Science - October 17, 2019

‘The p-value should not be all-important’

Coretta Jongeling

PhD student Jeroen Berg struck lucky. Or so he thought. Right at the start of his research he discovered a crucial gene in cucumbers. Sadly, replication proved the result insignificant. Which is why his proposition is:

PhD candidates are required to submit a few propositions with their thesis. In this feature, they explain the thinking behind their most provocative proposition. This time it’s the turn of Jeroen Berg, who was awarded his PhD on 11 October for his study on resistance to mildew in cucumbers.

Proposition: The term statistical significance, which is based on an arbitrarily chosen p-value threshold, should be banned from science.’

‘There is always a chance that you will discover something that is actually based on coincidence. The scientific community has agreed that the chance of this must not be more than five per cent, hence the p-value of 0.05. That means that one in 20 conclusions in a study are not actually true. That seems to me a rather random threshold. What is the difference between a p-value of 0.049 and one of 0.051?

This is a bit hypocritical, of course, given that I’ve written about 20 times in every chapter of my thesis: “p is this big, so this is significant”. You can’t leave it out either, because then the reviewers come down on you like a ton of bricks. It’s built in to the system.

There is a journal, Basic and applied social psychology, which has completely abandoned p-values. I think it’s fine to mention the p-value, but you shouldn’t let so much depend on it. It is more important to look at what the effect you have measured could mean. The gene I was researching did turn out to be important, but not in the way I first thought it was.

It is more important to look at what your result could mean

Researchers should check their own data much more often, and try to replicate studies. But that takes a lot of time and money, two things that are in short supply, sadly.’