Change over time

Analyzing change over time means that you want to identify shapes created by multiple data points, like trends or temporal patterns. The best geometry to find these shapes is the line. While you can use bars to display data over time, they favor point comparison, so this is not a true analysis of change. Things are much more blurred in practice, but this distinction between shapes and points is useful to clarify the message you want to communicate.

Five types of change

You can display at least five (plus two) types of temporal change:

  • Linear: This is the most common: a simple line chart where a straight line connects two adjacent data points and time progresses linearly from left to right, and time units are spaced evenly.
  • Jump: Similar to linear but, instead of a straight line, adjacent data points are connected using two lines with a right angle, making a step chart. Time units are less likely to be spaced evenly. The idea is that change is sudden, which makes it ideal for displaying price changes.
  • Cyclic. When there is a known marked and predictable seasonality (tourism is a good example), a linear timeline will not be very informative because it emphasizes what you already know and minimizes other changes. So instead of a linear timeline, you focus on sections of the temporal pattern. For example, how was tourism in January over the last ten years? It about February? Changes in each section become more explicit and insightful.
  • Relational: scatterplots usually display the relationship between two variables at a given point in time. But if you order your data points by date and connect them, you can see how the relationship between the two variables changed over time. The “connected scatterplot” is not always easy to interpret but can be very insightful. Several economics curves, like the Beveridge curve, take advantage of this.
  • Circular: When you display a time series in polar coordinates. This design is useful, for example, to compare daytime and nighttime since we are familiar with the clock metaphor. A continuous line circling in a display that represents a year is also frequent, if less useful.
  • Rank: Instead of displaying the actual data over time, you calculate rankings and show them instead.
  • Animation: in some cases, you don’t have to represent time. You can create a series of charts, each for one moment in time, and present them in sequence. For the animation to work, the temporal pattern must be simple enough so that the audience can make the connection. One of the most famous examples of using this technique is Hans Rosling’s first TED Talk.

The x-axis

For a correct display of time, it is necessary to understand time as a quantitative variable, ensuring that the spacing along the axis is proportional to time intervals, otherwise trend’s shape.

The y-axis

When reading a line chart, we get our cues from slopes. Except for being positive or negative, slopes don’t have an intrinsic and absolute meaning because the chart’s aspect ratio and the axes’ scales affect it.

The y-axis scale doesn’t have to start at zero, so you can focus on a more useful concept, resolution. Low resolution means that the scale is too wide and lines become almost flat, and relevant variation is hard to spot. On the other hand, high resolution means that the data’s minimum and maximum values are close to the scale’s beginning and endpoints. The high resolution makes variation easy to spot, but we risk focusing too much on the intensity of change. To minimize this, it is why it’s always advisable to represent more than one series so that you can compare slopes. Reference lines and bands of variation are also useful.

Number of series and the spaghetti problem

Your main goal when using a time series is to identify relevant trends and temporal patterns. Untangling this can prove not easy if you have too many intertwined series (the spaghetti). You don’t need to remove them: gray out the series that contribute little to your core message but are useful to provide context. Another option is to split the chart into small multiples.