Saturday, November 15, 2014

"Herding" of poll numbers

Nate Silver of fivethirtyeight has a blisteringly thoughtful article out about the reason why the 2014 American election polling results were so off the mark.

Silver picked up on a hunch by Harry Enten, in which Enten published his theory that the pollsters were “herding” each other toward an average result rather than doing their job and coming up with statistically valid yet completely independent results.

Especially noteworthy was Ann Selzer’s final poll number for the Iowa  US Senate race; her results were far from the clustered average, and other pollsters let her know that she was losing her touch.  Her figure was outside the standard deviation from the other polls – and yet closest to the true election night figures!

Silver makes a compelling case that the pollsters watch the poll-of-polls numbers and herd each other toward that average so that their own work looks respectable.  Here’s the link:


Afterword by the Blog Author

Tinkered and altered poll numbers look and act a lot like bad modeling of data!

No comments:

Post a Comment