I am running a paired samples t-test. n=29. Should I raise the alpha from .05 to .1? I was told this can increase statistical power for this sample size but should it be done?

Currently my paired sample t-test says, t(28) = 1.593, p = .061, one-sided. cohen's d 5.28835.


Thanks so much

  • 5
    $\begingroup$ Those are considerations that you should make before making your analysis. I recommend you check stats.stackexchange.com/search?q=p+hacking $\endgroup$
    – Tim
    May 15, 2021 at 16:47
  • $\begingroup$ You should set $\alpha$ before you look at the data based on your tolerance of false positives. If your experiment leads to to think you should have asked a different question (including asking the same question a different way) then you should ask it the new question with new data. If you are not getting enough power, your results next time will be more convincing to others if you use a larger sample size rather than shifting $\alpha$ beyond usual practice (you have already started down that road with a one-sided test) $\endgroup$
    – Henry
    May 15, 2021 at 16:55

1 Answer 1


I would have to agree with the comment that Tim made, and the link provided. Another consideration is what seems to be the "norm" in your field of study. As an ecologist, I've seen many instances in which alpha = 0.10, and authors declare they found a significant result at the p-value you reported. However, in other cases, you can keep alpha = 0.05, and report your p-value as a "trend" toward significance, or "marginally insignificant."

Here is an exhaustive list of the many other phrases scientists have used throughout the literature: https://mchankins.wordpress.com/2013/04/21/still-not-significant-2/

  • $\begingroup$ Thank you so much for your reply. This is in psychology/psycholinguistics. $\endgroup$
    – midori
    May 15, 2021 at 17:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.