I’ve been playing with box plots as a better way of visualizing and interpreting the results of our local employee engagement survey. Historically, our process has been to take the mean of the respondent’s answers to the questions and compare each against a benchmark- usually anything below a 4 on the Likert scale was good (or bad) enough to get it put onto the opportunity list. This has always struck me as a bit too generic for the purpose, particularly because it doesn’t take variance into account. The way the data has been used previously, however, has lent itself to this simple methodology as we have used the scores only to drive the right conversations with our employees, and as a basis by which to track general change as a whole. What this has left on the table, however, is our ability to better understand how the group really feels about a question. Ten people choosing a (4) should probably lead to a different course of action than four people choosing a (5), two people choosing a (4), and four people choosing a (3), even though the mean of both scenarios is (4). Here’s a sample data set I put together from a survey that would be scored on a seven point scale and the corresponding means:
At a glance, the averages do seem to tell a story; there are certainly some questions that are more highly scored than others. But look at what happens when we take the standard deviation:
That looks a little bit different, doesn’t it? Means that were quite similar to each other take on a new filter next to the SD. The differences are significant enough that we are almost certainly doing those who take the time to survey a disservice when using a mean only approach.
Let’s look at a box plot for the same data set:
Now I think we’re starting to tell more of a story. At a glance, we can quickly see exactly where there is general alignment around a particular topic, where people’s thoughts have a wide amount of variance, and where outliers are sitting well outside of the rest of the population. As a point of comparison, Q2 and Q3 have almost identical means (5.1 and 5.2 respectively) but look very different when graphed this way. I’m still playing around with how we can take more away from the numbers, but I think this is a solid next step. How else could you go about processing this data?
*Update: One caveat I’ve encountered is that far fewer people are able to derive information from the box plot than from some other visualizations. Although this is useful in analyzing the data, it will need to take a different format before it is widely distributed.