My former student Jared Gordon writes, apropos my post complaining about CNN’s reporting of two polls:
In your recent blogs on poll reporting at CNN, you made the statement “Someone at CNN must understand this better than most readers do.” Having worked on the TIME/CNN poll with Yankelovich, Inc. prior to law school (now the Harris Interactive TIME/CNN poll), I can confirm for you from personal experience that in fact several people on their political reporting staff are well-versed in polling statistics, in particular Keating Holland, the person who was then, and I believe still is, in charge of CNN’s polling. In fact, Mr. Holland routinely ran much more complicated analyses than a simple determination of the margin of error.
Because of my previous experience with the very competent polling staff at CNN, I have been considerably distressed by CNN’s recent reporting of public opinion polls, not because they are ignorant of the margin of error, but because they seem to be willfully ignoring the margin of error to create news. While a story that says “there’s no measurable change in the electoral picture, based on the latest poll” doesn’t generate much excitement, a story that cites a poll for who’s currently winning the presidential race draws considerably more eyeballs. I worry that CNN has given up on reporting the true stories in their surveys, and have instead decided to create stories in their surveys that they wish were there for purposes of improving their ratings or pageviews.
I would also note that in addition to the margin of error reported in the CNN story you referenced, there are other, less detectable, errors that result from the methodology used to determine “likely voters.” Every major poll has their own methodology for determining likely voters, and the methodology may even change from survey to survey, based on an attempt to conform to the percentage of the eligible voting population that the polling firm expects to vote in the next election. Thus, likely voter comparisons are not necessarily comparable from poll to poll, or even within a time series of surveys from the same poll. Other aspects of the way in which likely voters are chosen introduce other kinds of potential errors. These errors aren’t reported largely because they aren’t easily quantifiable errors.
Comments are closed.