ASSIGNMENT 2

CRITIQUE OF VISUALISATIONS

Based on 2 visualisation validation levels: Data / Task Abstraction → Visualisation Idiom

Example 1 - Dengue Trends Example 2 - World Fertility Rate

Example 1 - Dengue Trends in 2020

This visualisation was published by the National Environment Agency (NEA) in Singapore on 3 June 2020, as part of a press release to warn Singapore about the spike in dengue cases even as the COVID-19 pandemic was raging, and to urge collective efforts to prevent dengue. Broadly, the chart was intended to complement NEA's message that the trend of dengue cases from January to April 2020 was abnormally high in Singapore's history.

Weekly Dengue Trends

Source: NEA (3 June 2020)

Data (What they are showing)

Chart more suitable for data exploration than communicating multiple messages

The visualisation creator was trying to convey NEA's messages that in 2020, the annual trend of (a) weekly and (b) total dengue cases was abnormally higher than trend of dengue cases in specific years by plotting quantitative data - number of dengue cases every week - over 52 weeks in a year, from 2013 to 2020.

The chart shows only the weekly number of dengue cases. However, NEA's press release described the dengue trends both in terms of "weeks" and "months". This means the reader has to mentally calculate which month each week in the chart belongs to, while referring to the text. The creator could have aggregated the weekly number of dengue cases by month to reduce this cognitive effort.

Showing many years in the chart also makes it difficult for the viewer to understand which are the focal data in NEA's message.

Not showing the total number of dengue cases from 2013 to 2019 and projected total dengue cases in 2020 makes it harder for the reader to believe that the trend of high dengue cases from April to May 2020 would continue, given that it coincided with the 2020 Circuit Breaker period, where more people stayed at home and were exposed to mosquitoes in their neighbourhood.

Idiom (How they are showing it)

Chart junk

There are too many colours used, and it is not intuitive what colour is related to which year.

The colours of the 2014, 2017 and 2019 lines are also very similar, making it more challenging for users with some colour blindness to see.

The reader has to exert extra cognitive effort to link each year and colour on the legend, to the colour of the line on the chart, before they can see the trend of the target year.

Hard to focus

The lines are all of the same thickness and there is no salient colour that the reader can focus on. It is difficult for the reader to know which year(s) they are supposed to look at, or compare.

Small font

The original visualisation on NEA's webpage is of poor resolution and the text on the figure is small, making it difficult to see each data point clearly.

Given that the intended messages are of high public health importance, the data should be conveyed more efficiently with a higher font size and resolution.

Suggestions to Improve

1. 2 Charts to Show Expected Trend

Since this chart is intended for communication to the public, and not data exploratory analysis by NEA, perhaps the creator could have split this chart into 2 line charts with different data to convey different meanings. Specifically, NEA mentioned in its press release that:

  • "Singapore has not seen such a high weekly number of dengue cases since the peak years in 2013 and 2014"

    • Suggestion: Using a line chart, instead of showing every year of data, the creator can use interactivity to show data by year, or focus on the average or historical highest number of weekly dengue cases from 2013 to 2019 (black line), and compared this trend with 2020's historical number of weekly dengue cases (red line). This chart would have communicated more clearly that 2020 has an abnormal trend of dengue cases compared to previous years.

    The weekly case data can be aggregated into monthly data to be better aligned with NEA's messages and avoid confusing the user.

    If projected data is available, especially because of strong correlation between dengue and temperature, NEA could possibly illustrate that the trend of dengue cases from January to June 2020 was expected to continue in the rest of 2020 if action was not taken.

    I illustrate my suggestions in the following chart:
    Weekly Dengue Trends

  • "The number of dengue cases this year is expected to exceed the 15,998 cases reported in 2019, and may even surpass the 22,170 cases reported in 2013"

    Suggestion: Using a column chart, the creator could have focused on the total historical number of dengue cases from 2013 to 2020 (1 bar per year), and show the historical (solid line) and projected (stacked on top of the 2020 solid bar) total number of cases in 2020.

    I illustrate my suggestions in the following chart:
    Weekly Dengue Trends

Suggestions to Improve

2. Polar Plot to Show Abnormal Weekly Trends

Instead of a line chart, NEA could also choose to use a polar plot to show the weekly trends in dengue across months and years more clearly. The creator could bold the line for 2020 to make it more salient to the user. An example of a polar plot showing data with monthly seasonality across years is shown on the right.

Today's Chart
Image Source

Back to Top

Example 2 - World's Fertility Rate 2023

This visualisation is published by the user maps_us_eu on the Data is Beautiful subreddit that is dedicated to "aesthetically pleasing works of data visualisation".

It is intended to express the fertility rate per country and showcase countries with either growing, stable or declining population, with 2023 data.

World Fertility Rate

Source: maps_us_eu (19 February 2023)

Population data
Fertility data from UN?

Data (What they are showing)

Misleading data source

The creator cited that the population size and fertility rate data were obtained in 2023, with the latter from "2023 United Nations Fertility Rate".

