If I remember correctly, isn't the Margin of Error really just the % of

[WCGreen]

just how accurate the poll may be as dictated by the size and nature of the sample...

I seem to remember that it doesn't mean you ad 3.5 either way, it means that there is a 3.5 probability that the whole poll is wacked...

It's been so long since I took an 8 hour class in polling, ack in the early 80's, and I just sort of fell into the accepted interpretation of the MOE. But it's been nagging me that I am right about it. It has to do with the standard deviation, if I'm not mistaken.

Anyone...

[cbayer]

1. It means that you can be confident of the results within that margin.

So, if it's 45/55 with a MOE of 3, then the results are really 42-48/52-58. You are right, it is a calculation based on sample size and variability of answers.

Anyway, I think that's right.

[WCGreen]

2. Thanks....

I'm to lazy to try and find my stats book...

Hell, there is a +/- chance it's in hell if I know land...

[DCBob]

5. That's not correct. MOE means a difference above that value is statistically significant.

For example: 50-45 with a MOE of 4 would be statistically different. 50-46 with MOE of 4 would not be.

[pnwmom]

18. You're just saying the same thing with different words.

So you're both correct.

[slackmaster]

13. There is also a level of confidence usually expressed as a percentage

For example, a 95% probability that the true figure is within the margin of error.

90% is commonly used as a threshold of statistical significance; but it does mean that 10% of statistics expressed at that level are expected to turn out to be wrong.

[Speck Tater]

3. Margin of error is how much of a spread, plus or minus,

represents the range of values that you can say with 95% confidence that the "real" value falls in that range.

In other words, the probability that the real numbers are outside the margin of error is 5%, or 1 chance in 20.

[WCGreen]

4. that's pretty much what I thought...

I get it now.

I wasn't comfortable just adding/subtracting the margin to the result. Something was nagging me.

[thenext8seconds]

6. wikipedia has an article all about it

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

[Jackpine Radical]

7. OK, lemme try. The margin of error is the p=+/-.05 confidence interval.

We're usually talking about the standard error of a proportion. As long as the proportion is somewhere near 50%, the standard error depends only on the sample size.

Here is the formula: SE = sqrt <(p)(1-p) / n > where p is, say, the proportion of McCain voters and 1-p (i.e. all the rest) is the proportion of Obama voters and n=the sample size. Now suppose there are 1000 people in the poll, and 450, or 45%, prefer McNutso. The Standard Error for this sample is calculated as follows: Multiply p, or .45, times 1-p, or .55. That will give you .2475. Round it off to .25. Divide that by 1000, and get .00025. Take the square root of that and get .0158. In a normal distribution, the margin of error (i.e. the limits that contain 95% of the population) is equal to 1.96 times the standard error. 1.96 times .016 is approximately .03. Thus 95% of the time, the poll result will be accurate within a three-point range, 1.5 points on either side of the obtained percentage.

[dsc]

8. Usually the MOE is chosen at a 5% level

which means there is a 5% chance the real mean is outside the range. The size of that range is what is determined by sample size and that is your MOE.

[pnorman]

9. MOE is the reciprocal of the square root of the sample size.

That's a fraction, so multiply it by a hundred to convert it to a percentage. I assume that there are guidelines about the method of drawing the sample.

pnorman

[Silent3]

10. The typical confidence interval is 95%

The confidence interval, unlike the margin of error (MoE), is hardly ever explicitly stated.

If a poll says Obama is favored by 52% of voters, with a 3% margin of MoE, that means there's a 95% chance that actual voter preference is somewhere in the range 49%-55%. In one out of 20 such polls the real value will fall outside of the MoE.

Error distribution is typically a bell curve, so the most likely real values are closest to the reported value. If you see a poll you don't like, say Obama at 42% and McCain at 47%, it's possible, but not very likely, however, that this really means 45/44 in favor of Obama.

The big problem with interpreting polls and the margins of error is that the stated MoE is in some ways just a mathematical abstraction which doesn't necessarily capture the full range of reasons to doubt the accuracy of a poll, like bad polling methodology or bad demographic weighting of the raw polling data.

[dbmk]

11. It usually means that there is a 95% chance of the real number beng within the given MOE

As someone already said :)

Given their sample sizes, I can not see how they can get as low as they get on the MOE.

The actual results also seem to indicate that they usually are misrepresenting it and giving a way to low MOE.

[BzaDem]

12. If Obama has 50 with a MOE of 3, there is a 95% chance that his real support is between 47 and 53

[DCBob]

14. There is still some confusion here. Some polls use this value in different ways but generally...

the MOE is not added and subtracted from both values to determine significance. It is only added or subtracted from one or the other. For example poll values of 48% and 43% with a MOE of +/- 4 are significantly different. Any value 4 or more, greater or less from either number is significantly different from the original value. You only apply the MOE to one number at a time.

[Bonn1997]

15. That's what I always thought too BUT

I remember seeing a pollster on TV say the MOE applies to both #s simultaneously (basically what other posters here are saying). I didn't think that was correct, though, because the two values (McCain's and Obama's) are not independent. I had always thought that the margin of error was applied to the difference between the #s for the two candidates. So if it's a 52 to 48 Obama lead in a poll with a MOE of 3, I always thought that meant Obama's lead was anywhere from 1 to 7 points. But that wasn't how I interpreted thiat pollster.

[DCBob]

16. You would rarely have significance if the MOE is applied to both values simultaneously.

Most of these polls have MOE of +/- 4. If applied simultaneously then the real MOE would be 8, so 47-40 would not be significant. That does not make sense.

[Teaser]

17. no, you're thinking of significance and power

significance level is the probability of rejecting the hypothesis that your test population and your reference population are the same when they are in fact drawn from the same distribution. Power is the probability of failing to reject a false hypothesis of identify when the populations are, in fact, different.