[Part 3] Beyond average: How to visualise polarisation in data

Driving action from how polarisation affects different categories over time

In the 3rd case of this series, we dug into data from dentsu international’s proprietary survey about brand intelligence, engaging our clients’ stakeholders with complex data to approve a crucial shift in communication strategies.

Sometimes, when we’re trying to understand polarisation of opinions, the data available will involve different groups of people. Depending on what is the goal of the analysis, people may be segmented by age, gender, region, interests, among other categories. Besides, surveys can address different topics, such as different industries or different brands when it comes to marketing & media.

The first two parts of this study showed how visualising polarisation helps reveal insights that may be hidden when we look only at measures of central tendency like the average and the median. Besides, we’ve explored examples of how to address polarisation over time, especially with many data points. One more layer of complexity can be added to the analyses of polarisation, which is comparing across categories.

If it’s a simple comparison, among few categories and in a single period, it should be sufficient to use a similar chart to what was already seen on cases 1-A, 1-B and 2 (see Part 1 and Part 2 of this article), i.e. a stacked bar chart or a stacked column chart.

This example uses the same chart as in case 1-B (from Part 1 of this article), but with one bar for each age category instead of one bar for each year.

However, when there is a need to understand polarisation over time and across categories, the stacked bars and columns are not enough to properly handle the complexity of the data.

Case 3: Polarisation visualised in boxplots

My choice to visualise polarisation across categories and in more than one moment in time was a boxplot. This type of chart is used to show distribution. Even though it is commonly used in statistics and in finance, it’s not the simplest chart to interpret. In my years working with marketing and media data, I’ve experienced managers discourage or shy away from using boxplots and any other charts more complex than pie charts and bar charts. There’s no reason to fear boxplots, though.

The anatomy of a boxplot.

Taking one small step at a time is the key to addressing complex datasets that require more complex data visualisation to be understood. No good decisions come from stakeholders (managers or clients) that don’t really get your chart, whether they’re embarrassed or not to admit it. So the small-step-by-small-step is important both in the process of analysing the data and the process of presenting the conclusions. With a complex chart, this means taking time to understand or explain how to interpret the chart, and visually highlight each insight. This should be balanced, because the highest the rank of your stakeholder, the less time they’ll have to stop and interpret more complex charts. The payoff can be great, though.

The case in question compared a brand health indicator across multiple brands, and polarisation indicated how much more or less contrasting the brand perception became in 2 years. What we found out from the data and had to communicate to stakeholders was that our new client’s brand was underwhelming in how they connected emotionally with consumers, especially among younger generations. The goal was to support a shift in media strategies.

To capture each brand’s emotional intelligence in this proprietary study by dentsu international, survey respondents were asked questions that reflected the 5 components of the Brand EQ index:

1. Self-awareness: confidence and recognition of feelings;
2. Self-regulation: self-control, trustworthiness and adaptability;
3. Motivation: drive, initiative, commitment and optimism;
4. Empathy: understanding others’ feelings, diversity and political awareness;
5. Social skills: leadership, conflict management and communication skills.

In this case, because the data was available per brand, it was possible to analyse the dispersion of data, which is the extent to which the distribution of the data point plotted in a chart is squeezed or stretched. Because of that, the chart chosen was a boxplot. This was meant to be presented to an audience with varying levels of data savviness and visual literacy, so to guarantee that a boxplot wouldn’t seem too scary and hard to understand, it was important to go through the chart step-by-step. This approach can help unlock insights and help audiences less familiar with more complex charts engage with the data.

The comparison between the two surveys helped understand how consumer perception of brands varied in a conflicting period: the fieldwork from the first survey happened in 2020, but before the covid-19 pandemic started, and the fieldwork from the second survey happened at the end of 2021. The breakdown per age group showed how young audiences were being overlooked. However, it was important to trace the steps taken to get to that conclusion.

Polarisation involves understanding the extremes, the “poles”. The limits of the whiskers from a boxplot indicate the highest and lowest values in the dataset. And the line that divides the box in two indicates the middle value (the median).

Each dot plotted in the chart represented one brand. In a boxplot chart, half of the observations (in this case, half of the brands) will be positioned inside the box. The limits of the box represent the upper and lower quartiles. The line or “whisker” on the top and on the bottom of the box will each have 25% of the brands positioned within them.

To start looking into polarisation, the difference between the minimum and maximum values of each boxplot was highlighted, represented as a light blue area. Even though that difference decreased across age groups in the most recent survey, it was clear that young people continued to be the group with the most polarised opinions about brands.

Actually, all age groups got more polarised. The coloured areas represent the difference between the minimum and maximum values of each boxplot per age group, and all of them have a shape that shows that increase in polarisation.

In this view, the coloured areas are highlighting the interval between the maximum value and the lower quartile, which contains 75% of the brands. Because the areas show an upwards movement, it was concluded that 75% of the brands analysed most likely improved their brand perception within all age groups.

Unfortunately, the brand perception of the main brand being analysed in this case was declining in all age groups, especially among younger people. Young people spend less, feel more, and are often overlooked by brands. Our new client’s brand was struggling to communicate with all audiences, especially the younger generations, and the data made that clear.

By engaging the stakeholders with a step-by-step dive into this data, we told a story about polarisation of opinions over time across different brands. The audience was captivated and agreed to support relevant shifts in media strategy.

Conclusions

In a dataset, two categories can have the same average but quite different distribution.

While measures of central tendency like average and median are useful to summarise a whole dataset in a single value, their use as a single source of truth in business data analyses often hides valuable insights that can only be grasped when dispersion and polarisation of data are taken into account.

That doesn’t mean anyone should throw their averages out the window, but to explore survey datasets further by understanding polarisation may reveal more insights. Even when the audience doesn’t have a high level of data savviness and visual literacy, taking them through the discoveries step by step is a guaranteed way to engage people with data.