Mapping Convenience Stores with Waffle Grids
What waffle grid map is, and compare it with stacked bar chart
TL;DR
What is a waffle chart? What is a waffle chart on map? When should this type of map be used? This article serves as a gentle introduction to waffle chart and waffle grid map, using a dataset of convenience stores in Hong Kong. Part I explains what waffle chart is, while Part II compares the pros and cons of using waffle grid map and stacked bar chart from the lens of data visualisation and cartography.
Part I — Waffle Grid Map
Here’s a waffle grid map showing the number and proportion of two major convenience stores (CVS), 7-Eleven and Circle K, by district. It is a waffle chart on a map.
NOTE: My previous article gives a brief introduction to these two major CVS chains. Or, check Wikipedia descriptions about the operation of 7-Eleven and Circle K in Hong Kong.
But, first thing first, what is a waffle chart?
But what is a waffle chart?
In short, waffle chart is a squared version of pie chart. It shows the composition of various categorical as parts of a total. Each waffle slice includes many squares. All squares together form a grid, showing the total quantity.
Waffle charts could be easily interpreted as each square grid a quantity of something. The grids show quantities of various groups that form a whole.
A waffle chart shows progress towards a target or a completion percentage. There is a grid of small cells, of which coloured cells represent the data.
A chart can consist of one category or several categories. Multiple waffle charts can be put together to show a comparison between different charts.
from datavizproject.com
Square pie charts (a.k.a. waffle charts) can be used to communicate parts of a whole for categorical quantities. To emulate the percentage view of a pie chart, a 10x10 grid should be used with each square representing 1% of the total. Modern uses of waffle charts do not necessarily adhere to this rule and can be created with a grid of any rectangular shape.
from README of R waffle package
Example usages of waffle chart could be found here and here. In addition, the famous xkcd radiation dose chart is also a waffle chart!
One reason why waffle chart has a deep impression on readers is because of its nature of repetition — repeatedly drawing the grids of the same colour one by one. In that way, the visual impact to the reader is large that people will recover the meaning of what each grid means (alternatively speaking, they will be brainwashed). The iteration of grids makes people remember the data encoding, and in turn easier to read.
But what is a waffle grid map?
As explicit as it is, waffle grid map (or waffle map, I use these two terms interchangeably here) is literally putting waffles on the map.
When the variable you used to group each observation is spatially related (e.g. by county, by district, by any boundary, etc.), you could place each slice of waffle on the respective locations on the map, representing which area the waffle refers to. And congrats, you immediately upgraded the basic waffle chart to a waffle grid map!
The marvellous article from Kenneth Field showcases the use of waffle grid map. The map he created places waffle charts of COVID data of each country somewhere inside (or close proximity) to the boundary of that country.
Part II — Charts on map: Should it be used?
Waffle map vs. stacked bar chart
You may have seen maps with charts on top of them. Waffle map is applying the same token — putting a chart on top of each spatial variable. In most cases, the spatial variable implies countries or districts. Placing charts on a map is never a new cartography technique. Historic atlas heavily pursuit this type of data visualisation technique.
So waffle map belongs to the “ charts on map” category in the thematic map family tree. Does that mean we could replace the waffle chart on each district with other chart types? The answer is — why not?
In the end, this waffle grid map is using the same set of data for creating the stacked bar chart. Intrinsically, the stacked bars and waffle grids are derived from the number of 7-Eleven and Circle-K stores in each district.
As a quick sketch, I could replace the waffle grid with the corresponding bar in the bar chart I made in the previous article. And ta-da, we have successfully recycled the old materials to make a new thematic map.
Which method should I use?
What comes with multiple visualisation methods is the tricky question:
Which visualisation method should I choose, then?
Same as every data visualisation, there are no universal rules of choosing which method is the best. Still, listing out the pros and cons could help decision making. The table below compares where do stacked bar chart and waffle grid map plays well, and where do not¹.
Stacked Bar Chart
👍🏻 Pros: Easy to order the districts/county/unit of analysis by variable
For an array of stacked bar charts, you could order the bars by your objectives of visualising data. In most cases, the bars are ordered by the value of one variable. In the stacked bar chart I created, I order them by the proportion of 7–11 stores to all CVS in that district to emphasise the statement “Where you can find more convenience stores of one chain than the other”. That way, it is straightforward to see which district has the largest/smallest proportion of 7–11.
