So… How do you design a poll?  Well, here is the standard formula for determining the representative, random, sample size for an entire population.

n =  N / (1 + (N * e^2))   

WHERE:

n = sample population – This is the number of voters you need to RANDOM sample – if it is not random your results are skewed( i.e. Pew, CNBC, Fox News, et al).  EXAMPLES:  If conservative people are less likely to respond to a poll, then your results are skewed liberal. If left brained people are less likely to reveal their thoughts, then you will oversample artsy folks… and under represent engineering types.  The electoral college system also skews the results.

N = Total number of voters in the nation - From the 2008 election this number is 131M (http://www.census.gov/newsroom/releases/archives/voting/cb09-110.html )

e = precision  (+/- %) We are going to call this 3%.  Elections are won by less than 1% but, you  should try to complete sampling based on 1%... good luck with that!

And, confidence level is 95%.   We want to be 95% sure our sample represents the entire voting population. The formula is based on 95%.  (There are other formulae; see http://edis.ifas.ufl.edu/pd006 .  But, this is the standard.)

THEREFORE:

n =  131,000,000 / (1 + (131,000,000 * 0.03^2))   = 1,111 registered voters

The next two tests come from education.

Is the question valid?  Example:  Is your pet dog black or white?  What if you don’t have a dog?  What if your dog is brown… or spotted?  You cannot ask a true-false or a yes-no question when a subjective (essay) answer is required.

Is the question reliable?    Does it test for the condition you are sampling?  One of the greatest problems with political poling is the delivery of the question: “Mitt Romney’s father released 12 years of his taxes.  Mitt Romney has released only two years of his taxes. Is Mitt Romney hiding something from the American people?”  As we all know, both sides of the presidential race are being accused of being dishonest. This question does not measure anything but emotion.

SO, where do the polls differ.  Well, they

a) are not so random

b) they fail to be precise (What good is +/- 10%?)

c) they don’t sample enough registered voters (or the respondents are not truthful)

d) they don’t ask a valid question

e) the responses don’t measure the condition being surveyed.

Sorry… I am a geophysicist… what do you want the answer to be?