Skip to main content

View Diary: How the Associated Press used their poll this week to mischaracterize the race for the White House (170 comments)

Comment Preferences

  •  About outliers.... (0+ / 0-)

    It's important to note what an outlier is.

    An outlier is not a datapoint outside the MoE.

    Sampling error is distributed (more or less) normally - in a bell curve.  Most of the time, the result will be in the hump of the bell.  Some of the time it will be a bit outside the hump.  Sometimes it will be way out on the flanges.  None of these are "outliers", and as the MoE is based on 95% confidence interval, one in 20 polls will be outside their MoE.

    An outlier is a poll that is weird for some other reason - a mistake in the counting, sampling from the wrong population, whatever.

    And "house effects" are not randomly distributed, and not necessarily normally distributed either.  You can't assume that the firms with house effects near 0 (i.e. close to average) are more accurate than firms with big house effects.  It could well be that all firms are biased in one direction, and that the firm that most consistently favours Obama (or Romney) is the leased biased.

    Just thought it was worth noting...

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site