LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous (more recent) messageNext (less recent) messagePrevious (more recent) in topicNext (less recent) in topicPrevious (more recent) by same authorNext (less recent) by same authorPrevious page (June 2009, week 5)Back to main SAS-L pageJoin or leave SAS-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Mon, 29 Jun 2009 12:28:11 -0400
Reply-To:     Susan Durham <sdurham@BIOLOGY.USU.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Susan Durham <sdurham@BIOLOGY.USU.EDU>
Subject:      Re: Fisher's Protected LSD versus Tukey's HSD
Comments: To: Kevin Viel <citam.sasl@GMAIL.COM>

On Mon, 29 Jun 2009 10:12:39 -0400, Kevin Viel <citam.sasl@GMAIL.COM> wrote:

>Greetings, > > I performed a one-factor ANOVA with approximately 8 levels and 100 >observations (unbalanced). The omnibus F test suggested that at least one >mean is different. An investigator requested Fisher's protected LSD. >Originally, I used the Tukey's HSD. > > For a few comparisons, there is a difference. Does anyone have an >opinion concerning which test is more appropriate? > >Thank you, > >Kevin >Atlanta, Georgia, USA

Such a can of worms!

A useful reference for the standard multiple comparison methods is

RW Day & GP Quinn, 1989, Comparisons of treatments after an analysis of variance in ecology, Ecological Monographs 59(4), 433-463.

On pp450-451, Day and Quinn say "Carmer and Swanson (1973) and others have claimed that the Fisher's protected LSD test controls the EER [experiment-wise error rate] at the specified level by virtue of the preliminary overall F test. This will only be true when there are no real differences between treatments, as shown by Carmer and Swanson's (1973) own simulation results.... Where some mean differ, the protected test functions as a simple LSD test for comparisons within groups; it has a per-comparison type I error rate...."

In contrast, Tukey's HSD does control EER, although conservatively. Day & Quinn's simulations indicate that Tukey's comes in at about half the nominal rate in some comparisons (see their Fig. 1), so it is possible to have an overall test significant at <0.05 but no pairwise comparisons significant at <0.05, as Peter notes in his response to your post.

A choice between Fisher's protected LSD and Tukey's HSD is a choice between apples and oranges: the former controls comparison-wise Type I error, and the latter controls experiment-wise Type I error. Since you originally used Tukey's, you can argue that you've chosen to control EER, which the protected LSD doesn't do.

Cheers, Susan

--- Susan Durham Ecology Center Utah State University Logan, UT 84322-5205


Back to: Top of message | Previous page | Main SAS-L page