Monday, August 8, 2016

New Georgia Poll

There's a new poll out on Georgia.

Ignore it.

Well, kinda sorta ignore it.

Here's a brief AJC story on the poll, and if you're more interested go here for a better breakdown from the polling firm itself. The AJC story fails to mention this is a robo-poll, an automated phone survey that probably -- though it's unclear -- called only landline telephones (by federal law you can't robo-call cell phones).

Robo-polls are notoriously flawed, so be skeptical. How skeptical? 538's new Georgia forecast lists the poll but grades the polling firm only with a "C" for historical accuracy and quality of methodology, which is damn low. 538 weights polls in its overall average, with extra points for being more recent. Even with this new poll's recency, it's weighted only 0.49 while a poll from last week by the AJC (that used humans calling landline and cell phones) gets weighted 0.65 in 538's overall forecast. That should tell you something.

Back to that new Georgia poll. It gives Hillary Clinton a 44-37 advantage over Donald Trump. The margin of error is 4 percent so, technically, this difference falls just within being considered a statistical tie. That said, the handful of recent polls all point to a slight Clinton lead, all within the margins of error. Georgia may indeed be in play, which is good if you're a political junkie and bad if you hate to see the airwaves filled with political ads.

Note the breakdown provided does not include age. Often you end up with far too few young respondents in a robo-call to landlines. It's unclear that the data were weighted in any way.

No comments: