Note: Originally, we wrote this article in french following an invitation from Survey Magazine. It was first published in this quarterly magazine dedicated to the world of surveys and market research in November 2019 before being presented here in english on our website. 

When thinking of DataViz or data visualization, the first thing that pops into our minds is choosing the right chart. It is not easy to make your choice without getting lost in the infinity of visual possibilities.

Bar charts, pie charts, line charts, scatter plots, radars, box plots, treemaps and many more. How to choose the perfect chart that for your data story?  We will be revealing the secrets of our brain’s perception and cognitive biases by applying this knowledge to our chosen field: market research and surveys.

If you do not know what DataViz is and why it is important, we invite you to read our first article. If on the other hand you are rather looking for concrete advice to build them by choosing the best visual elements each single time, then you have come to the right place.

Encoding data to optimize their perception

Its quite simple in principle, a good DataViz must make it possible to read the data clearly, precisely and quickly. In order to do this, you will need to look at how we perceive the different visual components in order to use them to best convey the information behind the data.

Old but Gold: positions, lengths, areas and angles

The book Graphic semiology by Jacques Bertin was first published in 1967.  Visual diagrams, networks and maps contained in this book remain a reference in the field up to this day. You can look at these as a few of the concepts used in DataViz before this name even existed. All the extracts below come from his book.

Jacques Bertin’s visual variables: Size, Value, Grain, Color, Orientation and Shape.

In 1986, Jock MacKinley published another reference which enriched Bertin’s work on the aspects of cognitive perception for this visual encoding. The extracts below are from his book Automating the Design of Graphical Presentations of Relational Information. In the latter, he ranks, organizes and classifies the visual variables according to their precision and to their type: quantitative, ordinal (qualitative) and nominal (qualitative).

Jock MacKinley’s work on visual variables

About color choices

Colors are already mentioned in Bertin and MacKinley’s works but their particular role requires some more clarification. There are three fundamental roles colors can take for data to make better sense in data visulization:

  • Nuance, using similar color shades
  • Contrast, using very different or even opposite color tones
  • Contextualize, reflecting reality to categorize

Regarding contextualization, it can often be used in conjunction with nuances or contrast. For example, to visualize the results of a simple dichotomous question, with a yes and no answer, you will usually contextualize the no with a red and the yes with a green while using two practically opposite colors on the color wheel.

Storytelling with pictures and charts

If you want your data to convey the proper meaning and stick within the viewer’s mind, it must be simple but it should also include storytelling elements. It is not really about encoding data with images, but you can use pictures in a way that serves contextualization.

Some infographics use images extensively but for DataViz, you have to learn how to use them sparingly so that they do not steal the spotlight away from the actual data. These storytelling elements are an important part of creating data visualizations and knowing the story you want to tell can be useful when selecting the perfect chart.

To choose the ideal graphical representation for your data, the central question is: what do you want to highlight? The answer to this question can be: a comparison, a trend, a distribution, a hierarchy, a temporal evolution and so on. This is the starting point for your chart choices and the one chosen by the Extreme Presentation site and its brilliant Chart Chooser.

Finally, you must also ask yourself about the storytelling dimension that will always accompany DataViz:

  • What is the message you wish to convey?
  • What is the story you want to tell?

However, another starting point for visualizing the results of a survey would be quite simply: what type of question was asked?

That will give you a lot of information about the type of results and therefore about the best way to encode them effectively

The case of DataViz for market research and surveys

To illustrate the data from a market study, you must first look at how to better visualize the results of a single question, in other words a survey. In this case, here are all the parameters you should consider:

  • the type of question: dichotomous, multiple choice, likert scale, etc.
  • the type of answers and therefore of the actual data: discrete quantitative, continuous quantitative, qualitative nominal or qualitative ordinal
  • the number of choices in the answers
  • the discrepancies between data and orders of magnitude, how wide is your data
  • what you want to show: comparison, distribution, (de) composition, relationship, trend, evolution, etc.