This is misleading, as the author posted this on 19 February 2023. To our best research, statistics for world population size and fertility rate in 2023 by the United Nations were unavailable at this time (26 February 2023), and were at best from 2019 or projected. If they meant that the data was "projected" for 2023, they should have stated so or list the actual data sources.

Misleading linking of 2 datasets

It was misleading for the creator to directly link the concept of fertility rate with current population size, if the intent was to show population growth or decline.

First, a country's population growth or decline is not attributed only to fertility rate, but also the concepts of mortality rate (which is further dependent on life expectancy), immigration and emigration.

Second, fertility rate is not based on the actual number of live births in a given year (which birth rate might indicate), but the average number of children of an imaginary woman who lives till the end of her entire childbearing years (age 15 - 49) is expected to have, in line with the age-specific fertility rate of the given year (Source: World Bank). This has little bearing on the given country's population size in the given year, although it provides informaton about the population's expected growth rate.

Inappropriate data binning and legend

The dataset for fertility rate is expressed as "number of kids per female", which is on a continuous, ratio scale. However, the creator chose a discrete 6-colour scale to describe the fertility rate of a given country, expressed in [>1, >1.5, >2, >3, >4, >5] kids per female.

This is confusing for the reader as the categories are:

  • Not mutually exclusive - there is no upper cap per category according to the discrete colour scale used. After scrutinising the visualisation, it is likely that the creator meant that a country labelled with a fertility rate "more than 1.5 children per female" actually has a fertility rate of "more than 1.5, but less than 2 children" per female.

  • Not exhaustive - What colour would a country with a fertility rate of "1.0 or less" children per female would be encoded in? Based on the colour legend, such countries should not be on the chart. South Korea's fertility rate was one of the world's lowest at 0.78 in 2022. However, South Korea's polygon is coloured purple in the visualisation, although purple is supposed to symobolise "more than 1 kids per female". There is either an error with the labelling, or the data was inaccurate.

Fertility Rate Legend
fertility map

Idiom (How they are showing it)

Redundant channels

The visualisation has multiple channels, and yet only encodes 2 data scales:

  • Total population size per country (treated as ratio data), encoded as an area magnitude channel (2D size of polygons)

  • Fertility rate per country (treated as ordinal data), encoded as discrete colour scale

According to Tamara Munzner, a visualisation is considered to be "expressive" only if the visualisation's encodings express all of, and only, relevant information of the dataset. There are a few redundant channels that are meaningless:

  • Grouping of countries - The grouping of the polygons does not appear to follow any logical grouping. For example, the European Union countries are all grouped into a single polygon as if it is a single country (bottom of the circle), yet countries from the European continent or other international political or economic unions are not grouped.

  • Position of shapes - The creator tried to fit all countries into the shape of a globe. However, the positions of the countries on the visualisation do not appear to be aligned with their actual positions on a world globe. For example, the United States appears directly below India.

  • Patterns on shapes - Some polygons have scale patterns on them (for example, India , China, European Union), while others do not. There is no explanation to understand how the countries with scale patterns are related to each other.

Comparability is difficult

It is hard to compare the size of similarly sized polygons as there is no consistent shape. There is a wide range of shapes - some polygons have 4 sizes (e.g., United States) while others have 13 (e.g., Brazil). One has to rely on the figures to tell the difference.

Suggestions to improve

Encoding of fertility rate data

1. Bin the fertility rate into more useful, mutually exclusive and exhaustive categories:
If we accept that the replacement fertility rate needed to maintain a country's population size is 2.1 children per female, and any country with a fertility rate below this is expected to enter population decline over time, the creator could consider using a 3-point discrete colour scale (3 distinctly contrasting colours):

  1. Population growing: >=2.1 children per female

  2. Stable population: 2.1 children per female

  3. Population in decline: 2.1 children per female

2. Use a continous colour scale for fertility rate:
Instead of a discrete colour scale, the creator could use a more ordinal/sequential colour scale to express ascending fertility rates across countries e.g., in terms of ascending colour saturation.

3. Consider a different visualisation to present population change rates across countries
The creator could have subtracted the crude death rate from the crude birth rate to obtain the population change rate in the absence of migrations, to express which countries are in danger of population decline. As the obtained population change rate would be a continuous data type, the creator could consider a choropleth map to show population change rate per country in terms of colour saturation.

Remove redundant channels

The creator can consider removing the redundant channels from the current visualisation, in accordance with the expressiveness principle as earlier mentioned.

The visualisation should only have a single channel encoding each dataset:

  • Total population size per country (treated as continuous data), encoded via a single channel such as area magnitude or position. If a globe shape is still desired by the creator, perhaps the creator could try to keep the shape of each polygon representing each country within the globe consistent (e.g., circle), while using purely an area magnitude channel to encode population data.

  • Fertility rate per country, which the author can encode as a discrete or continuous colour scale (see "Encoding of fertility rate data" on left). For example, higher fertility rate can be expressed as higher colour saturation.

There should be no patterns on some polygons, unless the patterns are encoding another dataset and represent a relationship between polygons with patterns.

Back to Top