Question for anyone trained in statistics…
Is the margin of error reduced by combining multiple polls? Isn’t the margin of error a result of sample size? Hypothetical example:
If sampling 500 people results in a 5 point margin of error, 1000 people 4 points and 1500 people 3 points, wouldn’t combining the results of 3 polls using 500 people have a margin of error of 3 points?
You cannot take dissimilar polls and combine them to create a larger sample size to decrease the MoE. Now from the same polling organization over close periods of time this is OK, as tracking polls are done this way. For instance, the Washington Post tracking poll has around 1,200 likely voters, with around 400 'new; poll respondents added each day, with 400 of the 'older' poll respondents dropped from the survey. This allows the polling firm to maintain their desired sample size.
Here are some tools to help you with MoE calculations. First is the US Census Bureau tables with voting age population, registered voters and percentages. From this you can see that in 2002 out of a 18 year or older population of 210,421,000 there were 66.5% of said population who are both registered voters AND citizens of the United States. This gives us N=139,929,965 for 2002 which is good enough for MoE calculations for 2004.
Source: U.S. Census Bureau, Voting and Registration, Historical Times Series Tables
Here is a Margin of Error calculator located at the American Research Group web site. You put in 139,929,965 as the Population size, put in the sample size of the poll in question in the sample size box, hit the Calculate button and you have the answer for the margin of error.
Source: ARG Margin of Error Calculator
Hope this helps,
dvwjr
I understand, I think.
dvwjr, you’re saying that from a scientific perspective, dissimilar polls can’t be combined to reduce the MoE. And comebacknewt points out that because most polls over time are reporting very similar results, the results are probably true, just statistically ummeasurable.
So the real uncertainty of the results is smaller than the MoE, we just can’t measure it precisely? Wouldn’t that mean that the claim that “polls show the race in a dead heat” is only technically correct but misleading?