Improving Tables

1. Problem Statement

Tables and graphs are the two main approaches we have to communicate quantitative information. They both have different roles: tables are usually best suited to displaying information where we want to look up individual values and compare individual values. Precise values are usually shown. Graphs, on the other hand, are best suited to showing trends where the message is contained in the shape of the values, eg, showing patterns over time.

The merit of tables or graphs for different visualisations is covered extensively in data visualisation literature – see some suggestions on our “Resources that helped us” tab – and won’t be mentioned further here.

Much focus in data visualisation is on graphs or info-graphics. Much less discussion is centred on the humble table perhaps because they are overly familiar and we have an assumption, considered or not, that the default Word “gridline” table is “good enough”. However, as this blog post will show, well thought out and formatted tables can greatly improve the clarity and therefore impact of your message.

2. Suggested Approach and Commentary

This post assumes that we have chosen a table as our means of visualisation. Straightforward table design decisions include:

1. Delineating columns and rows; 
2. Arranging data;
3. Formatting text;
4. Summarising values;
5. Page information; and
6. Additional table characteristics

The rest of this section discusses each of these features in turn.


There is good discussion on all aspects of table design in Stephen Few’s book, Show Me the Numbers (Designing Tables and Graphs to Enlighten) – see our Resources page linked above. 

First, a standard Word or PowerPoint approach via (Insert > Table).  What is presented is a straightforward “grid” where the gridlines have been highlighted for emphasis.  In fact, the grid itself gets too much attention and detract from the information in the table. Few of us ever go much beyond amending this default view and yet there are compelling reasons to do so as the rest of this blog post will illustrate. 

This brings us to our first point.

  

2.1 Delineating columns and rows

In general, removing gridlines and replacing them with white space can dramatically improve the effectiveness of a table.  In general, the fewer gridlines the better, meaning there is less clutter for the reader to work through to understand the content in the table.  Individual data items are found more easily. 

Experiment with the amount of white space but don’t add too much.  Notice the difference between the table presentations below which experiment with white space, rules and fill color to highlight particular cell values.

Notice the difference in clarity between A) and then to D).  In A) table cells appear to be equally important.  In D) we have helped the reader focus straightaway on the most important cell(s).

In large tables with many rows, highlighting alternate rows can greatly help the reader scan across the table to pick out values of interest.  See the example below where a pale background fill has been applied to alternate rows.  Alternatively, or in addition, you can use a blank row and column after every (say) fifth row/column to help guide the reader.

Scanning along the row for Product 6, for example, is a lot easier than if no fill had been applied to the table. 

2.2 Arranging data

How do we best arrange data in a table to tell its story?  Western convention is that time-series data works best from left to right, eg columns headed Q1 2018, Q2 2018… Q4 2018 working from left to right across the table.  Equally, columns that show data values derived from another column should be placed to the right of that column to ensure a logical data flow when reading the table left to right. 

If we want to display a ranking arrangement, it usually works best to display items vertically from largest at top to smallest at bottom.   Some examples are given below:

The tables above show conventional time order left to right (Q1, Q2, Q3, Q4).  In the lower example, columns are clearly grouped as "2018 Sales by quarter", plus the units (£m) are shown only once, rather than four times, which is also helpful.

2.3 Formatting text

Western convention is to prefer horizontal orientation, unless reasons of space force us to do otherwise, eg a table with many columns might justify vertically-orientated column headers. 

Text should be aligned to the left and numbers aligned to the right.  Columns that contain simple Y/N text, for example, work well when centred in a column. 

Some examples are shown below: 

In A) the first column of text is left-aligned and the numbers in the second column are aligned to the right.  In B) the text in the second column are centred.

Fonts – Choose a font that is as legible as possible.  The same font should be used throughout a table.  Each numeric digit (0-9) should have the same width to enable numbers to align within a column.  Good choices of font are Arial (the font used in this post) or Times New Roman. 

2.4 Summarising values

Column and row summaries can be very helpful to avoid the reader mentally needing to, or attempting to, add up row or column totals themselves.  Consider making them visually distinct from the regular columns, eg through use of bold formatting.  An example is shown below:

2.5 Page information

Table title and sub-title: A clear table title helps anchor your table within the document and means its content can be unambiguously referred to at a later date.  For example, a title might be:

2019 Q3-to-date regional sales (as at 25/08/19)

If this table is referred to at a later date, the reader will have no problem understanding what the table is showing.  As a general principle, the aim should be to limit the possibility that the reader needs to ask follow-up questions. For example, in the above table, if “sales” could mean any of “booked orders”, “delivered orders” or “payments received”, the title could be improved further to clarify which measure is being reported.

