26 September 2014

What you need for strong hypotheses

Given how important hypothesis testing is in science, I am continually frustrated by how much trouble students have in making good, strong hypotheses ones.

When I ask students, even our graduate students, “What’s your hypothesis?”, the answer often start with something like, “I’m studying,” “I’m looking at,” or, “My question is...” Those are not hypotheses.

Some more advanced students (particularly in proposal seminars) will say, “My hypothesis is that my control group will be different than my experimental group.” Okay, you’ve learned the concept of the null and alternate hypotheses. That’s a useful thing to understand for statistical analyses. I suppose “experiment will be different than the control” that counts as an hypothesis, but it’s such a weak one.

That’s not hypothesizing, that’s just hoping.

Strong hypotheses have two components:

First, strong hypotheses incorporate some kind of mechanism, whether implicitly or explicitly. A strong hypothesis is based on investigating causal mechanisms. Without that, you have a fishing expedition.

Second, because strong hypotheses are based on assumptions about mechanisms, they make explicit predictions. Do you think the mean of the experimental group will be higher or lower than the control group? Both are valid alternate hypotheses under the “null versus alternate” scheme, but they are not the same. It’s even better if you can predict the magnitude of the difference.

No comments: