Adding regression lines to each factor on a plot when using ANCOVA
Steve, Thanks for providing an example (which does, however need a bit of tweaking; BTW, it's usually not a good idea to cbind your data when what you really want is a data.frame). Your guess about some clever way of using abline() is unfortunately not correct - as the help page indicates, the slope and intercept must be given as single values. So you will have to extract each (intercept, slope) pair from the model coefficients and call abline() on them. A convenient way to do this is to specify the model as mod <- lm(y ~ f/x + 0) (which I first learned from MASS, the book). Here f is your grouping variable. As the book says, this gives "separate regression models of the type 1 + x within the levels of f". The "+ 0" removes the usual intercept which is replaced by individual intercepts for each level of f. For your example this will give 12 intercepts as the first 12 coefficients and 12 slopes as the remaining coefs. Then you can use cof <- coef(mod) for(i in 1:12) abline(a=cof[i], b=cof[12 + i]) to plot the 12 lines. -Peter Ehlers
On 2010-04-01 16:21, Steven Worthington wrote:
Dear R users,
i'm using a custom function to fit ancova models to a dataset. The data are
divided into 12 groups, with one dependent variable and one covariate. When
plotting the data, i'd like to add separate regression lines for each group
(so, 12 lines, each with their respective individual slopes). My 'model1'
uses the group*covariate interaction term, and so the coefficients to plot
these lines should be contained within the 'model1' object (there are 25
coefficients and it looks like I need the last 12). The problem is I can't
figure out how to extract the relevant coefficients from 'model1' and add
them using abline. I imagine there's some way of using the relevant slopes
abline(model1$coef[14:25])
together with the intercept, but I can't quite get it right. Can anyone
offer a suggestion as to how to go about this? Ideally, What i'd like is to
plot each regression line in the same color as the group to which it
belongs.
I've provided an example with dummy data below
best,
Steve
# ===========================================================
# hypothetical data
species<-
c(1,1,1,2,2,2,3,3,3,3,4,4,4,5,5,5,5,6,6,6,7,7,7,8,8,8,8,9,9,9,9,9,10,10,10,11,11,11,11,12,12,12,12,12)
beak.lgth<-
c(2.3,4.2,2.7,3.4,4.2,4.8,1.9,2.2,1.7,2.5,15,16.5,14.7,9.6,8.5,9.1,9.4,17.7,15.6,14,6.8,8.5,9.4,10.5,10.9,11.2,11.5,19,17.2,18.9,19.5,19.9,12.6,12.1,12.9,14.1,12.5,15,14.8,4.3,5.7,2.4,3.5,2.9)
mass<-
c(45.9,47.1,47.6,17.2,17.9,17.7,44.9,44.8,45.3,44.9,39,39.7,41.2,84.8,79.2,78.3,82.8,102.8,107.2,104.1,51.7,45.5,50.6,27.5,26.6,27.5,26.9,25.4,23.7,21.7,22.2,23.8,46.9,51.5,49.4,33.4,33.1,33.2,34.7,39.3,41.7,40.5,42.7,41.8)
dataset<- cbind(groups, beak.lgth, mass)
# ANCOVA function
anc<- function(variable, covariate, group){
# transform data
lgVar<- log10(variable)
lgCov<- log10(covariate)
# separate regression lines for each group
model1<- lm(lgVar ~ lgCov + group + lgCov:group)
model1.summ<- summary(model1)
model1.anv<- anova(model1)
# separate regression lines for each group, but with the same slope
model2<- lm(lgVar ~ lgCov + group)
model2.summ<- summary(model2)
model2.anv<- anova(model2)
# same regression line for all groups
model3<- lm(lgVar ~ lgCov)
model3.summ<- summary(model3)
model3.anv<- anova(model3)
compare<- anova(model1, model2, model3) # compare all models
# plots
par(mfcol=c(1,2))
boxplot(lgVar ~ group, col="darkgoldenrod1")
# plot the variate and covariate by group
plot(lgVar ~ lgCov, pch=as.numeric(group), col=as.numeric(group))
legend("topleft", inset=0, legend=as.character(unique(group)),
col=as.numeric(unique(group)),
pch=as.numeric(unique(group)), pt.cex=1.5)
abline(model1) # Need separate regression lines here
list(model_1_summary=model1.summ, model_1_ANOVA=model1.anv,
model_2_summary=model2.summ,
model_2_ANOVA=model2.anv, model_3_summary=model3.summ,
model_3_ANOVA=model3.anv, model_comparison=compare)
}
# call function
anc(beak.lgth, mass, species)
# ===========================================================
Peter Ehlers University of Calgary