Solved – How to predict or extend regression lines in ggplot2

ggplot2prediction

I have a data frame that contains two time series: the dates and version numbers of Emacs and Firefox releases. Using one ggplot2 command it's easy to make a chart that uses loess (in a way that looks a bit amusing, which I don't mind) to turn the points into lines.

How can I extend the lines into the future? I want to determine where and when Emacs and Firefox version numbers will cross, and if there's a way to show an error range, all the better.

Given that ggplot2 is plotting the lines, it must have a model, but I don't see how to tell it to extend the lines, or to get the model out and do something with it.

> library(ggplot2)
> programs <- read.csv("http://www.miskatonic.org/files/se-program-versions.csv")
> programs$Date <- as.Date(programs$Date, format="%B %d, %Y")
> head(programs)
  Program Version       Date
1   Emacs    24.1 2012-06-10
2   Emacs    23.4 2012-01-29
3   Emacs    23.3 2011-03-10
4   Emacs    23.2 2010-05-08
5   Emacs    23.1 2009-07-29
6   Emacs    22.3 2008-09-05
> head(subset(programs, Program == "Firefox"))
   Program Version       Date
18 Firefox      16 2012-10-09
19 Firefox      15 2012-08-28
20 Firefox      14 2012-06-26
21 Firefox      13 2012-06-15
22 Firefox      12 2012-04-24
23 Firefox      11 2012-03-13
> ggplot(programs, aes(y = Version, x = Date, colour = Program)) + geom_point() + geom_smooth(span = 0.5, fill = NA)

Emacs and Firefox versions by date

(Note: I had to fudge the early Firefox versions and turn 0.1 onto 0.01, etc., because "dot one" and "dot ten" are equal arithmetically. I know Firefox is releasing every six weeks now, but they don't exist yet, and I'm interested in a general answer to this prediction question.)

Best Answer

As @Glen mentions you have to use a stat_smooth method which supports extrapolations, which loess does not. lm does however. What you need to do is use the fullrange parameter of stat_smooth and expand the x-axis to include the range you want to predict over. I don't have your data, but here's an example using the mtcars dataset:

ggplot(mtcars,aes(x=disp,y=hp)) + geom_point() + xlim(0,700) +
stat_smooth(method="lm",fullrange=TRUE)

Related Solutions

Solved – How to add horizontal lines to ggplot2 boxplot

Found solution myself. Maybe someone could use it:

#step 1: preparing data
ageMetaData <- ddply(data,~group,summarise,
            mean=mean(age),
            sd=sd(age),
            min=min(age),
            max=max(age),
            median=median(age),
            Q1=summary(age)['1st Qu.'],
            Q3=summary(age)['3rd Qu.']
            )
#step 2: correction for outliers
out <- data.frame() #initialising storage for outliers
for(group in 1:length((levels(factor(data$group))))){
	bps <- boxplot.stats(data$age[data$group == group],coef=1.5) 
	ageMetaData[ageMetaData$group == group,]$min <- bps$stats[1] #lower wisker
	ageMetaData[ageMetaData$group == group,]$max <- bps$stats[5] #upper wisker
	if(length(bps$out) > 0){ #adding outliers
		for(y in 1:length(bps$out)){
			pt <-data.frame(x=group,y=bps$out[y]) 
            out<-rbind(out,pt) 
        }
    }
}
#step 3: drawing
p <- ggplot(ageMetaData, aes(x = group,y=mean)) 
p <- p + geom_errorbar(aes(ymin=min,ymax=max),linetype = 1,width = 0.5) #main range
p <- p + geom_crossbar(aes(y=median,ymin=Q1,ymax=Q3),linetype = 1,fill='white') #box
# drawning outliers if any
if(length(out) >0) p <- p + geom_point(data=out,aes(x=x,y=y),shape=4) 
p <- p + scale_x_discrete(name= "Group")
p <- p + scale_y_continuous(name= "Age")
p

The quantile data resulution is ugly, but works. Maybe there is another way. The result looks like this:

ggplot2 boxplot with limit lines

Also improved boxplot a little:

added second smaller dotted errorbar to reflect sd range.
added point to reflect mean
removed background

maybe this also could be useful to someone:

p <- ggplot(ageMetaData, aes(x = group,y=mean)) 
p <- p + geom_errorbar(aes(ymin=min,ymax=max),linetype = 1,width = 0.5) #main range
p <- p + geom_crossbar(aes(y=median,ymin=Q1,ymax=Q3),linetype = 1,fill='white') #box
p <- p + geom_errorbar(aes(ymin=mean-sd,ymax=mean+sd),linetype = 3,width = 0.25) #sd range
p <- p + geom_point() # mean
# drawning outliers if any
if(length(out) >0) p <- p + geom_point(data=out,aes(x=x,y=y),shape=4) 
p <- p + scale_x_discrete(name= "Group")
p <- p + scale_y_continuous(name= "Age")
p + opts(panel.background = theme_rect(fill = "white",colour = NA))

The result is:

advanced boxplot with ggplot2

and the same data with smaller range (boxplot coef = 0.5)

boxplot with outliers

Solved – How to ‘dodge’ the position of geom_point in ggplot2

The group should = INDEX instead of ntrunc in the aes.

plot =  ggplot(data, aes(x=ntrunc, y=beta_best, group=INDEX, colour=INDEX)) +
   geom_point(aes(shape=detectable), na.rm=TRUE, position="dodge") +
   geom_errorbar(aes(x=ntrunc, ymax=beta_high, ymin=beta_low), na.rm=TRUE, position="dodge")

The plot looks better now.

enter image description here

Best Answer

Related Solutions

Solved – How to add horizontal lines to ggplot2 boxplot

Solved – How to ‘dodge’ the position of geom_point in ggplot2

Related Question