IBM Support

Does SPSS Statistics offer multiple comparisons using the Benjamini & Hochberg method to control the false discovery rate?



Bejamini & Hochberg (1995, Journal of the Royal Statistical Society, Series B, Volume 57, No. 1, pp. 289-300) introduced a simple method of handling the multiplicity problem in statistical testing that controls the false discovery rate rather than the familywise error rate. Does SPSS offer this method?

Resolving The Problem

The following commands will compute the desired results according to Benjamini & Hochberg's method. They assume that you have an SPSS file containing one case per p value, with a variable named p holding the p value or significance level of interest for each comparison. Make sure that there are no other cases in the data file, as the number of cases in the file is used to count the number of comparisons involved.

sort cases by p (a).
compute i=$casenum.
sort cases by i (d).
compute q=.05.
compute m=max(i,lag(m)).
compute crit=q*i/m.
compute test=(p le crit).
compute test=max(test,lag(test)).
formats i m test(f8.0) q (f8.2) crit(f8.6).
value labels test 1 'Significant' 0 'Not Significant'.

The primary result of interest is the variable test, which will have a value of 1 for all significant comparisons, and 0 for nonsignificant comparisons. These commands use a .05 level for the false discovery rate. You can change that by changing the value of q in the second COMPUTE command.

This has been submitted as an Enhancement Request to IBM SPSS Development and will be considered for a future release of IBM SPSS Statistics

[{"Product":{"code":"SSLVMB","label":"SPSS Statistics"},"Business Unit":{"code":"BU001","label":"Analytics Private Cloud"},"Component":"Not Applicable","Platform":[{"code":"PF016","label":"Linux"},{"code":"PF014","label":"iOS"},{"code":"PF033","label":"Windows"}],"Version":"Not Applicable","Edition":""}]

Historical Number


Document Information

Modified date:
16 June 2018