From there and from the previous information about visual encoding, we have created the following summary table:

# choicesType of answerPreferred visual variablePreferred chart
Dichotomous2-3Nominales OpposéesColor hue, angle, areaPie charts
Multiple choice with a single answer3-10NominalPosition, color hue, length, angle, areaBar charts, treemaps
Multiple choice with multiple answers3-10NominalPosition, color hue, length, areaBar charts, treemaps
Likert scale and other semantic differential scales to measure frequency or nuance answers4-6OrdinalPosition, color saturation, color hue, angle, length, areas

Pie charts, bar charts, 100% stacked column charts

 

 

Rating scale (NPS)3-100NumericPosition, color saturation, color hue, length, areas

Figures, 100% stacked column charts

 

Ranking3-5OrdinalPosition, position, position, color saturation, images, color hue, length, areaBar charts, treemaps, scatter plots on 1 or 2 dimensions
Open-endedTextualPosition, area and volume, color saturationWord clouds

Let’s dwell deeper into more details with concrete examples for simple survey / poll questions. Each of the following question types will be accompanied with visual choices and each choice will be explained. 

Choosing the right chart for each of the 7 simple question types

There are is an infinite number of charts and visuals that can be used to represent the results of a survey. The summary below does not mean to be exhaustive or even set in stone. However, this is another thought starter for making your choices to visualize survey data.

When in doubt, never go for the more complicated solution. Moreover, in an interactive DataViz, data varies and sometimes widely, so it is better to select a chart that can bear the magnitude and variation.

Enough with the chatter, here is our thought starter to view the results of a survey question.

1 – Dichotomous questions (DQ)

Dichotomous questions have the distinction of having two generally opposite answer choices. Whatever the results are, there is rarely anything better than a good old pie chart to visualize them.

As for colors, choose the ones that better reflect the answers while creating enough contrast so as to clearly distinguish them.

Question example: Do you use Facebook?
Answer choices: Yes / No / DNA

Click to zoom!

2 – Multiple choice with a single answer (MCSA)

These questions have answer choices that are generally nominal. There are usually between 3 and 6 choices, which makes the use of the adjacent charts quite practical.

Question example: If a small local public facility should open tomorrow near you, which of the following would best suit your needs?
Answer choices: Parks / Sports elements / Outdoor games for children / Outdoor games for senior

If the difference between data points is very large, treemaps will be better suited as they allow more precision when judging the order of magnitude. Conversely, they should be avoided when the proportions are close as in this example.
Between a pie chart and a bar chart, it’s interesting to note that whenever the difference between data points is small like here, you should select the bar chart which is better to appreciate these minor differences.
Given the importance of position in perception, when one uses a bar chart, it is often better to rank it in ascending order.

Click to zoom!

It is sometimes interesting to choose a pie chart if you want to represent the shares of an oligopolistic market or more generally if you want to show that all the answers are part of a whole (percentage).

3 – Multiple choice with multiple answers (MCMA)

Quite similarly to MCSA, these questions often have between 3 and 6 nominal answer types.

The only difference, however important, with the MCSA type is that the set of answers does not represent a whole. Some respondents may opt for 4 choices while others only for 1. As a result, the total sum of responses is biased by the number of choices made by each participant.

Exemple: Why do you not want to install a payment termnal in your business?

Click to zoom!

Therefore, a pie chart will rarely an interesting choice for these questions. We will rather choose bar charts or treemaps depending on scattered the results are. If the differences in magnitude are large, then you should opt for a treemap and if they are small, choosing a bar chart will get you that additional precision.

Click to zoom!

4 – Differential semantic scales (DSS)

This type of questions includes likert scales, if you don’t already know, likert scales are those questions whose answers range, for example, from strongly disagree to strongly agree.

There are generally 4 to 7 ranked choices, which is why the use of color hues are often very meaningful. When it comes to frequency for example, you can choose color hues from lighter to darker tones and also contextualize your data with your main color choice.

