[R] Plot binomial regression line

[Achim Zeileis]

On Thu, 26 May 2011, Jörgen Svan wrote:

> Dear all,
>
> I am quite new with R and I have a problem with plotting a binomial
> regression line in a plot.
>
> This is what I type in:
>> model<-glm(Para~Size,binomial)
>> par(mfrow=c(1,1))
>> xv<-seq(3.2,4.5,0.01)
>> yv<-predict(model,list(area=xv),type="response")
>> plot(Size,Para)
>> lines(xv,yv)
>
> The error message that I get is:
>
>> Error in xy.coords(x, y) : 'x' and 'y' lengths differ

I assume that setting "area" is not correct.

> My txt-file is attached. Could someone please help me to find out what I 
> did wrong.

For a plain scatterplot you could do this:

## read data (and code Para as factor)
d <- read.table("PerBinom.txt", header = TRUE)
d$Para <- factor(d$Para)

## model fit and predicted probabilies
m <- glm(Para ~ Size, data = d, family = binomial)
s <- seq(3.2, 4.5, by = 0.1)
p <- predict(m, data.frame(Size = s), type = "response")

## scatterplot
plot(as.numeric(Para) - 1 ~ Size, data = d)
lines(p ~ s, col = 4)

A similar display can be obtained easily with the "effects" package:

library("effects")
plot(allEffects(m), ask = FALSE, rescale = FALSE)

which also works if there is more than one regressor. See 
http://www.jstatsoft.org/v08/i15/ and http://www.jstatsoft.org/v32/i01/ 
for more details about the "effects" package.

Finally, you could also use exploratory displays, e.g., a spinogram or a a 
conditional density plot:

cdplot(Para ~ Size, data = d, ylevels = 2:1)
plot(Para ~ Size, data = d, ylevels = 2:1)
plot(Para ~ Size, data = d, ylevels = 2:1, breaks = quantile(Size))

Adding a regression line to the latter is not completely straightforward 
due to the rescaled x axis. You could do something like this:

lines(sapply(s, function(x) mean(d$Size <= x)), p, col = 4)

hth,
Z


> Thank you on beforehand,
> Jörgen
>