CNN reporting an ABC News Poll: Harris up by 6 Points

[BlueInID]

28. Misinterpretation of Confidence Intervals

TLDNR: A confidence interval is a measure of repeatability, not a measure of accuracy.

This is a bit stats-nerdy, but this is a misinterpretation of what a confidence interval is. It's based in frequentist statistics and addresses the probability of finding the same result were the experiment (in this case, sampling opinions) to be repeated. In the context of polling, and assuming a sample size of 2000, with a 52% favoring Harris. The plus/minus 3% defines the interval (49 - 55) that we would expect to contain the estimate derived from 95/100 independent samples of 2000 and _using the same methodology_.

So treating a poll's margin of error as a predictor of the true value, which would be the percentage of folks in the whole US who favor Harris at the time the sample was drawn (for a national poll like this one), is a misinterpretation of what the CI actually estimates.