Physicist Amok

Stranger Fruit points out a poll on belief in evolution/creation at USA Today. The numbers are disturbing, but what we really learn from this is that the folks at USA Today don’t set very good polls. There’s a common problem in software engineering: what users tell you they want bears no resemblence to what they actually want. However, a person’s beliefs should influence their actions and decisions, so to measure belief, it’s better to measure actions and decisions than to ask about beliefs. A better approach would have been:

1. Humans and bacteria share a common ancestor (True/False)
2. Homo sapiens first appeared on Earth more than 10,000 years ago. (True/False)
3. Would a presidential candidate declaring that they DO NOT believe that bacteria and humans have a common ancestor make you more or less likely to vote for him? (more/no difference/less)

The first two questions immediately establish whether a person believes in evolution. The second question adds another tidbit of information: if you answered yes to 1 and no to 2, then we can conclude that you don’t understand numbers.

I took out the questions about how familiar the respondant is with evolution and creation, since I can think of only one interesting measurement I could make off of them with any confidence, namely the distribution of these quantities conditioned on their answer to question one, and it’s not particularly informative. If I were to ask them, I’d phrase them as:

• How would you rate your knowledge of evolution? (Nonexistent/Weak/Strong)
• How would you rate your knowledge of creationism? (Nonexistent/Weak/Strong)

I would also want to add some additional questions to establish a baseline for the respondant on this scale. The formulation of such a set of baseline questions is a research project in its own right, though perhaps someone has already done it.

You’ll notice that I took all the fine gradations out. If anyone has some research on how well such gradations work, I’d like to know about it. My unsubtantiated impression is that humans are extremely bad at fine discrimination, while they’re fairly accurate in big, robust blocks.