2.6 Additional table characteristics

Repeat column headers at the top of each page

This is particularly valuable when a long table breaks across one or more pages of a report.  If the column header repeats at the top of each page then the user doesn’t need to refer back to the very top of the table, on the preceding page or even further back, for the column headers.  Instructions to do this in Word are as follows:

3. Applicability and Alternatives

This blog post assumes that a Table is our preferred means of visualisation – no alternatives were considered.

4. Implementation

All tables presented in this post have been created in Excel 2010 but will work in any version of Excel.

5. Context

Tables are ubiquitous in printed reports.  Few of us go beyond making simple changes to the default Word or PowerPoint default “grid view” table.  This post has shown that straightforward changes such as removing gridlines and replacing with whitespace, and thinking about row and column order and alignment can greatly improve the readability of your tables and hence the impact they make on your audience.

Climate change

To coincide with the UN Climate Change Summit 2019 on September 23rd we highlight some striking climate-change related data visualisations we have seen across the Web:

Climate Stripes

Source: Prof Ed Hawkings, University of Reading, UK

Professor Ed Hawkins' warming stripes for 1850 (left side of graphic) to 2018 (right side of graphic). Progression from blue (cooler) to red (warmer) annual readings indicates long-term increase of average global temperature.

Fossil Fuel Emissions

The progress achieved in the last 60 years has come with a massive 400% rise in CO₂ emissions. All countries can take #ClimateActionNow to reverse this trend. See the link below for a striking animation from Carlos Razo - in particular, the dramatic rise in emissions from China since the mid-2000s.

CO₂ concentrations

Source: https://e360.yale.edu/digest/co2-concentrations-hit-highest-levels-in-3-million-years

Charles Keeling was instrumental in developing the technology to accurately measure CO2 concentrations in the atmosphere, and he began doing so in 1958 at the Mauna Loa Observatory. That 61-year record has proven to be the gold standard globally for measuring the rise in anthropogenic carbon dioxide. The “Keeling Curve” is a graph that vividly depicts how atmospheric concentrations of CO2 have shot upward since the mid-20th century.

For further background, see this one-minute YouTube video: https://www.youtube.com/watch?v=rEbE5fcnFVs

Temperature increase consensus

Source: https://climate.nasa.gov/climate_resources/9/graphic-earths-temperature-record/

Data Visualisation on Twitter

Some suggested Twitter accounts to follow

There are many Twitter accounts that showcase data visualisations.  Some of the accounts we have found helpful for ideas included:

Account	Description
@theboysmithy	Alan Smith, Head of Visual and Data Journalism at the Financial Times
@storywithdata	Cole Knaflic, author of book "Storytelling with Data"
@visualisingdata	Andy Kirk, Data visualisation designer, consultant, trainer, lecturer and author
@maartenzam	Maarten Lambrechts – Data journalist | Data designer | Visualisation consultant
@simongerman600	Simon Kuestenmacher – A geographer/demographer sharing maps and data that explain how the world works
@neilrkaye	Neil Kaye – Climate data scientist at the UK Met Office
@jschwabish	Jon Schwabish – Economist, DataViz & Presentations
@OurWorldInData	Data to understand the big global problems and research that helps to make progress against them.
@NatGeoMaps	Since 1915, National Geographic Maps have been illustrating the world around us through the art and science of mapmaking

Are there other #dataviz Twitter accounts you follow that you'd like us to recommend? Please get in touch using the comments below.

Principles of Data Visualisation

Introduction

The focus of this blog site is on sharing examples of data visualisation that we hope will be useful to actuaries.  But we have also been asked about the principles underlying data visualisation. 


In preparing the examples on this site, we have found it useful to consider some questions to test whether we have achieved the desired goal of each visualisation.  These questions are available under the "Hints and Tips" tab.  Here we consider them alongside a suggested process for developing a solution. 

Data visualisation process

Step 1 - Context

Who is your audience?  What message or story are you conveying?  What data do you need to tell that story?


a) Does the information within the visualisation answer the question posed by your audience?

Step 2 - Choose your chart carefully

What is the best chart for your audience and story?


b) Have you chosen the right type of visualisation?  There is nothing wrong with a simple bar chart?

