Using heatmaps to visualise survey data

Thursday, Sep 11th, 2014

We're happy to announce that we've added another client to our portfolio for whom we worked on analysing survey results. The survey targeted C-level executives and the goal was to understand the company's position in innovation. In this blog, I'd like to dive a bit deeper into one of the visualisations we used, namely the heatmap:

This heatmap shows the respondents in columns and the topics in rows. Each topic is composed out of several questions. To further improve the chart, we ordered both the rows and the columns so that low-scoring cells appear on the top-left and high-scoring cells on the bottom-right. What can we see on this chart?

  1. The executives feel that the company is scoring really well on areas like 'Commitment to learning', 'Strategy alignment' and 'Independent'.
  2. 'Risk taking' and 'Competitve aggressiveness' are key areas with opportunities to improve.
  3. Executive 'O81' is somewhat of an outlier. He / she has a more pessimistic view on the organisation.
  4. There is a significant disagreement over 'Business results' and 'Innovation results'.

Advantages

Heatmaps visualise a lot of information while not overwhelming the user. This one contains about 400 data points. Yet, there are no numbers and no complex shapes. The only metric you evaluate as a user, is the hue of the cells.

If organised correctly, heatmaps present the big picture in an instant. You instantly see which topics score well and which ones don't.

Last but not least, heatmaps allow you to detect outliers. Some cells stand out next to their neighbours. This is something that the human brain notices immediately.

Drawbacks

Heatmaps are less precise than, say, a bar chart. With a bar chart, we can easily compare lengths and know exactly which datapoint is bigger. Comparing hues on a heatmap is less precise. If the values are more or less the same, it is hard to distinguish the colour of the cells. We think that's acceptable in this case, because we are not really interested in those small differences. We are looking for general patterns and outliers to these general patterns.

Another drawback is that a heatmap wouldn't work if we had a 1000 respondents. Maybe we could group the respondents by department, to get a reasonable amount of columns. But then you're aggregating again. In fact, we aggregated the questions already, in topics. For each respondent, we show the average of all the questions in a topic. Whenever there is aggregation, an underlying pattern might get lost.

Dive into the cells

To discover underlying patterns in each topic, we built another chart to accompany the heatmap. This one is a bit more technical.

What you see here, is the score distribution per area. "Risk taking", for instance, scores consistently low. And "Commitment to learning" scores consistently high. That's what we saw in the heatmap as well. The red line indicates the average.

The interesting thing about this chart is that we also see another pattern. A lot of the areas are bi-polar. This means they get a lot of high scores, and a lot of low scores. But only few center scores. In such a case, an 'average' is not telling the entire story. We see the bi-polarity in the heatmap as well, in the 'Innovation results' for instance. Indeed, there is strong disagreement between respondents on that row in the heatmap.

There are exceptions though. Have a look at "NEEDS". It is very bi-polar in the linecharts, yet very average in the heatmap. Apparently, everybody agrees in the heatmap but there are 2 different groups in the line chart. That's weird. How can we explain that? This one required a deep-dive into the questions of the "NEEDS"-area. These questions asked for different kinds of needs. Need for new ideas, for different communication methods, for entering new markets, ... . It turned out that everybody had strong opinions on these topics. Either they identified the need as very low, or as very high. Yet, nobody identified every need as either very low or very high. So they all average around the center. The line chart illustrates the strong opinions. And the heatmap illustrates the averaging around the center.

Conclusion

Heatmaps are an effective way of communication for survey results, if the number of data points is not too large. If you're aggregating results, it's worthwhile to investigate underlying patterns.

Kris
Data architect

Add new comment

Image CAPTCHA