Below is a brief snippet of R syntax. If you run it, you will see that model 2 v model 3 produces the same F ratio in each output. However, model 1 vs model 2 produces different F ratios. My understanding is that these should be the same. Any insights as to what is happening would be appreciated.
prob1 <- scan() 99.9 81.2 77.8 66.4 99.9 80.4 81.6 80.6 42.2 50.9 46.9 83.0 91.2 85.8 82.0 74.1 47.4 47.9 40.5 60.2 70.1 88.2 57.8 61.7 54.3 45.8 52.1 88.0 71.9 57.6 59.2 99.9 60.8 60.9 55.5 99.9 96.0 71.2 75.3 99.9 80.0 77.2 64.9 99.9 92.3 80.7 73.3 99.9 52.7 57.5 56.8 48.6 prob1 <- data.frame(t(array(prob1,dim=c(4,13)))) names(prob1) <- c("pchem1","mvcalc","pchem2","molbio") lm.1 <- lm(pchem2 ~ pchem1, data=prob1) lm.2 <- lm(pchem2 ~ pchem1 + mvcalc, data=prob1) lm.3 <- lm(pchem2 ~ pchem1 + mvcalc + molbio, data=prob1) anova(lm.1,lm.2) anova(lm.2,lm.3) anova(lm.1,lm.2,lm.3)
FOLLOW-UP #1:
The first anova() call is doing this calculation: $\frac{R_2^2-R_1^2}{1-R_2^2}·\frac{n-3}{1}$
The second anova() call is doing this calculation: $\frac{R_2^2-R_1^2}{1-R_3^2}·\frac{n-4}{1}$
If anyone can provide a reference for this, it would be appreciated. Additionally, if anyone could explain why the step-by-step single paired answer might be incorrect would be welcome. (I have taught it this way, and all the texts I've referenced indicate doing it this way.)
FOLLOW-UP #2:
SPSS (default) reports model-comparisons pair-wise...not in reference to the last model in the list.