Step 3 - Test it out and experiment

Can you improve the visualisation?  Reduce clutter to make it easier for your audience to understand the story.  Does it help to add any emphasis of boxed text to tell the story?  Seek feedback from your audience.


c) Can your audience understand and interpret the visualisation quickly (eg, within 15 seconds)?  If not, then it's possible that your graph is too complicated - remove any clutter.  Does it help to label specific features or add key messages in boxed text?


d) Will the user ask a subsequent question after viewing your visualisation?  If so, then do we need some supplementary visualisation (eg, a separate graph or an overlaid line graph) or more detailed labelling of values?


e) Would your visualisation benefit from any form of data grouping?  For example, would plotting cash flows in annual buckets be clearer than monthly buckets?


f) Does the visualisation make due consideration to all your users?  Eg, a visualisation with lots of different colours may not be understood by individuals who are colour blind


g) Is the style and design of your visualisation sufficiently future-proof or may it change next time?  Users get used to seeing certain types of information.  If this visualisation changes month-on-month then the communication may be weaker.

Improving on a graphic in a news article

1. Problem Statement

We see ever more graphics around us – in newspapers, online, and social media.  These inevitably range in quality – most of them are very good, but occasionally we see examples which obfuscate rather than illuminate.  On the positive side these present an opportunity to think about how they can be improved, and as a case study, actually to show what the improvement(s) might be. 

The offending graphic in this case is the one towards the end of the following online article: https://www.theguardian.com/commentisfree/2019/jul/04/post-brexit-election-boris-johnson-polls-jeremy-corbyn

This is one such example – we take the original graphic, replicate it, then iteratively improve it.

2. Suggested Approach

Critiquing an existing graphic is essentially similar to creating a new one.  Start off by asking what the purpose of it is – what is the key message which the author is seeking to convey?  In this case the answer is clearly given above the graphic – “How Brexit would suit Boris Johnson”.  In this case there is probably no need to ask whether this is the right message – the key question is given that that is the message, how well does the graphic convey it, and how could it be better conveyed?

Here is the original graphic:

Why is it bad?

• The main issue is that the bars add up three things that are mutually exclusive - they are results from three different polls. So total Conservative support from three different polls add up to well over 60%, Labour to just over 60%, etc. This is meaningless.  The core message, about the impact of Brexit on the respective shares of the parties, requires comparison between the three polls, which is nearly impossible if they are stacked in this way.
• The colour scheme is unhelpful – the use of red and two shades of grey.  Given that red is associated already with the Labour Party, using it in a different way on a chart which includes Labour is confusing.  It might only take a second or two for the reader to figure this out, but those seconds are an unnecessary waste.

It’s also worth recognising the good aspects of the graphic, however:

• Good use of title and subtitle – it is good practice to give the key conclusion/message in the title or the subtitle of a chart – it is clearly stated here (“How Brexit would suit Boris Johnson”), along with the description of what the numbers actually represent, ie “Respondents were asked:…”.
• Good placement of the legend – knowing what the three different polls/scenarios are is core to understand this data, so placing them above the chart is helpful.
• Using a bar chart rather than a column chart – this enables the data labels (ie party names) to be shown horizontally and hence be more legible than on a column chart.

3. Rationale and Commentary

This section runs through the iterations of the chart, each one trying to improve it.  To put things in context, these iterations took about 15 minutes – ie it wasn’t a time consuming exercise.

1. Original Guardian presentation 

This is the same presentation replicated in Excel:

1. Iteration 1 - unstack the bars 

This at least enables the three different polls/scenarios to be more easily compared, as they all now start at 0% on the same axis.  However the number of bars, and the colour scheme, still get in the way of interpreting it.

1. Iteration 2 - flip rows/columns and recolour the segments 

 Stacking the bars in the other way is a fairly obvious way of reconfiguring the data, given that the results of each poll add up to 100%.  The three polls/scenarios read logically down the left hand axis.  And using the colours associated with each political party makes the presentation more intuitive. 

Returning to the core message of the graphic, the impact on the Conservative share of support in the third scenario is much clearer.

1. Iteration 3 - flip rows/columns, recolour the segments, and reorder the parties:

As a final improvement to emphasise the core message, shifting the Brexit party segments to be next to the Conservative segments shows that the increase in the latter seems to be a direct result of the decrease in the former, with the shares of the other parties staying roughly the same (which is clearer because their segments now line up).

4. Applicability and Alternatives

There may well be further iterations and alternative presentations which get the message across better, and which bring out other messages altogether – this was intentionally a “quick and dirty” exercise to show what can be done in a short period of time, so broader alternatives weren’t considered.

5. Implementation

The original graphic was replicated in Excel, which was then used to create the iterative improvements (the original spreadsheet is available on request). Excel 2016 was used for the creation of both charts, however the column/bar charts will work in any version of Excel.

6. Context

It is often quick, easy and instructive to critique graphics/visualisations which you come across in any sphere of life, and can be even more useful if you take a short amount of time to make improvements to it, if you feel that the original doesn’t convey its central message very effectively.