A Significant Message

Significance: a danger word

Personal bug-bear or genuine message? Let’s go with with option 2. Significant when used in general conversation means “meaningful” or “important” or sometimes just “large”, but when you are reporting data it means that the observation is unlikely to be due to sampling error. Or, if you prefer, that your P value is below whatever critical value you decided.

Because of this potential for ambiguity, in science writing don’t use significant to mean anything other than statistical significance unless you preface it with whatever you want to emphasise e.g. “biologically significant”.

And now the kicker. Don’t say significant often at all, even if you are talking about statistics. I know this may not be what you learned in school, but there are three good reasons for this:

  • It’s not usually the confidence in your findings that is the most interesting, it is the findings themselves. Change your word order to emphasise the real-world importance.
  • You should be reporting the P value from your test anyway, which means you are already reporting statistical significance so there is no need to repeat yourself.
  • If you use “statistically significant” in every sentence then your results paragraphs will be really hard to read and absorb.

Of course, there is a final point. Often people consider the line they have chosen for statistical significance to be 0.05 without really thinking what that means. P values of 0.051 and 0.049 are almost identical, your confidence that the results are not due to type I error are just about the same.

Editing for Impact

If you use “significant” rarely, then you have the option to use it for impact. Being judicious with your application will mean that you can highlight where your data are particularly robust. You can use significant like an underline to say that you are confident the results reflect the true population.

“These data demonstrated that body mass index positively correlated with calorie intake (r=0.82, P<0.01) and proportion of processed food within diet (r=0.77, P<0.05), and negatively correlated with minutes of exercise per week (r=-0.79, P<0.01). However, surprisingly, there was also a statistically significant negative association with hours worked in the laboratory (r=-0.9, p<0.001)*.

Found this useful?

You might like the book.

*if only this were true.