R Likert chart with secondary axis

Question

I want to display data that comes into pairs in a way similar to a likert chart, but with a secondary axis. I found this chart that looks exactly what I want.

Does anyone know how one can create s.th. similar or has an example on how to add a secondary axis to a Likert chart or any idea on how to present similar data in a clever way?

What lines of code did you try on what example data? I mean, there are plenty of examples available on stackoverflow on how to create charts with two axes (although many will frown upon that being bad practice). IMHO some things are quicker done in Excel; those charts beeing one of them. — lukeA, May 30 '16 at 17:15
This is a chart I found on google (no source how this was done), the closest I found so far was the Likert ones but they only have one x-axis. I will have different categories of data with two data points each, as in the chart above. — ThatQuantDude, May 30 '16 at 17:19
Speaking of Google: https://www.google.com/?q=site%3Astackoverflow.com+r+barchart+two+axes . I suggest you read through those answers + [How to provide a minimal reproducible example in R](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example#answer-5963610). Then edit and improve your question accordingly. I.e., provide input data (e.g. add the result of `dput(mydata)`), what lines of code you tried and in what way the result is not satiasfying. — lukeA, May 30 '16 at 17:26
I think you are confusing Likert scale, a scale used in questionnaires, with a horizontal bar chart that uses an overlay. One problem with the above chart is that the different scales are not comparable. It would be better to use the same scale in a [grouped bar plot](http://www.statmethods.net/graphs/bar.html#content). Using different scales is misleading - in the current example the difference of magnitude is 2x! — noumenal, May 30 '16 at 17:46

bgoldst · Accepted Answer · 2016-05-31T01:22:27.420

I enjoy trying to build these kinds of plots from primitive graphics elements and manual coordinate computations. It gives you complete control over all aspects of the plot, at the expense of a fair amount of complexity and intricacy in the plotting code.

Here you go:

## generate data
df <- data.frame(region=rep(c('OECD','Non-OECD Europe and Eurasia','China','Bunkers','Asia','Non-OECD Americas','Africa','Middle East'),each=2L),year=c(1973L,2010L),emissions=c(68,41,17,9,7,23,5,4,4,12,3,4,2,3,1,7));
df;
##                         region year emissions
## 1                         OECD 1973        68
## 2                         OECD 2010        41
## 3  Non-OECD Europe and Eurasia 1973        17
## 4  Non-OECD Europe and Eurasia 2010         9
## 5                        China 1973         7
## 6                        China 2010        23
## 7                      Bunkers 1973         5
## 8                      Bunkers 2010         4
## 9                         Asia 1973         4
## 10                        Asia 2010        12
## 11           Non-OECD Americas 1973         3
## 12           Non-OECD Americas 2010         4
## 13                      Africa 1973         2
## 14                      Africa 2010         3
## 15                 Middle East 1973         1
## 16                 Middle East 2010         7

## set configuration parameters
xlim <- c(0,1);
ylim <- c(0,1);
ybararea.pct <- 86;
ybargap.ratio <- 2.2; ## bar/gap; really describes larger bars, i.e. 2010 bars to gaps
ybar1973.pct <- 55; ## 1973 pct height of 2010
x.range.1973 <- c(0,100);
x.ticks.1973 <- seq(x.range.1973[1L],x.range.1973[2L],20);
x.range.2010 <- c(0,51.5);
x.ticks.2010 <- seq(x.range.2010[1L],x.range.2010[2L],10);
y.wrap <- 16;
col.1973 <- '#3377BB';
col.2010 <- '#BBCCEE';
col.axis <- '#888888';
legend.x <- 0.77;
legend.y <- 0.2;
legend.overlap <- 0.01;
legend.spread <- 0.033;

## precompute derived parameters
h <- (1-ybararea.pct/100)/2*diff(ylim); ybararea.range <- c(ylim[1L]+h,ylim[2L]-h);
regions <- unique(df$region);
NB <- length(regions);
ybar.height.max <- diff(ybararea.range)/(NB + (NB-1L)/ybargap.ratio); ## this formula can be derived
ybar.height.2010 <- ybar.height.max;
ybar.height.1973 <- ybar.height.2010*ybar1973.pct/100;
ybar.centers <- rev(seq(ybararea.range[1L]+ybar.height.max/2,ybararea.range[2L]-ybar.height.max/2,len=NB));

## helper function
xscale <- function(x,r,xlim=c(0,1)) xlim[1L]+(x-r[1L])/diff(r)*diff(xlim);

## plot
par(mar=c(5,7,4,2)+0.1);
plot(NA,xlim=xlim,ylim=ylim,xaxs='i',yaxs='i',axes=F,ann=F);
## bars
with(df[df$year==2010L,],rect(xscale(0,x.range.2010),ybar.centers-ybar.height.2010/2,xscale(emissions,x.range.2010),ybar.centers+ybar.height.2010/2,col=col.2010,border=NA));
with(df[df$year==1973L,],rect(xscale(0,x.range.1973),ybar.centers-ybar.height.1973/2,xscale(emissions,x.range.1973),ybar.centers+ybar.height.1973/2,col=col.1973,border=NA));
## x-axes
segments(xlim[1L],ylim[1L],xlim[2L],col=col.axis);
axis(3L,xscale(x.ticks.1973,x.range.1973),x.ticks.1973,col=col.axis,padj=0.8);
mtext(expression(paste('Percentage of total ',CO[2],' emissions for 1973')),3L,1.75);
segments(xlim[1L],ylim[2L],xlim[2L],col=col.axis);
axis(1L,xscale(x.ticks.2010,x.range.2010),x.ticks.2010,col=col.axis,padj=-0.8);
mtext(expression(paste('Percentage of total ',CO[2],' emissions for 2010')),1L,1.75);
## y-axis
mtext(lapply(regions,function(x) paste(collapse='\n',strwrap(x,y.wrap))),2L,0.3,las=2L,at=ybar.centers);
## legend
rect(legend.x-legend.overlap,legend.y-ybar.height.2010/2,legend.x+legend.spread,legend.y+ybar.height.2010/2,col=col.2010,border=NA);
rect(legend.x-legend.spread,legend.y-ybar.height.1973/2,legend.x+legend.overlap,legend.y+ybar.height.1973/2,col=col.1973,border=NA);
text(legend.x-legend.spread,legend.y,'1973',pos=2L,offset=0.1);
text(legend.x+legend.spread,legend.y,'2010',pos=4L,offset=0.1);

References

awesome thanks I will go through the code and try to fully understand it — ThatQuantDude, May 30 '16 at 19:08

R Likert chart with secondary axis

1 Answers1

References