👍🏻 Pros: Allow direct comparison between districts
Making a stacked bar chart means every bar sums up to 100%. And using percentage means every district (i.e. unit of analysis) has the same “base” and allow direct comparison between different groups. Regardless of the total number of stores in each district, it is possible to directly compare them by the percentage of 7–11 stores in terms of all convenience stores found in the district.
Say, if we compare the distribution of stores between Shum Shui Po District (many CVS) and Islands (few CVS). The absolute numbers do not help to tell the trend (Shum Shui Po (152 stores) has around 4 times more CVS than Islands (35 stores)!). However, using percentages, we could ignore the absolute number of stores and allow direct comparison.
👎🏻 Cons: Cannot distinguish the total number of each district
What comes with percentage means that the raw total is dropped. Sometimes, the raw total of each group also tells an important message. The total number of convenience stores, in this case, could tell the differences in store density in each district.
If the total is an important point for the readers, it is necessary to add some footnotes. A method is to create an extra column, indicating the total number of stores in each district.
Waffle Grid Map
👍🏻 Pros: Easy comparison between sub-totals of districts
The difference in the waffle size could impose a powerful visual impact on readers. A large waffle grid immediately lets readers know that district has a significant amount of stores. And this visual element is much stronger than throwing a total number in a summary table.
👍🏻 Pros: Show spatial distribution of the data behind
The spatial distribution of CVS are of course not randomly distributed — stores are usually located in densely populated areas. Using maps could show the differences in the density of stores between downtown, new towns and the countryside.
👍🏻 Pros: Possible to retrieve raw data
The raw data are not completely thrown out in waffle chart — people could count the number of 7–11s and Circle-K by just counting the grids (if they are that bored).
👍🏻 Pros: Uncommon visualisation method
Let’s be honest — sometimes we get bored with bar charts. Even though showing charts are better than throwing tables with numerous large numbers, bar charts are simply not enough.
Waffles help to create a “wow factor” into the data visualisation. It helps to refresh the eyes of the readers from thousands of bar charts they have read before. Maps are eye-candy for readers. The maps add a squeeze of freshness to the story. Even if the spatial element is not the essential element of telling the story, adding the counties and locations behind adds a few drops of fun.
👎🏻 Cons: Longer preparation time
It takes time to make a map that could effectively convey the message. When not designed properly, the message behind is hindered by the superfluous chartjunk. What level of spatial detailedness should be added to the map to effectively blend the non-spatial and spatial parts of the data, and then tell the story?
Placing the waffle charts in a suitable area is somehow a non-trivial task. And do not forget the time required for adding those annotations. Again, it takes time to prepare an effective map².
👎🏻 Cons: Are maps always needed?
Is the geographic element important in the data? Does the map lying behind helps convey the message more clearly? Map could be a double-edged sword — it could tell you more behind the story of the data if the trend is spatially related. It could also veil the story and hinder you from discovering the trend (by making wrong perceptions) if the trend is not spatially related.
Think twice, and more than twice before mapping.
Hiatus
This article compares waffle charts and stacked bar charts. From the lens of data visualisation and cartography, you could notice where does each data visualisation method outplays, and where it does not play well. Choosing a data visualisation method is, in general, not a black-or-white decision. There are no panacea and a single chart that could visualise all the stories behind the data. At the same time, it is difficult to quantify which method is absolutely better than the other.
Never forget that we are creating charts and maps with a purpose. Waffle grid maps is a great tool and could create nice maps, but it should be used with valid reasons.
Further Readings
Waffle Chart — Data Viz Project
https://datavizproject.com/data-type/percentage-grid/
Mapping coronavirus waffles
https://www.esri.com/arcgis-blog/products/arcgis-pro/mapping/mapping-coronavirus-waffles/
¹ We could have four chart types (stacked bar chart, waffle chart, stacked bar chart on map, waffle grid on map) if we perceive map as one of the geometry layers (with longitude and latitude being x and y aesthetics respectively) using the grammar of graphics approach. But this may make the whole story too complicated. Here, I compare only two types of “charts” to keep things simpler.
² I will write a behind-the-story to jot down my thoughts and experiments on creating this map. Stay tuned!
