[Part 1] Beyond average: How to visualise polarisation in data

Making peace with polarisation and its value to explain divergent opinions

The first case of this 3-part study tackled a challenge of visualising polarisation in data that quantified the impact of Diversity & Inclusion efforts in a company, achieving the goal to ensure continued support from stakeholders.

Sometimes it takes getting in touch with my contentious, former-lawyer side, to admit that diverging opinions are normal in healthy debates. Before switching careers to work with data, in my formative years in Law school, I was especially attracted to mediation, conciliation and arbitration — methods that are alternatives to a court of justice, where a neutral facilitator helps the parties to resolve disputes and make agreements. Studying those methods includes recognizing how extremely polarised opinions often cause an unsettling feeling that there’s no way to find middle ground. On a political level, severe polarisation may damage institutions essential to democracy, and that perception was one of the reasons why I decided to leave my home country. Learning how to navigate a world where perspectives about important topics are often highly polarised involves comprehending the data that quantifies that polarisation.

More recently, my work with data analyses has included analysing survey data, which also requires understanding contrasting groups of opinions or beliefs. Even if you never analysed survey data, there’s a good chance you have at some point answered a survey with answer options that look like a scale of agreement.

Example of agreement scale. There are sometimes variations on how the extremities are named or if the middle options are named.

The goal of this kind of question is usually answering if, on average, a group of people agree or disagree with a certain statement. It is highly important to analyse the central tendency of opinions by using measures like the average and the median. However, we may be missing relevant insights about human behaviour by not looking into polarisation. A step further from just calculating the average agreement with each statement could be to explore how polarised they are, i.e. how big is the gap between the extremes. Polarisation goes beyond the central tendency to show how neutral opinions are about a topic, and just how different opinions can be.

Data from the proprietary study of Brand EQ from dentsu international.

Data visualisation plays a big part in making data tangible. And what makes me love it more: data visualisation is an intrinsic part of analytical thinking, especially when trying to understand large volumes of data. How can polarisation be better represented visually? This should aim to help unlock more insights when analysing the data, and to better communicate the data behaviour to an audience. Choosing the best chart to show polarisation depends on the data available. This study explores some possibilities of chart types to represent polarisation. The examples use marketing data, from surveys about customer attitudes and brand perception, but the concepts discussed can be applied to analyse other types of data as well.

Case 1-A: Static polarisation visualised in stacked bars

One of the simplest ways to visualise polarisation is using stacked bars. This case originally used data from a Diversity & Inclusion survey realised by the Benelux team from dentsu international, where respondents were asked about gender identity, sexual orientation, cultural background, spirituality, physical and mental health, discrimination experiences, among other topics. Here are some examples of the survey section that explored the feeling of belonging:

To what extent…

1. Do you feel welcome at our company?
2. Do you feel safe at our company?
3. Do you feel you can be yourself within the company?
4. Do you feel your ideas, opinions and feedback are heard within the company?

However, the original data is confidential and protected by the General Data Protection Regulation (GDPR). Because of that, the data has been slightly altered and the questions have been changed into a context of colours, so that the charts could still be used as examples while still respecting the respondents’ privacy.

Example inspired by the presented results from a Diversity & Inclusion survey. The data has been altered to respect privacy.

The choice of horizontal bars intended to mimic the sequence of the agreement scale in the survey, making it easier to interpret the data. A divergent colour scale from red to green was used to represent the diverging opinions. With this, we were able to analyse to what extent people felt welcome and safe at the agency, or in the privacy-friendly example, to what extent people liked the colours blue and orange. Opinions tended to the highest scores: 4 or 5, which meant feeling welcome and safe, or liking blue and orange.

In this case, opinions weren’t much polarised, because the negative (1 or 2, red) and neutral options (3 — white) are just a small portion of each bar. This helped the team involved in Diversity & Inclusion understand what kind of initiatives were most needed to make a true impact in the company.

When the same Diversity & Inclusion survey was repeated a year after the first results were analysed, it posed the challenge of understanding how opinions changed over time. This will be addressed as Case 1-B on Part 2 of this article, which will also bring other examples of polarisation over time and how to visualise it.

Conclusions

Measures of central tendency like the average and the median can hide important information about human behaviour, and visualising polarisation helps reveal those insights. There are multiple types of charts that can help visualise polarisation, and the simplest way to start is with stacked bars.

By adding more levels of complexity to the analyses of polarisation, such as the passage of time, other chart types may be more appropriate — this will be discussed further in Part 2.