An analogy to understand statistical power

The concept of statistical power is a slippery one. Here is an analogy that might help (courtesy of John Hartung, SUNY HSC Brooklyn).

You send your child into the basement to find a tool. He comes back and says "it isn't there". What do you conclude? Is the tool there or not? There is no way to be sure.

So let's express the answer as a probability. The question you really want to answer is: "What is the probability that the tool is in the basement"? But that question can't really be answered without knowing the prior probability and using Bayesian thinking. We'll pass on that, and instead ask a slightly different question: "If the tool really is in the basement, what is the chance your child would have found it"?

The answer depends on the answers to these questions:

•	How long did he spend looking? If he looked for a long time, he is more likely to have found the tool.

•	How big is the tool? It is easier to find a snow shovel than the tiny screw driver you use to fix eyeglasses.

•	How messy is the basement? If the basement is a real mess, he was less likely to find the tool than if it is super organized.

So if he spent a long time looking for a large tool in an organized basement, there is a high chance that he would have found the tool if it were there. So you can be quite confident of his conclusion that the tool isn't there. If he spent a short time looking for a small tool in a messy basement, his conclusion that "the tool isn't there" doesn't really mean very much.

So how is this related to computing the power of a completed experiment? The question about finding the tool, is similar to asking about the power of a completed experiment. Power is the answer to this question: If an effect (of a specified size) really occurs, what is the chance that an experiment of a certain size will find a "statistically significant" result?

•	The time searching the basement is analogous to sample size. If you collect more data you have a higher power to find an effect.

•	The size of the tool is analogous to the effect size you are looking for. You always have more power to find a big effect than a small one.

•	The messiness of the basement is analogous to the standard deviation of your data. You have less power to find an effect if the data are very scattered.

If you use a large sample size looking for a large effect using a system with a small standard deviation, there is a high chance that you would have obtained a "statistically significant effect" if it existed. So you can be quite confident of a conclusion of "no statistically significant effect". But if you use a small sample size looking for a small effect using a system with a large standard deviation, then the finding of "no statistically significant effect" really isn't very helpful.