Forums | OpenR

14 April 2022, 11:34 PM

As preparing for group presentation of Bonferroni t-test, I met a problem: How exactly the pairwise.test figure out or calculate the result p-value numbers. 

My understanding is that when we apply p.adj = 'none', the result would be the t-test p-values for each pair, and  when apply p.adj = 'bonf', the result would different from the former.

But as I try to use t.test() in R to verify the p-values, and I got different results from pairwise.test( p.adj = 'none') and I donot know why, should they be the same?

And as for results from pairwise.test( p.adj = 'bonf') , it is more confusing how the result numbers are calculated. 

Could you please help me with this question? Many thanks.

my code for comparing the pvalues from t.test and  pairwise.test( p.adj = 'none') is below.

data(airquality)

#test for ozone value for month 5 and 7
var.test(airquality$Ozone[airquality$Month == '5'],airquality$Ozone[airquality$Month == '7'], 
         alternative = "two.sided", ratio = 1)

#variance is equal
t.test(airquality$Ozone[airquality$Month == '5'],airquality$Ozone[airquality$Month == '7'],
       var.equal = TRUE, 
       alternative = "two.sided", mu = 0)

#test for ozone value for month 5 and 9
var.test(airquality$Ozone[airquality$Month == '5'],airquality$Ozone[airquality$Month == '9'], 
         alternative = "two.sided", ratio = 1)

#variance is equal
t.test(airquality$Ozone[airquality$Month == '5'],airquality$Ozone[airquality$Month == '9'],
       var.equal = TRUE, 
       alternative = "two.sided", mu = 0)

#pairwise.test, but use 'none' for adjustment.
pairwise.t.test(airquality$Ozone,airquality$Month, p.adj = "none")

15 April 2022, 11:22 AM

You have to specify not using pooled sd and the variances are equal:

pairwise.t.test(airquality$Ozone,airquality$Month, pool.sd = FALSE,var.equal = TRUE, p.adj = "none")

See the help document for details.

15 April 2022, 11:49 AM

Thanks for replying! It solved and the result matched when applying no adjustment in pairwise.test. 

And as for the bonferroni adjustment I figure out in R.