Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
correction for multiple comparisons when testing many Pearson correlations for significance
1,906 views
Skip to first unread message
Esther
Aug 10, 2014, 2:44:36 PM8/10/14
to
I know that SPSS provides many types of corrections when performing ANOVA (Bonferroni, plus others harder to calculate). Are similar corrections available for the situation of performing many Pearson correlations in the same experiment? I want something more lenient than Bonferroni, but cannot calculate these by hand.
If not in SPSS, even a table could help - I saw a reference to a table of corrections for multiple significance tests for correlations, but could not find it.
Thank you.
If not in SPSS, even a table could help - I saw a reference to a table of corrections for multiple significance tests for correlations, but could not find it.
Thank you.
Rich Ulrich
Aug 10, 2014, 11:32:58 PM8/10/14
to
On Sun, 10 Aug 2014 11:44:36 -0700 (PDT), Esther <aliz...@gmail.com>
wrote:
>I know that SPSS provides many types of corrections when performing ANOVA (Bonferroni, plus others harder to calculate). Are similar corrections available for the situation of performing many Pearson correlations in the same experiment? I want something more lenient than Bonferroni, but cannot calculate these by hand.
>If not in SPSS, even a table could help - I saw a reference to a table of corrections for multiple significance tests for correlations, but could not find it.
>Thank you.
For "something more lenient than Bonferroni", I think
you must be thinking of False Discovery Rate which
was introduced by Benjamini and Hochberg. See
http://en.wikipedia.org/wiki/False_discovery_rate
The first Google hit for <SPSS FDR> is
http://www-01.ibm.com/support/docview.wss?uid=swg21476447
which gives a formula.
The third hit is from the SPSSX-L list and Bruce Weaver
points to code, and describes how to simplify it.
http://spssx-discussion.1045642.n5.nabble.com/SPSS-syntax-for-Benjamini-Hocberg-FDR-procedure-td5714855.html
Be aware that there are not all FDRs are created equal -
there are legitimate reasons for making stronger sets
of assumptions about the patterns expected -- most of
these are for such purposes as code-breaking and pattern
detection-- and you can get tests with r=0.0 promoted
to "probably p < 0.05" if you choose inappropriately.
--
Rich Ulrich
wrote:
>I know that SPSS provides many types of corrections when performing ANOVA (Bonferroni, plus others harder to calculate). Are similar corrections available for the situation of performing many Pearson correlations in the same experiment? I want something more lenient than Bonferroni, but cannot calculate these by hand.
>If not in SPSS, even a table could help - I saw a reference to a table of corrections for multiple significance tests for correlations, but could not find it.
>Thank you.
you must be thinking of False Discovery Rate which
was introduced by Benjamini and Hochberg. See
http://en.wikipedia.org/wiki/False_discovery_rate
The first Google hit for <SPSS FDR> is
http://www-01.ibm.com/support/docview.wss?uid=swg21476447
which gives a formula.
The third hit is from the SPSSX-L list and Bruce Weaver
points to code, and describes how to simplify it.
http://spssx-discussion.1045642.n5.nabble.com/SPSS-syntax-for-Benjamini-Hocberg-FDR-procedure-td5714855.html
Be aware that there are not all FDRs are created equal -
there are legitimate reasons for making stronger sets
of assumptions about the patterns expected -- most of
these are for such purposes as code-breaking and pattern
detection-- and you can get tests with r=0.0 promoted
to "probably p < 0.05" if you choose inappropriately.
--
Rich Ulrich
Bruce Weaver
Aug 16, 2014, 3:43:05 PM8/16/14
to
http://udel.edu/~mcdonald/statmultcomp.html
HTH.
--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM."
0 new messages