The
Bonferroni correction is a stalwart of statistical and empirical reasoning. Statistics has its flaws and its benefits. Statistics are everywhere but not always understood. Statistics are used to answer research questions...but they can sometimes be employed in an incorrect fashion or in a very
BIASED fashion. Mark Twain said, "There are lies, damn lies, and statistics."
The Bonferroni correction is used to account for increased
experimentwise error rates when testing multiple hypotheses. Experimentwise error rates are used to describe the increased chances of committing a
Type I error when running multiple chi-squares, t-tests, ANOVAs, and other statistics concurrently. You are simply more likely to detect statistical significance
by chance with the more statistical tests that you run.
The Bonferroni correction keeps researchers
HONEST in regards to reporting significant main effects of
clinical merit. It further deters researchers from making
erroneous conclusions based on large sample sizes and implausible effect sizes.
In order to calculate the Bonferroni-corrected alpha value to achieve statistical significance when testing multiple hypotheses concurrently,
divide the alpha value of .05 by the number of hypotheses you are testing. So, if I was assessing the differences between men and women on four (4) different outcomes, (.05 / 4) = .013. This means that the inferential statistic for any of our four outcomes would have to be less than .013 to be statistically significant (rather than just being lower that the normal .05).
Publications have caught on to the utility and relevance of the Bonferroni correction. Some journals specify its use in the
author guidelines and will
reject manuscripts automatically if the correction is not used for multiple hypotheses.
In conclusion, use the Bonferroni!
