Splus for Two-way Analysis
Setting up the Data
- fnames <- list(treatment = LETTERS[1:4], poison = c("I", "II", "III"))
- creating names for the levels of the two cacorigal variables
- poisons.design <- fac.design(c(4, 3), fnames, rep = 4)
- setting up the design matrix in the form of 4x3 table repeated 4 times
- poisons.df <- data.frame(poisons.design, surv.time)
First Look at Data
- X11()
- Opening the graphic window
- plot.design(poisons.df)
- Dot plots of means of groups, for each factor separately, on the same scale
- plot.design(poisons.df,fun=median)
- same with medians of groups
- par(mfrow=c(1,2))
- Both plots on one page!
- plot.factor(poisons.df)
- boxplots, split by each factor separately
- attach(poisons.df)
- attaches the data frame so we can use the variables directly by name
- interaction.plot(treatment, poison, surv.time)
- profile plot, treatment on the X axis.
The function aov
- aov.poisons < - aov(surv.time ~ treatment + poison, poisons.df)
- fits two-way additive model and stores in a special aov-results object
- aov.poisons < - aov(surv.time ~ treatment + poison, poisons.df)
- same plus interaction. Works also when formula is simply: surv.time ~ treatment* poison, poisons.df)< /i>
- summary(aov.blood)
- The familiar ANOVA table
- model.table(aov.poisons)
- The estimated effects
- model.table(aov.poisons, type="means")
- The estimated means
Diagnostic Plots
- hist(resid(aov.poisons))
- Histogram of residuals
- qqnorm(resid(aov.poisons))
- Normal QQ plot of residuals
- plot(fitted(aov.blood),resid(aov.blood))
- Residuals vs predicted: checking homogeniety of variance
More on formulas
- surv.time ~ (treatment + poison)^2
- same as befor
- surv.time ~ (treatment + poison + stam)^2
- for a three-way analysis with all pairwise interactions
- surv.time ~ (treatment + poison +stam)^3
- for a three-way analysis with all interactions
- surv.time ~ treatment*poison*stam
- same as above