Tuesday, July 15, 2008

Fun With Biased Polling

I recommend my statistics series for a deeper discussion on what polls mean. You can find it on the website given at the top of the blog.

But for a very, very bad example of polling, Lou Dobbs can't be beat. Here's today's question:

Do you find it outrageous that the Bush Administration refuses to give our Border Patrol agents the respect and support they deserve?

Internet polls of this kind are never based on random sampling, because those who respond are the ones who care enough to respond. Those who don't care that much will not respond. Besides the people who read, say Dobbs' website, are not a random sample of all Americans to begin with.

All this means that the polls on various websites tell us nothing what voters or web-users or Americans and so on think; only what those people who answered think. It's also quite possible to "freep" those polls so that one person can vote many, many times.

In short, such polls are garbage. But this particular poll would be garbage even if it used a random sampling scheme, because it employs a leading question. The word "outrage" leads the voter towards what he or she is supposed to feel about the lack of "respect" and "support" the Border Patrol agents "deserve." So you are told how you should think and then you are asked to agree with that, pretty much.

One reason why even proper polls on various topics (such as opinions about abortion) so often disagree in their findings is that the way the questions are worded does matter.
Added on July 16: Check out this FAIR post about how wording changes the conclusions of a poll even when the changes are quite subtle.