Sunday, August 09, 2015

Question on Margin of Error

When we see a margin of error in a poll, is it really the same for any points within a poll? In other words, if a poll has a 3% margin of error, does that mean that someone who has 5% in the poll has a 95% chance of having between 2% and 8%, or is the 5% 3% only valid for someone who gets 50% in the poll? I am under the impression that the further away you are from 50%, the smaller the margin of error should be, but I am not certain about this.

Any statistic geeks want to explain it to me?

That is all.


Anonymous said...

I'll try, but i am very rusty on math.

As n increases, a binomial distribution approximates a normal distribution, so i will sloppily use its math.

Standard deviation is sqrt (n * p * (1 - p))

Playing around with numbers, p = .5 and n= 30, sd = 2.73
And p = .05, n = 30; sd = 1.19

Art Deco said...

The statistic is an estimate of the actual value. It is conventional to make use of a '95% confidence interval', which is to say you've determined that you can say with 95% confidence that actual value is within the range you offer. (It's been 2/3 of a generation since I've studied it, and it was mostly statistics-as-a-social-science tool and not mathematical statistics, so do not ask me by what method you construct that confidence interval).