that pollsters are sampling from a known population. There are many, many assumptions that go into poll results, and if you understand them, you'll know what to distrust about them.

The biggest, and least trustworthy, assumptions are that the people sampled for a poll are a random and representative sample from the overall voting population.

Another huge assumption is that the pollsters know the characteristics of the population from which they're selecting samples. For example:

a. They know how to factor in things like party loyalty
b. They know the proportions of R's to D's to Independents
c. They can actually reach people who represent those proportions

d. And the BIGGIE: that those responding to a poll are giving honest answers.

The extreme variability among the various poll results, and the secretive polling processes used by each organization, show that nobody is meeting those assumptions.

The MOE, depending on the size of the sample, means that the poll results are correct, within that range about 95% of the time.

NOT TRUE!!!!!

It means only that you'll get the same poll results 95% of the time -- there is still the minor problem of having to assume that your samples are of the true voting population.

There is a famous polling disaster from 1936. An organization did a nation-wide telephone poll and determined that Landon would beat Roosevelt in a landslide. Their polling technique was pretty good, but they left out one major factor: they weren't polling the voting population, but rather the population of people who had phones. And in 1936, that population tended to be much more wealthy, and heavily Republican.

The same problem is going on today: the people who will answer polls are not necessarily representative of the voting population. There have been numerous articles in the past few years discussing the increasing difficulty of getting representative samples: things like unlisted cell phones and caller-ID tend to exclude many people.

Go to the link at my post above, to see how very little the MOE really means in terms of poll results.

Basically, the margin of error is the likely range in which a larger poll of people sampled the same way would fall. So if you telephone a certain number of people by randomly dialing telephone numbers between 1:00 and 2:00pm and ask them which presidential candidate is less of a weasel, then if the MOE is 3% that means that if you were to select a larger group by the same means and question them the same way, then 95% of the time your results should be within 3% of your smaller sample.

Note that your poll says nothing about how people would actually vote, or even about people in general. It merely says that of those who will answer the phone between 1:00pm and 2:00pm, the percentage that will say that Bush is less of a weasel than Kerry is likely to be within 3% of the sampled value.

As for your question about combining polls, if two polls are of equal size and MOE, have similar sampling methods, and yield about the same results, the MOE of the average will be about 70% of the MOE of the polls individually. Combine four polls and the MOE of the aggregate will be about half that of the originals.

Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.