Summary of "The Reign of the
p-Value is Over: What Alternative Analyses Could
We Employ to Fill the Power Vacuum?"
- Hasley, L. G. (2019). The reign of the p-value
is over: what alternative analyses could we employ to fill the power vacuum?
Biology Letters, 15. doi 10.1098/rsbl.2019.0174
http://doi.org/10.1098/rsbl.2019.0174
p values, all by
themselves, are not very informative, especially when they are treated
dichotomously ("significant" or not). They give you no estimation of the
size of the effect, or how much error there is in that estimate. p
values can vary a lot from one sample to another. Getting significance in
one sample is no guarantee that you are likely to obtain significance when
testing another sample from the same population. Lewis has several suggestions
to help address the problems with p values.
- Put a confidence interval on the p value to show how much
uncertainty there is in the estimation of the p value.
- Put a confidence interval on the expected value of p were you to
repeat the analysis on a new sample of data
- I have an Excel spreadsheet to calculate these confidence intervals.
My students can find it in BlackBoard. Others can request it by
email.
- Report the Bayesian estimated false positive risk, the probability that
your rejection of the null hypothesis is actually a Type I error.
- Report effect sizes with confidence intervals.
- Report the Bayes Factor Upper Bound -- an upper bound for the how much
more likely the alternative hypothesis is true than is the null hypothesis.
Back to
Wuensch's Stat Help Page