Icon Array

Data visualization using icon array charts to communicate validity. @AJThurston

Introduction

IN DEVELOPMENT
as of 2023-02-19

This is essentially a follow-up to the data visualization on expectancy plots. Research has demonstrated icon arrays are better understood by non-technical audiences (Zhang et al., 2018)

Setup and Libraries

As the waffle package is not available on CRAN, you will need Rtools installed to install the package from Github.

library(personograph)
library(summarytools)
library(tidyverse)
library(ggplot2)
library(ggExtra)

Data Import

In this example, these simulated data represent test scores pre and post training intervention. First, let’s import the example data and quickly orient ourselves to it. There are three variables in the file:

1. id: This is just an identifier or a unique participant ID
2. pre: this is the individual’s pre-training score
3. post: this is the individual’s post-training score

# df <- read.csv("https://raw.githubusercontent.com/AJThurston/iconarray/master/iconarray.csv")
df <- read.csv("iconarray.csv")
df %>%
  select(pre,post) %>%
  descr(.)

## Descriptive Statistics  
## df  
## N: 1000  
## 
##                        post       pre
## ----------------- --------- ---------
##              Mean      0.58      0.29
##           Std.Dev      0.15      0.16
##               Min      0.19      0.01
##                Q1      0.48      0.16
##            Median      0.58      0.27
##                Q3      0.68      0.40
##               Max      1.14      0.87
##               MAD      0.15      0.18
##               IQR      0.20      0.24
##                CV      0.25      0.55
##          Skewness      0.18      0.47
##       SE.Skewness      0.08      0.08
##          Kurtosis     -0.09     -0.30
##           N.Valid   1000.00   1000.00
##         Pct.Valid    100.00    100.00

p <- ggplot(df, aes(x = pre, y = post)) +
      geom_point() +
      theme(legend.position="none")
p <- ggMarginal(p, type="histogram")
p

Observe the fact that pre training scores are heavily positively skewed, the post-training scores are normally distributed, and the correlation between scores is about .66.

Determining Pass/Fail

Based on the number of quantiles indicated in the parameter above, now we actually need to calculate the thresholds between each of the quantiles then assign each predicted score to a quantile group. The first bit of code here calculates the quantiles with the quantile function.

df$pre_passed <- cut(df$pre, .5)
quantiles

In another approach, we may want to use a logical grouping of scores to make the decision. For example:

• Group 1: Scores from 0 to 39
• Group 2: Scores from 40 to 49
• Group 3: Scores from 50 to 59
• Group 4: Scores from 60 to 69
• Group 5: Scores from 70 to 100

It’s best to work with the client in advance to set expectations and understand what model will best meet their needs.

Icon Array Plot

In many cases it is easier to simply take these values and display them in, for example, a PowerPoint plot. However, if you have to automate this tast for multiple criteria, the code below may be useful for this purpose.

pre <- list(failed=0.73, passed=0.27)

svg("iconarray_pre.svg",
    width = 4, 
    height = 8)

personograph(pre,  
             colors=list(failed = "gray", passed="#336666"),
             draw.legend = FALSE, 
             dimensions=c(10,10))
# trace(personograph, edit = T)


dev.off()

## png 
##   2

post <- list(failed=0.42, passed=0.58)


svg("iconarray_post.svg",
    width = 4, 
    height = 8)

personograph(post,  
             colors=list(failed = "gray", passed="#336666"),
             draw.legend = FALSE, 
             dimensions=c(10,10))

dev.off()

## png 
##   2

Export

Export the pass/fail data for use elsewhere.

write.csv(data, "iconarray_appended.csv")

References

Zhang, D. C., Highhouse, S., Brooks, M. E., & Zhang, Y. (2018). Communicating the validity of structured job interviews with graphical visual aids. International Journal of Selection and Assessment, September 2017, ijsa.12220. https://doi.org/10.1111/ijsa.12220

Appendix: Alternatives

The personograph function is quite old at this point, and could not have taken advantage of the ggplot functionanility at the time it was developed. In creating this tutorial, I spent nearly two days trying to get other, more recent R packages which do leverage ggplot to work and, unfortunately, neither of the two I tried resulted in a satisfying conclusion.

Originally, this tutorial was designed to make heavy use of the waffle R package, namely the geom_waffle function: https://github.com/hrbrmstr/waffle. The ggwaffle package is another alternative: https://github.com/liamgilbey/ggwaffle. If you wanted to make waffle charts instead of icon arrays, I highly recommend both, but my personal preference is ggwaffle.

Two briefly recap my frustration with trying to use these packages for icon arrays, it really boils down to two main problems:

1. Spending a considerable amount of time diagnosing issues related to installing icons using geom_icon
2. Spending a considerable amount of computing time waiting for images to render using geom_images

Plainly, I was never successful in getting fonts to install and pull into R appropriately for geom_icon which does plot quickly. This is more driven by the issues of propriety related to font formats. Even with Font Awesome successfully downloaded and installed. With geom_images, my 24 core CPU struggled to render icon arrays in a timely manner.

Finally, if you want a no code solution to icon arrays, please consider University of Michigan’s iconarray.com. I typically steer people away from GUI-based solutions, but I modeled my R based solution off what I was able to produce there.