Deconstruct

Original

The original data visualisation selected for the assignment was as follows:


Source: Australian Institute of Health and Welfare (2022c).


Objective and Audience

The objective and audience of the original data visualisation chosen can be summarised as follows:

Objective

The page containing the data visualisations states that it “highlights some of the key issues in Indigenous health” (Australian Institute of Health and Welfare, 2022c).

So the practical question being asked is:

What are the key issues impacting the health of Indigenous Australians?

Burden of disease is used as a measure of “the impact of different diseases and injuries in terms of the number of years of healthy life lost due to illness or premature death” (Australian Institute of Health and Welfare, 2022c). The visualisation shows the fatal burden (years of life lost), non-fatal burden (years lived with disability) and total burden (disability adjusted life years), by disease group for Indigenous Australians (Australian Institute of Health and Welfare, 2022a).

In other words, burden of disease “is the difference between a population’s actual health and its ideal health, where ideal health is living to old age in good health (without disease or disability)” (Australian Institute of Health and Welfare, 2022a).

Audience

The data visualisation is published as a web article called Indigenous health and wellbeing on an Australian Government site. As the title contains recognisable, easily searchable terms, the audience is assumed to be diverse as anyone from the general public can look for and view the page (in contrast, the data is sourced from a report called the Australian Burden of Disease Study 2022 which uses more technical language).

The generally non-technical language of the accompanying text, brief description of burden of disease, and the high-level summary of the data show that the page is probably designed for general consumption, with links to other sources if the reader requires more detail.

The simplicity of the data visualisation supports this assumption - the use of colours and simple, summarised data.

Critique

The visualisation chosen had the following three main issues:

  • It fails to answer a practical question: The visualisation fails Kaiser Fung’s (2014) trifecta check-up question What is the question? The title - Disease group contribution to total, fatal and non-fatal burden among Indigenous Australians, 2018 - describes the data, but does not provide the audience with an understanding of what question the visualisation is trying to answer.

    The only reference to the visualisation is the conclusion that “Respiratory diseases also contributed substantially to non-fatal burden, and infant & congenital conditions to fatal burden (Figure 1)” (Australian Institute of Health and Welfare, 2022c), but without this text, an audience without previous exposure to the concept of burden of disease would be totally lost.

  • It fails to represent the data clearly: While there are some positives elements to the visualisation (Total burden, as the main data for the purpose of answering the question has been placed on the top of the visualisation; the use of different colours helps differentiate the values for each disease), overall, the visualisation is not effective as the graph choice.

    While it is easy to navigate, it is difficult to draw conclusions for a number of reasons, including:

    • The x-axis starts at 0, but this is only relevant for the first disease group on the graph, so it is impossible to tell the percentage of the second disease group onwards without relying on the hover text.
    • The reader is required to rely on the hover text to provide an understanding of what is presented, and requires having a good memory to remember what was contained in each visualisation in order to compare the contributions of the different disease groups.

  • The colour choice is not meaningful: The colours seem to have been arbitrarily assigned and are not useful except to differentiate between each disease groups.

    It is hard for the reader to easily grasp the information without having to constantly refer to the key and because the order of the disease groups is different across the three bars, it is difficult to compare them between Total, Non-fatal and Fatal.

Reconstruct

Data wrangling

The original data source used for this visualisation was Data tables: Australian Burden of Disease Study: impact and causes of illness and death in Aboriginal and Torres Strait Islander people 2018, sourced from: https://www.aihw.gov.au/reports/burden-of-disease/illness-death-indigenous-2018/data. As this is an Australian Government website, the source data is assumed to be reliable, so could be used for the reconstruction.

To reconstruct the visualisation, the following data was sourced from the above link:

  • Fatal burden figures straight from Table S4.2, column H
  • Non-fatal burden and Total burden figures from Table S2.3, columns C and D, respectively, using the formula: disease group value/total of burden.

Once the contributions were captured for the variables, the data set was converted from wide to long format, changing the original variables from Disease group, Fatal, Non-fatal and Total, to Disease group, Burden (with each burden given a letter indicator to assist with further manipulation - A = Total, B = Fatal, C = Non-fatal) and Number.

A Percentage variable was derived from the Number variable, and the data set was imported:

library(readxl)
# import the data
ATSI_burden <- read_excel("/Users/Sian/RMIT/MATH2404_A2/Burden_ATSI_2018.xlsx")

# view the first 5 rows of data to ensure it has imported correctly
head(ATSI_burden, 5)
## # A tibble: 5 × 4
##   Disease_group    Burden Number Percentage
##   <chr>            <chr>   <dbl>      <dbl>
## 1 Blood/Metabolic  A         1.6      0.016
## 2 Cancer           A         9.9      0.099
## 3 Cardiovascular   A        10.3      0.103
## 4 Endocrine        A         3.3      0.033
## 5 Gastrointestinal A         3.3      0.033

The final manipulation required prior to reconstruction was to convert the Burden values to factors so that they have meaningful names and could be ordered appropriately:

ATSI_burden$Burden <- factor(ATSI_burden$Burden,
                             levels = c("A", "B", "C"),
                             labels = c("% of Total Burden (disability-adjusted life years)", "% of Total Fatal Burden (years of life lost)", "% of Total Non-Fatal Burden (years lived with a disability)")) 

str(ATSI_burden)
## tibble [51 × 4] (S3: tbl_df/tbl/data.frame)
##  $ Disease_group: chr [1:51] "Blood/Metabolic" "Cancer" "Cardiovascular" "Endocrine" ...
##  $ Burden       : Factor w/ 3 levels "% of Total Burden (disability-adjusted life years)",..: 1 1 1 1 1 1 1 1 1 1 ...
##  $ Number       : num [1:51] 1.6 9.9 10.3 3.3 3.3 2.4 5.1 2.7 12.4 2.7 ...
##  $ Percentage   : num [1:51] 0.016 0.099 0.103 0.033 0.033 0.024 0.051 0.027 0.124 0.027 ...

Code

The following code was used to fix the issues identified in the original:

library(ggplot2)
library(magrittr)
library(dplyr)
library(viridis)
library(forcats)

burden_by_disease <- ATSI_burden %>% 
  mutate(Disease_group = fct_reorder(Disease_group, Percentage)) %>%
  ggplot(aes(x = Disease_group, y = Percentage, fill = Disease_group)) + 
  geom_bar(stat = "identity", position = "dodge") + 
  facet_grid(~ Burden) +
  labs(x= "Disease Group", y = "Percentage per Burden", title = "Key Issues Impacting the Health of Indigenous Australians (2018)", subtitle = "Disease group contribution to total, fatal and non-fatal burden*", caption = "*Disease burden measures the difference (in years) between actual health and ideal health.
  Source: AIHW (2022) Australian Burden of Disease Study: impact and causes of illness and death in Aboriginal and Torres Strait Islander people 2018
       www.aihw.gov.au/reports/burden-of-disease/illness-death-indigenous-2018/data") +
  scale_y_continuous(labels = scales::percent) +
  scale_fill_viridis(option = "H", discrete = TRUE)+
  theme_bw() + 
  theme(axis.title = element_text(size = 12, face = "bold"), legend.position = "none", plot.title = element_text(size = 16, face = "bold"), plot.subtitle = element_text(size = 14), plot.caption = element_text(size = 9, face = "italic"), plot.caption.position = "plot") + 
  geom_label(aes(label = scales::percent(ATSI_burden$Percentage, accuracy = 1, suffix = "%")), fill = "white") +
  coord_flip()

Reconstruction

The following plot fixes the main issues in the original.

References