Time is a common dimension for visualising data and people have created temporal visualisations for a very long time. The ‘earliest known attempt to show changing values graphically’ is a planetary movements chart, shown below, with time running horizontally dating from the 10th or 11th century. The timeline — mapping a sequence of events by time — is a classic visualisation form. (For a history of the timeline, have a look at the book Cartographies of Time by Daniel Rosenberg and Anthony Grafton, which has a wonderfully rich selection of illustrations.)
But visualising data in timelines can come up against the strange ways humans understand and experience time. We may agree that time is passing at a uniform rate, but there are complications. There are time zones and daylight savings. The human experience of time passing does not always match up to the rate of the clock (“time flies when you’re having fun”). Different cultures conceive of the shape and orientation of time differently. Which direction feels “natural” to draw the arrow of time can be influenced by the writing direction we use (this timeline in Arabic has time going right to left). Looking back on the past, more recent events appear in greater focus than those farther back.
In this post, I discuss how the ways we think about time shape the data we create (even if it’s not immediately obvious), taking the example of historical time. Illustrated by my own work visualising data from digitised museum collections, I explore how the designer can choose to either emphasise these peculiarities or to conceal them, highlighting other characteristics of the data instead. Time is fundamental to making sense of digitised museum collections (data that describes museum holdings: objects, artworks, texts, etc.) and visualisation can be a powerful way to analyse, explore, and present patterns and stories in this data, but it is a domain where the oddities of time data can really rear their heads.
Take uncertainty, for example. A common issue with historical dates is that they are often not precisely known, but estimated to a stretch of time. Date information for an item may be given as a span (eg. 1940–45) or accompanied by some qualifier like “approximately” or “circa.” For artefacts from a very long time ago, this uncertainty can be very large.
The purpose of your visualisation and what you want to draw attention to will inform how best to deal with this. If visualising a large number of data points with uncertain date information, if you start slapping error bars on everything you’re in danger of that uncertainty becoming the entire communication of your visualisation. And your intention in visualising the data may not be quantitative analysis or to present an indisputable temporal order. You may, instead, be interested in revealing general patterns and connections across a large collection.
In a project visualising Cooper Hewitt Smithsonian Design Museum collection data (which I’ve written more about here and here) I experimented with using a collage layout to position data with uncertain dates. Photos serve as the data points, positioned at a random place within their timespan (time running horizontally). All the images are then spread out so that nothing is overlapping while keeping some portion of each photograph within its timespan (see the diagram below). The design does not draw attention to the fact that many of the dates are uncertain, but the overall impression also does not encourage the viewer to read the visualisation for precise date information. In a similar vein, others have explored using a normal distribution or spreading items in a grid structure to organise items within time spans like this.
Another issue is that, when we think about the past, we tend to chop time into chunks, for example historical periods (eg. the Ming dynasty 1368–1644). There is also a tendency toward round numbers; we organise the past into centuries or decades.
It’s with this kind of structure—round numbers or named periods—that date estimates are often made. It may not be immediately obvious this is at play as time in museum collection data may still be expressed in numbers or a time format. But these can be the result of museum database software translating from a textual input, for example “18th Century” or, indeed, ‘Ming dynasty’.
Misleading shapes can result. For example, if positioning data points at the middle of their date span, many items dated “18th Century” will form a stripe at 1750, suggesting something interesting is happening when in fact it is just an artifact of how time is conceived of and recorded.
Visualisation can be a way to expose this kind of bias in data; the most dramatic changes will be visible at boundaries. In a project visualising a digitised collection of historical photography from the V&A Museum, London, (which I’ve written more about here and here) I organised the data vertically by photographic technique and horizontally by time. Similar to the Cooper Hewitt project, I positioned items at a random point in their date span.
Visualising the data this way reveals a dense cluster of daguerrotypes (a type of early photograph) at 1840–60 because many of these records are dated “mid 19th century” (which is translated by the database software to 1840–60). But there is also a sparser scattering either side of this. Checking the records revealed many of the items outside the middle cluster are dated “mid-19th century” with a hyphen. The software script translating this to numbers had tripped on the hyphen, converting the span to the full century: 1800–1899. The visualisation betrays this way of thinking of historical time in discrete chunks, coupled with a metadata error.
Drawing attention to the unexpected shapes historical time data can form, however, may not be your focus. In a project visualising a dataset of historical portraits from the Nordic Museum, Stockholm, I explored positioning example artworks, selected at random, along timelines. (You can try out the prototype here). Visualising the data this way offers tastes of illustrative images produced at different times, instead of revealing the overall quantitative shape of the dataset.
Many of the artworks in this dataset were dated to the century. And while this is not emphasised in the visualisation design, that does not mean its effect does not seep through. Two portraits — one dated 18th Century (translated in the data to 1700–1799) and the other 19th Century (translated to 1800–1899) — can end up right next to each other, as in the example below. From the clothes depicted in each painting, they are from very different times. But if we were to go strictly by the data, this pair could have met at a 1799 New Year’s Eve party — it seems unlikely!
Multiple dates for a record
A further complication with historical data is there may be multiple dates attached to a record, even just for the item’s creation. A photograph, for example, can have a different date for when the photograph was taken and when it was printed. Items can be reproductions of something made much earlier. Historical buildings may have many dates attached capturing rebuilding, extensions, remodelling etc. Which is the “right” one to visualise? The question here really is: What are you hoping to see/show with your visualisation design? These cases likely need to be dealt with on a case by case basis.
With artworks there can even be a mismatch between the time an image depicts and the date it was made. These prints in the Nordic Museum collection were made 1850–60, but they all depict earlier historical figures in earlier costume. If you are visualising this dataset hoping to see patterns in historical clothing through time this is going to be a problem.
The Nordic Museum actually includes a category in their data, “tidsanda,” which roughly translates as “time spirit.” It is used to record the historical time an artwork or design depicts, if this is at odds with its production date. This unusual category has creative potential for visualisation!
How we think about the world shapes the data we create to describe it. Historical time data is a classic example of this: the situation is more complicated than it may initially seem. Visualisations can be designed to deliberately expose the workings of this kind of data, but may also be designed to conceal, drawing out different stories in the data instead.