0
$\begingroup$

I have a table with 11 columns and 2 rows. Column 1 is the control group and I want to compare all the other columns with it. I'm not interested in compraing columns 2 - 11 with each other and I'm concerned that Bonferroni's correction (which seems to be the only option available) is overcompensating by multiplying the p value by 55 or however many comparisons it is. Is there something analogous to Dunnett's so that I can compare just with the control group? Or should I just do 10 chi-squared tests and multiply each p value by 10?

Also, when I do use Bonferroni'sin a chi-squared, is it possible to see the exact p values?

Thank you

$\endgroup$
  • 1
    $\begingroup$ You are right with your grievance, but it is not Bonferroni correction's problem; it is SPSS options problem. Unfortunately, SPSS doesn't allow to specify user-defined number of comparisons by hand, e.g. via syntax. So, you should just produce the table with uncorrected p-values and then multiply it according to the number of comparisons you actually imply. $\endgroup$ – ttnphns Jan 13 '14 at 13:38
  • $\begingroup$ @ttnphns, I think that's the correct answer, why not make it official? $\endgroup$ – gung - Reinstate Monica Jan 13 '14 at 15:39
  • $\begingroup$ Thanks. Is it possible to display the (corrected or otherwise) p values for multiple comparisons or will a series of 2x2 tables do the trick? I'm assuming I'm right that Bonferroni is as simple as multiplying the p value by the number of comparisons made? $\endgroup$ – Matthew Jan 13 '14 at 15:41
  • $\begingroup$ @Matthew, the Bonferroni correction is not equivalent to multiplying the observed p-value by the number of comparisons. That would quickly lead to adjusted p-values >1, which is non-sensical. Instead, you divide alpha (typically .05) by the number of comparisons. Eg, if you want to adjust for 5 comparisons, you would use .05/5 = .01 as your new threshold for 'significance'. $\endgroup$ – gung - Reinstate Monica Jan 13 '14 at 17:42
  • $\begingroup$ I'm with you. Altman, in his textbook, seems to multiply by n comparisons. With this particular table, the uncorrected p values were either way above anything anyone would consider significant (and so not worth correcting anyway) or extremely small (and so multiplying by 10 rendered something apparently sensible)--hence why it seemed to make sense. Is there a good text on multiple comparisons? $\endgroup$ – Matthew Jan 13 '14 at 21:08

Your Answer

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

Browse other questions tagged or ask your own question.