Skip to content

Plotting confidence intervals

5 messages · Ben Tupper, phii m@iii@g oii phiiipsmith@c@, Jim Lemon

phii m@iii@g oii phiiipsmith@c@
# 
I want to show little bell curves on my bar chart to illustrate the 
confidence ranges. The following example from Paul Teetor's "R Cookbook" 
does what I want, but shows I-beams instead of bell curves. The I-beams 
suggest uniform, rather than normal distributions. So I am looking for a 
way to plot normal distribution curves instead.

# Example from Paul Teetor, "R Cookbook", page 238.
library(gplots)
attach(airquality)
heights <- tapply(Temp,Month,mean)
lower <- tapply(Temp,Month,function(v) t.test(v)$conf.int[1])
upper <- tapply(Temp,Month,function(v) t.test(v)$conf.int[2])
barplot2(heights,plot.ci=TRUE,ci.l=lower,ci.u=upper,
          ylim=c(50,90),xpd=FALSE,
          main="Mean Temp. By Month",
          names.arg=c("May","Jun","Jul","Aug","Sep"),
          ylab="Temp (deg. F)")

Does anyone know a package that does this or, alternatively, can anyone 
suggest a direction to go in if one were to write R code to do this?

Philip
Ben Tupper
# 
Hi,

Would something like yarrr do the trick?
https://ndphillips.github.io/yarrr.html

Or gghalves?  https://github.com/erocoar/gghalves


Cheers,
Ben
On Sat, Dec 7, 2019 at 9:32 PM <phil at philipsmith.ca> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      phii m@iii@g oii phiiipsmith@c@
    

    

      
      #
    

  


  

      
Thanks for these helpful suggestions.

These options don't work in my case because I don't know the individual 
observations (the dots). A statistical agency collects the observations 
and keeps them confidential. It provides the mean value and the standard 
deviation, plus the fact that the observations are normally distributed. 
So I have enough information to draw the distribution function.  Mean 
values and standard deviations are provided for several cases 
(geographies). I can plot the mean values for all cases in a bar chart. 
I can show the confidence intervals as I-beams, as in my example. But I 
would prefer to show the confidence intervals as truncated bell curves, 
referring to, say, 95% of the unseen observations.

Philip

        
On 2019-12-07 22:07, Ben Tupper wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Jim Lemon
    

    

      
      #
    

  


  

      
Hi Philip,
This may be a starter:

attach(airquality)
heights <- tapply(Temp,Month,mean)
temp_sd<-tapply(Temp,Month,sd)
lower <- tapply(Temp,Month,function(v) t.test(v)$conf.int[1])
upper <- tapply(Temp,Month,function(v) t.test(v)$conf.int[2])
library(plotrix)
barp(heights,ylim=c(0,100),names.arg=month.abb[5:9],
 main="Air quality (May-Sep)",xlab="Month",ylab="Temperature")
dispersion(1:5,y=heights,ulim=upper,llim=lower,intervals=FALSE)
ci95<-seq(-1.96,1.96,length.out=40)
norm_curve<-rescale(dnorm(ci95),c(0,0.4))
for(i in 1:5)
 polygon(c(i-norm_curve,i+norm_curve),
  c(heights[i]+ci95*temp_sd[i],heights[i]+rev(ci95*temp_sd[i])))

Jim

        
On Sun, Dec 8, 2019 at 8:18 PM <phil at philipsmith.ca> wrote:

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      phii m@iii@g oii phiiipsmith@c@
    

    

      
      #
    

  


  

      
Thanks so much Jim. Yes, this is giving me what I want.

Philip

        
On 2019-12-08 05:00, Jim Lemon wrote: