Hi, Please forgive cross postings - I wasn't sure which group was most appropriate for this.
I'm using a 2-treatment by 3-period cross-over design, Design 4.1.1 described by Ratkowsky et al. (1993) on page 122, to examine treatment and carryover effects during the latter 3 periods of a 4 period feeding trial. This is an AABB (for sequence 1), BBAA (for sequence 2) trial, where response data from the first period were not used in the analysis. The reason for only considering the last 3 periods of the 4 period trial was: my experimental treatments were two commercial fish feeds fed to tanks of fish. Because it is not possible to apply no treatment before the beginning of the experiment (that would involve not feeding then re-feeding the fish, which would cause strong carryover effects), the first period was designated an "acclimation period", during which the fish were fed whatever feed they were scheduled to be fed during the second period of the experiment. The idea was to allow any carryover effects caused by the feeding practices before the experiment (before period 1) to dissipate by the end of the first period. My very basic question is: Ratkowsky et al. examples always assume that there is no carryover effect in the first period of the experiment to be analyzed. Is there any reason in terms of the statistics or the SAS model why I shouldn't list a potential carryover effect in my data (column 5 of the data below would be all zeros in a Ratkowsky et al. example)? The set up of my data and SAS code, which are below, follow Ratkowsky et al., p 126-127. Thank you in advance for any comments or suggestions! Laurel Ramseyer [EMAIL PROTECTED] data aABB_bBAA; input sequence subject period treat $ carry $ y @@; if carry = '0' then carry = 'b'; cards; 1 1 2 a a 0.001086 1 1 3 b a 0.000262 1 1 4 b b 0.000581 1 3 2 a a 0.001097 1 3 3 b a 0.000214 1 3 4 b b 0.000577 1 5 2 a a 0.001126 1 5 3 b a 0.000248 1 5 4 b b 0.000638 2 2 2 b b 0.000616 2 2 3 a b 0.000911 2 2 4 a a 0.000776 2 4 2 b b 0.000691 2 4 3 a b 0.000975 2 4 4 a a 0.000907 2 6 2 b b 0.000564 2 6 3 a b 0.001054 2 6 4 a a 0.001070 run; proc glm; class sequence subject period treat carry; model y=sequence subject(sequence) period treat carry / solution ss1 ss2 e1 e2; random subject(sequence); lsmeans treat carry / pdiff stderr; output out=dres student=stdres r=resid p=fitval; run; proc univariate data=dres plot normal; var resid stdres; run; proc plot data=dres hpct=90 vpct=50; plot resid*fitval; run; Reference: DA Ratkowsky, MA Evans and JR Alldredge. 1993. Cross-over experiments : design, analysis, and application. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
