An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.
Choosing a chart
Chart choice follows from two things, the question being asked and the type of each variable. Most business questions belong to a small set of families. Comparison between categories, change over time, composition or part-to-whole, distribution of one variable, relationship between two, and deviation from a target. Each family has a few forms that read well. Variable type then narrows the choice, because a categorical variable needs an identity channel and a continuous one needs a magnitude channel. Starting from a gallery of chart pictures inverts this, and it is why so many charts answer a question nobody asked.
Why it matters
Say the sentence you want the reader to walk away with, out loud, before you draw anything. If the sentence is that the north branch grew fastest, you need a comparison of growth rates, which is a sorted bar chart. If it is that orders cluster at the end of the month, you need a distribution over days. The sentence names the chart. When you cannot say the sentence yet, the chart is not the problem you have.
A manager wants to know whether delivery times are becoming more consistent. Which form best answers the question as asked?
Formulas
Worked examples
A retail team asks which of twelve stores improved most on same-store sales this year, and receives a stacked area chart of monthly revenue by store.
The question is a comparison of one derived quantity, the change per store, across twelve categories. The stacked area chart shows level over time with eleven of the twelve stores sitting on shifting baselines, so no store other than the bottom one can be read. The form that matches the question is a sorted bar chart of percentage change, one bar per store, largest at the top. A slope chart connecting last year to this year also works, and adds the levels back at the cost of some crowding.
A finance analyst is asked whether invoice payment times have got worse, and prepares a line chart of the monthly average days to pay.
An average answers a level question, not a distribution question. If a small group of very late payers has grown while most invoices are still paid on time, the mean moves and nobody can see why. Plotting the distribution of days to pay, as a histogram for each of two periods or as a box plot per month, shows whether the whole distribution shifted or only the tail thickened. Those two findings call for completely different responses, one a process change and one a collections escalation.
Common mistakes
- ✗There is a correct chart for every dataset. There is a defensible chart for every question. The same dataset supports different forms depending on whether the reader needs a comparison, a shape, a trend or an exception.
- ✗Pie charts are always wrong. They are weak for comparing similar-sized segments, because readers judge angle poorly, but they communicate a single dominant share against the rest perfectly well. The real constraints are that the parts must be exclusive, exhaustive and non-negative, and that there should be few of them.
- ✗A time series always calls for a line chart. A line implies that the values between the plotted points are meaningful. For quantities measured in discrete unrelated buckets, such as revenue by product category, connecting the points suggests an interpolation that does not exist.
- ✗More series on one chart is more informative. Past roughly five or six lines, readers cannot trace individual series and the chart becomes a texture. Small multiples, one small panel per series on identical axes, keep the comparison readable.
Revision bullets
- •Question family first, variable type second, chart form last
- •Families: comparison, change over time, composition, distribution, relationship, deviation
- •Categorical variables need identity channels, continuous ones need magnitude channels
- •Part-to-whole forms need exclusive, exhaustive, non-negative parts
- •Small multiples beat a crowded multi-series chart past about six series
- •Say the takeaway sentence out loud before drawing anything
Quick check
A manager wants to know whether delivery times are becoming more consistent. Which form best answers the question as asked?
A survey lets each respondent tick every social platform they use. Which presentation of the results is defensible?
Connected topics
More in Visual Communication
Sources
- Munzner (2014)Munzner, T. Visualization Analysis and Design. CRC Press, 2014.Separates the task a chart serves from the data type and from the encoding chosen.
- Mackinlay (1986)Mackinlay, J. "Automating the Design of Graphical Presentations of Relational Information." ACM Transactions on Graphics, 5(2), 110-141, 1986.Introduced expressiveness and effectiveness as formal criteria for matching an encoding to a data type.
- Cleveland, W. S., & McGill, R. "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods." Journal of the American Statistical Association, 79(387), 531-554, 1984.The accuracy evidence behind preferring bars and dot plots to pies and bubbles for comparison tasks.