I have a dataset with 700ish samples of survey results. Things like height, weight, etc and someone wants to know if there is a statistical difference between the people who said yes to question A (~300) and those who said no (~400).

My initial thought was a 2 sample t-test between the two subgroups, but should I be using a paired t-test instead? (or should I be using something else entirely?)

Also, they want me to tell if there is a statistical difference across 17 variables, and I said I would need to account for false significance with the bonferroni correction to the .05 p-value threshold. Was that the correct correction?

  • 1
    $\begingroup$ Maybe too many issues here for one Q // What foundation for 'pairing' between subjects who Yes to A and No to A? // As for use of a two-sample t test, are data close to normal? If these are Likert scores, then it might even be somewhat controversial to treat ordinal categorical scores as numeric. Two-sample Wilcoxon test might be appropriate. // Bonferroni correction for 17 tests might be over-conservative (require too small a P-value for rejection). A more detailed discussion of how the 17 tests arise might enable someone here to suggest a more feasible method of avoiding 'false discovery'. $\endgroup$
    – BruceET
    Nov 17, 2020 at 23:26
  • 1
    $\begingroup$ They're results from a medical survey about whether or not they told their health provider about certain types of pain. I didn't think about normality, I will use a non-parametric test! Most of the statistics are body/health metrics. The 17 tests is also a bit missleading, as its 12 on the question A, then another 4 on a subset of the people who said yes to A. $\endgroup$ Nov 18, 2020 at 0:30


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.