This article was written by Patrik Lundblad and originally appeared on the Qlik Blog here: https://www.qlik.com/blog/combo-chart
In some of my earlier blog posts, I’ve gone through some more novel visualizations, such as the Sankey chart and distribution plot. But, for this post, it’s time to go back to the more common Combo chart, also sometimes referred to as a dual axis chart.
Already from the name, we can understand what the purpose of the chart is – to have some type of graphical combination. Most often, it’s a combination of bars, lines or markers, and, depending on the data, it could have one or two quantitative scales which the data is mapped against.
Let’s start with what makes up the chart.
In the same way as is done in a bar chart, the data encoding in a Combo chart uses bars and their height to represent a value. This means that we are mainly looking at the magnitude of each individual value as well as the comparison between values – learn more in my post: “Third Pillar of Mapping Data To Visualizations: Usage.” Therefore, to understand the magnitude of a value, we require a qualitative scale that starts from zero (or a negative value) and goes to the maximum value (or zero).
For a line, we use the position of a point on the line to represent the value. Hence, a line is not as good at showing magnitude; rather, it’s the difference in position between the values that are of interest. We can see a trend between the values, as well as over the full measurement. Now, if you are looking at trends, the most common thing to do is to adjust the quantitative scale to go between the minimum value and the maximum value.
An alternative is to use zero as a scale, but then the trend is not as easy to see.
Using a scale that doesn’t start at zero is different from the scale we used for bars. Therefore, a Combo chart often has dual axes with bars mapped against one axis and lines against the other.
Markers are often symbols where the position is mapped against the value. Hence, they are neither great at showing magnitude nor trends. But, they can be good as a comparison between the markers or between the marker and a bar or line. For markers, we can use any type of scale, and it doesn’t matter if it starts at zero or at a minimum.
For this blog, I'll leave out any more references to markers or other methods to add data to a Combo chart and instead focus on the bars and lines.
When To Use Combo Charts
The main purpose of a Combo chart is to compare the difference between two or more values and to look at the correlation between them. Because the magnitude and the trend analysis uses different scales, we often end up with two axes in the chart – either on the same side or, as is more often seen, with one axis on each side.
Since we are looking at trends, you should use Combo charts when your data has time values, so that the qualitative axis shows the order of the data. In the example below, we have now taken our sales data and combined it with the knowledge we have about our discount, so that we can see some pattern starting to emerge.
The month of April has been bad for sales, despite (or because of) the increase in discounts. Since the data is aggregated, it might be worth studying this month closer to see who got discounts and why. We can also see that both sales and discounts fluctuate over time, but there doesn’t seem to be any correlation between these variables.
When Not To Use Combo Charts
For me, a Combo chart is best used when looking at time data and trying to distinguish trends and map them against other measurements to find correlations. If you use data that isn’t ordinal/time-based, then you’re probably more interested in correlations than trends. For this purpose, you can use a scatter plot instead.
Here is what nominal data looks like in a Combo chart.
Compare this to what it looks like when we use a scatter plot. From the scatter plot, we can potentially see a linear pattern. We can also see that there are groups of data.
Issues With Combo Charts
There are, of course, some issues using a combo chart with dual axes. First and foremost, it’s a question of which measurement is mapped against which quantitative axis, especially if you use the same encoding for your measurement. In the picture below, I've added max discount as a measure using bars. Suddenly, it becomes tricky to know if the middle bar should be measured against the right or the left axis.
Another issue that can occur is if you use only lines in your combo chart. As each axis can be scaled either from zero or from a minimum value, we can get three different visualizations based on how we scale the axis. As you can see from the picture below, it can become messy to pick the right one.
If you use a Combo chart correctly with the right data, the right encoding and the right scale on the axis, it may serve you well and unlock hidden insights. If, on the other hand, you get it wrong, chances are you’re just adding more confusion than needed and should go with another visualization.
Lastly, something I learned a couple of years ago is that, when you have measurements of different scales or units and you want to compare them to each other, what you can do is to compare their change as an index. In the visualization below, both of my measurements start at an index value of one – which means that the value at the first position divided by itself equals one. What you do next is to take the value of the second position and divide it by the first value to see how the index has changed. Continue doing this across all your values, and you can see both how the trend changes, as well as the correlation against another measurement, all using the same unit and scale.