I am working with a RCB 2x2x3 ANCOVA, and I have noticed a difference in the calculation of sum of squares in a Type III calculation.
Anova output is a follows: > Anova(aov(MSOIL~Forest+Burn*Thin*Moisture+ROCK,data=env3l),type=3) Anova Table (Type III tests) Response: MSOIL Sum Sq Df F value Pr(>F) (Intercept) 22.3682 1 53.2141 3.499e-07 *** Forest 1.0954 2 1.3029 0.29282 Burn 2.6926 1 6.4058 0.01943 * Thin 0.0494 1 0.1176 0.73503 Moisture 1.2597 2 1.4984 0.24644 ROCK 2.1908 1 5.2119 0.03296 * Burn:Thin 0.2002 1 0.4764 0.49763 Burn:Moisture 1.0612 2 1.2623 0.30360 Thin:Moisture 1.6590 2 1.9734 0.16392 Burn:Thin:Moisture 1.1175 2 1.3292 0.28605 Residuals 8.8272 21 However, I would like to calculate some a priori contrasts within the Moisture factor as follows: Transect_moisture_contrasts<-matrix(c(-1,2,-1,1,0,-1),3,2) dimnames(Transect_moisture_contrasts)<-list(levels(env$Moisture),c("I vs. X&M","X vs. M")) contrasts(env$Moisture)<-Transect_moisture_contrasts > contrasts(env3l$Moisture) I vs. X&M X vs. M X -1 1 I 2 0 M -1 -1 soilmodel<-lm(MSOIL~Forest+Burn*Thin*Moisture+ROCK,data=env3l) > linearHypothesis(soilmodel,"MoistureI vs. X&M") Linear hypothesis test Hypothesis: MoistureI vs. X&M = 0 Model 1: restricted model Model 2: MSOIL ~ Forest + Burn * Thin * Moisture + ROCK Res.Df RSS Df Sum of Sq F Pr(>F) 1 22 9.4106 2 21 8.8272 1 0.58333 1.3877 0.252 > linearHypothesis(soilmodel,"MoistureX vs. M") Linear hypothesis test Hypothesis: MoistureX vs. M = 0 Model 1: restricted model Model 2: MSOIL ~ Forest + Burn * Thin * Moisture + ROCK Res.Df RSS Df Sum of Sq F Pr(>F) 1 22 9.6359 2 21 8.8272 1 0.80871 1.9239 0.18 The sum of squares for these two contrasts do not add up to the sum of squares of the main effect Moisture > .80871+.58333 [1] 1.39204 > 1.39204-1.2596 [1] 0.13244 Checking them together produces the correct sum of squares for the main effect > linearHypothesis(soilmodel,c("MoistureI vs. X&M","MoistureX vs. M")) Linear hypothesis test Hypothesis: MoistureI vs. X&M = 0 MoistureX vs. M = 0 Model 1: restricted model Model 2: MSOIL ~ Forest + Burn * Thin * Moisture + ROCK Res.Df RSS Df Sum of Sq F Pr(>F) 1 23 10.0869 2 21 8.8272 2 1.2596 1.4984 0.2464 So my question is: Should the sum of squares for the two contrasts add to the main effect here? If they should, maybe we can figure out why mine do not. Thanks in advance for any assistance. Cheers, John John J. Wiley, Jr. PhD Candidate State University of New York College of Environmental Science and Forestry Department of Environmental and Forest Biology 460 Illick Hall Syracuse, NY 13210 315.470.4825 (office) 740.590.6121 (cell) [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.