How Not To Do a Poll in One Example

There is a new poll out in Georgia. The poll has a ±3.4% margin of error. It is of 803 voters. But it is not of just any 803 voters. It is of 803 voters from a list of voters “determined likely to participate in the 2016 primary.” In other words, it is from a registered voter list of people, most likely, who have voted before.

That forces voters into a predetermined bracket out of the gate, which pollsters should not do.

But wait. It gets better.

The 803 people were all contacted “on the evening of January 17th, 2016 using an interactive voice response system.”

Yes, that would be a robo-poll, not a live operator poll.

But really, the most damning thing about the poll is that it was conducted in a Southern state on a Sunday evening when a lot of people are in church or with family.

In other words, the polling by default generated a poll sample of people not really indicative of who will actually vote in the Republican Primary.

But a news organization wanted the poll and so it will be big news because a news organization wants it to be big news.

By the way, it has Trump and Cruz ahead of all other comers, but still . . .

About the author

Erick Erickson

View all posts