When it comes to the type of chart, bars and 100% stacked bars work pretty well for this type of question.

Exemple: How often would you say you eat meat? Answer choices:

  • At least once a day
  • Multiples times per week
  • At least once a week
  • Less often
  • Does not eat meat at all
Question example: About the following sentence: “Music is an important element in my life”. Would you say that you …?

Likert scale answer choices:

  • Strongly disagree
  • Disagree
  • Agree
  • Strongly agree

Whenever both sides on the scale are opposed, for example with a likert scale, you can use divergent color hues as in the visuals above.

5 – Rating scales (RS)

For a simple rating scale with numerical answers, a good way to visualize the results is sometimes to simply display a figure. You can also use simple statistical calculations such as the sum, the average or even the median when necessary. The numbers themselves are getting a lot of attention from the brain. Therefore, it is better to select only a few and choose them well. If there are too many of them your DataViz will only be more confusing.

For an NPS (Net Promoter Score) or even for other scores that you wish to categorize, you can use all the tricks from the semantic scales by taking advantage of the ranking:  Detractors / Passives / Promoters.

Question example: on a scale of 1 to 20, how would you rate this service / product?

Average
11.65 / 20

Question example: On a scale of 0 to 10, how likely are you to recommend the brand / product to a friend or colleague?

Click to zoom!

6 – Ranking questions (RQ)

For these questions, the respondent is asked to rank several elements according to a criterion. The data is ordered but often have a nominal characteristic. You should encode the nominal aspect with different color tones and the ordered aspect with a bar chart for example, not forgetting to display the most important element first.

You can also use a more classic two-axis chart, keeping in mind that when it comes to ranking, the highest position will be immediately perceived as the most important.

Question example: rank these 3 fast-food restaurants by preference
Answer choices:
– Burger King
– McDonald’s
– Quick

Click to zoom!

Question example: name the first 3 brands that come to your mind

7 – Open-ended questions (OEQ)

Finally, the last type of questions we will be dealing with here are open-ended questions. They have a specificity compared to the previous ones: they are 100% textual.

Analyzing an open-ended question is not an easy task, trying to visualize then answers without taking the time to grasp its substance is tricky at best. There are nowadays a few tools that allow advanced intention analysis. However, a simple way to view the content of an open-ended question without analyzing its meaning is to use a word cloud like the one here.

The sample word cloud above is based on a the article you are reading

Note that all of these choices are suitable for the results of a simple question. However, when it comes to cross-referencing data between questions or adding a temporal or geographic component, the choices will obviously differ although the principles used will remain applicable. In theses cases, you will have to use other kinds of charts such as: line charts to show an evolution, scatter plots when you are looking for trends or maps when you are looking into geographical data.

All the information you read here can be used with infographics which are also quite useful when communicating complex data. You might wonder about the difference with DataViz then? The single most important difference is also the game-changer, and that is interactivity,

Riad Mawlawi, Digital & Business Development Director at Sunergia

DataViz Example

Most posts in the Market Insights blog offer DataViz upon simple free subscriptions. Most of them will be in french but If you wish, you can still easily have access to those. They have been created by our market research team and can be found at the end of our blog posts.

Finally, here is one of the latest DataViz published for our post about the most famous brands in Morocco in 2020. Do not hesitate to click on the images which represent the socio-demographic criteria (sex, age, income, housing area, region) or on the categories (Automobile, Cosmetic, etc.) to see the treemap evolve with to your choices.

DataViz Example

Most posts in the Market Insights blog offer DataViz upon simple free subscriptions. Most of them will be in french but If you wish, you can still easily have access to those. They have been created by our market research team and can be found at the end of our blog posts.

Finally, here is one of the latest DataViz published for our post about the most famous brands in Morocco in 2020. Do not hesitate to click on the images which represent the socio-demographic criteria (sex, age, income, housing area, region) or on the categories (Automobile, Cosmetic, etc.) to see the treemap evolve with to your choices.

Share This