When does it make sense to split a visualization into multiple charts?

[k25]

I recently faced a visualization task that turned out to be a bit trickier than I first anticipated. The goal was to provide an overview of the workload of selected departments in the organization measures in number of tasks pr month. It was requested that the visualization went as far back in history as the data allowed.

Initially, displaying all selected department in a time series chart seemed like a simple and reasonable approach. However, I quickly realized that the visualization was cluttered and difficult to interpret. In terms of comparing departments and evaluating tendencies over time the chart simple wasn’t very useful. It felt like it was trying do much more than its capacity allowed.

To improve visualization, I tried out different approaches:

• Compressing the data: Aggregating to quartiles or years to simplify the chart. It definitely made the time series easier to interpret, but with the cost of lacking details.
• Splitting the visualization into multiple charts: This approach led to one chart pr department and thereby the focus became each departments individual development rather than comparison between departments.

To me it seems impossible to provide both details, overview, comparison, and individual trends in one single visualization, but I wonder, at what point does splitting a visualization into multiple charts become more effective than keeping everything in one?

Splitting the visualization into multiple charts, clearly introduces some challenges as well. While it becomes much easier to follow the development of each department individually, comparing departments with one another suddenly becomes more difficult, and instead of comparing lines or bars within the same chart, the viewer must shift attention between multiple charts.

Another issue is the axis scales of the charts. Using different scales might improve the readability of individual charts, but if scales on axes are no longer the same, comparison becomes an even more difficult task.

This issue leads to one important question. What is the purpose of the visualization? If the main goal is to compare the departments to each other, one single visualization would be preferred, even if it would mean a loss of detail. However, if the main goal is to understand how each departments workload evolves over time, multiple charts would be the better choice. Ultimately, the answer to when it makes sense to split a visualization into multiple charts, depends on the question it is intended to answer.

In my case, I spoke to the users, trying to figuring out what was more important. In the end I decided to do the split. I would be interested to hear how others approach the discussed trade-off. When does the complexity of a single visualization outweigh the advantages of having all the data in one place? And how do you decide whether to prioritize comparisons between categories or the individual trends within them?

[RossC0d3]

I agree that the purpose of the visualization should be the starting point, and I think you raise an interesting question. However, I would argue that the choice between a single chart and multiple charts is not necessarily a binary one.

I am currently working on a project where I analyze production data from a manufacturer of glued laminated timber. The aim is to identify factors that influence production speed and, more importantly, to understand why production is more stable during some periods than others. The data covers different machines, shifts and production conditions, and I have therefore encountered several of the challenges you describe.

One aspect worth considering is the effect of aggregation. As you point out, grouping observations by month or year can make a visualization considerably easier to read. The question, however, is what information is lost in the process.

In my project, variation in production speed is one of the main things I want to investigate. If the data were aggregated into monthly averages, the result could be a clean and simple chart suggesting fairly stable production, even though production speed varies considerably from day to day. In such cases, aggregation risks hiding exactly what I am trying to understand.

An alternative would be to visualize the variation directly, for instance through boxplots or rolling standard deviations. These approaches could make the visualization easier to interpret while still preserving important information about variability.

Your point about axis scales is equally relevant. Independent scales can make individual trends easier to perceive, but they may also lead to misleading comparisons. Two production shifts could, for example, appear to have similar fluctuations simply because each chart is scaled to its own range, even though one is considerably more stable than the other.

Conversely, identical scales can make smaller fluctuations almost invisible when the categories differ substantially in magnitude. There is therefore no universally correct solution. Depending on the purpose, normalizing the data may be a reasonable compromise, although it introduces an additional layer of interpretation for the viewer.

Coming back to your main question, I wonder if a combination of approaches could offer a better balance. One option is small multiples with shared axis scales, possibly supplemented by a common reference line where the departments are comparable. This would allow the viewer to focus on each department individually while retaining a common baseline for comparison.

Interactive visualizations offer another possibility: presenting an overview first and allowing users to select specific departments for closer examination. While this does not resolve every trade-off, it reduces the need to fit all information into a single static chart.

I think your decision to discuss the purpose with the users was probably the most important step in the process. A visualization that addresses the questions users actually have is more valuable than one that simply includes as much data as possible.

Did you consider combining an overview with more detailed charts, rather than choosing one approach over the other?