I could be mistaken in my reading of this, but don't these calculations depend on whether the "margin of error" is based on the measured
individual favorability of each candidate as opposed to the measured
difference between them, what we usually refer to as the "lead?" In this Wisconsin poll, if what is being measured is the lead held by one candidate (here, Biden), and that lead is 17 points with a MOE of +/- 4 points, then wouldn't the pollsters be saying with 95% confidence that the true margin is between 13 and 21 points, rather than between 9 and 25?
Seems like the way you get to the 9-to-25 margin is by looking at the
individual numbers of each candidate. So in the Wisconsin poll at issue, Trump was at 40%. If you apply the +/- to his numbers individually, that would mean his support lies between 36 and 44%. Biden, who came in at 57%, would have his support lie between 53 and 61%. Only taking Trump at his worst and Biden at his best would yield a 25 point margin (61-36) while taken Biden at his worst and Trump at his best would leave you with a margin of 9, that being 53-44%. This is kind of a double counting of the margin of error -- applying 4 points to Biden
and 4 points to Trump.
Is that really what the numbers mean when they are reported? Or is it actually that the margin of error should be considered once, not twice, that is, apply it to the measured
margin between the candidates?
Here is an article on this issue:
https://www.pewresearch.org/fact-tan...lection-polls/