Stats Hacks #4

Playing a little catch-up here … Hack #4 is: Reject the Null.

Scientific method is all about taking guesses and then finding ways to prove them in the real world, through experimentation. Sometimes, though, it is just as effective to prove that a guess is wrong as it is to prove that it is right. And usually, it is a whole lot easier. When a guess is made in opposition to what a researcher thinks is true, it is a null hypothesis. Proving a null hypothesis is false is very helpful to proving the actual research hypotheis is likely to be true.

As I have flashbacks to fifth grade (“Michelle doesn’t not like you” … “Solid! I’ll ask her to the soc-hop.”), I’ll restate this in a more outcome-driven way. Science is benefited when research thinks a hypothesis is right and the data supports it. It is also benefited when research thinks a hypothesis is wrong, and the data agrees. Anything else, and that academic paper doesn’t seem quite as sexy.

In statistics, the modus operendi is usually the null hypothesis. Statisticians take the research hypothesis, turn it upside-down, and then go about trying to prove through number-crunching that that the null can’t possible be true. In Bruce Frey‘s words, “statisticians make a statement about whether a hypothesis opposite to the research hypothesis is likely to be correct.” It is called a “null” because the hypothesis opposite the research hypothesis is typically a statement that no relationship exists between variables. Nada = Null.

This is effective because proving something is true is a lot more difficult than proving something is false. For something to be true, it has to always be true. For something to be false, it really just has to be false one time. Intelligent Designers are waiting for things to fall up to disprove gravity, for instance. If they just get it to happen once, it takes a big bite out of Newton’s apple. (Well, not entirely true, since any hypothesis — including a null hypothesis — has to be testable, and testable means it has to be able to happen more than once.)

Also: ,

Some definitions:

hypothesis

an unsubstantiated but testable guess about how the world works

research hypothesis

a guess by the researcher about how the world actually works, posed as a statement that a relationship exists between variables

null hypothesis

the opposite of the research hypothesis, posed as a statement of no relation between variables