Monday, October 24, 2016

Lies, Damned Lies, and Statistics

Here's a piece at Zero Hedge showing a Podesta e-mail describing how polling data can be fudged with a minimal chance that the publisher would get caught. Early on the point of polls is to encourage your own people and discourage the other side and if you can gin up a 12 point lead, that surely does the job. This is why when polls are published, it behooves the public to know at least some rudimentary math and maybe a bit about the makeup of their own country before breaking out the champagne or the seppuku knife.

If the polling sample breaks out 36-27-31 for D, R, I, that might be valid for parts of upper Manhattan, but not for the nation as a whole. The Dems have a slight lead on registered voters with their very high hold on blacks and a virtual lock on the prisons and graveyards, but the enthusiasm just doesn't seem to be there this year. Changing the break to say 34,33,31 will give a much more likely result and if it takes a 9 point bias to get Hill a 2 point win, a one point bias is going to look like  a prison sentence.

Update: And what do you do with dishonest polling data?

No comments: