geom_smooth中的自定义lm公式

Question

I'm working with faceted plots, and adding lines using the lm method in geom_smooth()

d<-data.frame(n=c(100, 80, 60, 55, 50, 102, 78, 61, 42, 18),
              year=rep(2000:2004, 2), 
              cat=rep(c("a", "b"), each=5))

ggplot(d, aes(year, n, group=cat))+geom_line()+geom_point()+
  facet_wrap(~cat, ncol=1)+
  geom_smooth(method="lm")

I would like to set up a function to apply a polynomial where appropriate. I've worked up a function:

lm.mod<-function(df){
  m1<-lm(n~year, data=df)
  m2<-lm(n~year+I(year^2), data=df)
  ifelse(AIC(m1)<AIC(m2), "y~x", "y~poly(x, 2)")
}

But I'm having trouble applying it. Any ideas, or better ways to approach this?

Answer 1

It's not possible to apply different smooth functions with a single geom_smooth call. Here is a solution which is based on smoothing subsets of data:

First, create the base plot without geom_smooth :

library(ggplot2)
p <- ggplot(d, aes(year, n, group = cat)) +
       geom_line() +
       geom_point() +
       facet_wrap( ~ cat, ncol = 1)

Second, the function by is used to create a geom_smooth for each level of cat (the variable used for facetting). This function returns a list.

p_smooth <- by(d, d$cat, 
               function(x) geom_smooth(data=x, method = lm, formula = lm.mod(x)))

Now, you can add the list of geom_smooth s to your base plot:

p + p_smooth

The plot includes a second-order polynomial for the upper panel and a linear smooth for the lower panel:

Answer 2

lm.mod<-function(df){
  m1<-lm(n~year, data=df)
  m2<-lm(n~year+I(year^2), data=df)
  p <- ifelse(AIC(m1)<AIC(m2), "y~x", "y~poly(x, 2)")
return(p) 
}
# I only made the return here explicit out of personal preference

ggplot(d, aes(year, n, group=cat)) + geom_line() + geom_point() +
  facet_wrap(~cat, ncol=1)+
  stat_smooth(method=lm, formula=lm.mod(d))
# stat_smooth and move of your function to formula=

# test by reversing the condition and you should get a polynomial.
# lm.mod<-function(df){
#   m1<-lm(n~year, data=df)
#   m2<-lm(n~year+I(year^2), data=df)
#   p <- ifelse(AIC(m1)>AIC(m2), "y~x", "y~poly(x, 2)")
# return(p)
# }