how is the "margin of error" of a poll determined?

[Syrinx]

And how accurate is the determination of that margin?

Can someone explain that in simple terms? Thanks!

[Trajan]

1. It's too late for me to study it

http://en.wikipedia.org/wiki/Margin_of_error

[Syrinx]

3. thanks for the suggestion

Sometimes I forget that Wikipedia has an article on just about every phrase or figure of speech. :D

[lurky]

2. The more people in the survey, the smaller the MOE.

If the people chosen for the poll are truly a random cross-section of the voters, then it's pretty accurate. That's the hard part.

[ColbertWatcher]

4. Here's a link to a pretty good explanation ...

http://www.roanoke.com/columnists/angleberger/wb/wb/xp-53032

[liberati]

7. Excellent explanation!

I like the soup analogy.

[ColbertWatcher]

8. I did too. It's something everyone can understand. n/t

[Richard Steele]

5. History. They compare a given poll to similar polls from the past....

...and they look at how often similar polls were WRONG in the past
to determine how likely a current poll might be wrong. They compare
a single current poll to many, MANY past polls, and by AVERAGING
how much the old polls were wrong, they can predict how much
a current poll will be wrong with amazing acuracy.

The exact FORMULAS they use to compare & contrast old polls
with current ones are closely-held 'trade secrets'.

EVERYONE has the same access to the old-poll data; the various
polling corporations compete for contracts based upon claims
that THEIR formula for crunching old poll-numbers is slightly
better than the next Corporation's formula.

[pnwmom]

6. Actually, it's directly related to the sample size used.

And so different pollsters will report the same M.O.E. for the same sample size.

I worked for a pollster and we had thick sets of tables showing exactly what the M.O.E. was for any result, based on the sample size. The bigger the sample size, the smaller the margin of error.

[Richard Steele]

10. Yup- they're all working on the "perfect formula" that will let them use the smallest sample size...

...to derive accurate predictions.

The smaller the sample size, the FASTER they can do a survey.
Back when the first "Surveys" were done, it took weeks to
get general answers to general questions.

These days, they've got it down to such a SCIENCE that we're
inundated with specific numbers that are updated hourly.

It's pretty impressive, when you think about it.

[Lucky2017]

9. Actually it's pure mathematics.

When you call for example 800 people to ask for their political preference, there's always a (very very small) possibility that you end up with 800 republicans or 800 democrats or more realistic that the sample is not an exact replication of the situation that is examined. There no way to prevent that because if you want to take a good poll, you've got to have a random sample of the population. So, there's always a chance the sample and the results of the poll won't be a good representation of the reality.

The margin of error can be mathematically calculated. It is dependant on the size of the sample (if you poll almost every American, you will always have a good representation of the country and the margin of error will be almost zero, if you ask only 2 people for their presidential preference that can't be a good representation of the country and the margin of error will be very big).

An example: If a poll shows Obama with 52% en Mccain with 42% and the margin of error is 3%, this means that 95% of the times a poll like this one is taken the real situation will differ less than 3% from the poll. So there is a 95% chance that obama's real support is between 55% en 49% and McCain's between 39% an 45%. There's always a little chance (5%) that te real situation differs more than 5% from the poll's results.

I hope this explains it a bit. My excuses if my English is rather poor.