Bonferroni correction

[Paolo Costa]

Hi everyone, I'm comparing for a summary statistic analysis a bunch of variables among 5 groups. All are categorical variables, so in the table1 are displayed p values from chi2 test. Nonetheless, for a such a number of groups, I think it would be more correct to "adjust" the p values by using a Bonferroni correction. Am I right? If I correctly understood how the correction works, I can simply multiply p values * 5 so my new alpha cutoff will be:0,05*5 = < 0,25. The p value that I obtained are significant if irrespective of Bonferroni correction are still below the "true" alpha of 0,05, right?

Thank you everyone.

[Bruce Weaver]

Hello Paolo. What is the context? You mentioned "Table 1", which makes me wonder if some multivariable analysis comes later, and if the so-called "Table 1 fallacy" is relevant. Thanks for providing more info.

PS- If some kind of correction for multiplicity is really needed, I think you can find a method that is far less conservative than a Bonferroni correction. Bonferroni-Holm, for example.

[Rich Goldstein]

if by "Table 1" you are referring to a table of descriptive statistics, then I would argue that a multiple comparisons correction is not only not needed, it is not even desirable

[Paolo Costa]

Bruce: by table 1 I mean the table of summary statistic presented at the beginning of an analysis. Thank you for replying me

[Carlo Lazzaro]

Paolo:
in addition to previous helpful replies, please note that summary statistics are usually descriptive (so, no p-value and related stuff).
That said, if you want to go inferentilal, and you have, sa, 5 multiple comparison at your arbitrary cut-off is 0.05, you should diveide it by 5 (0.05/5=0.01). Every resulting p-value<0.01 shows that the result of your comparison reaches statistical significance at 0.05.

[Maxence Morlet]

In addition to the very helpful response, you may want to check out

ssc install rwolf2