Do you really need another time series plot?

I often get reason to think about Atwood’s law: “Any application that can be written in JavaScript, will eventually be written in JavaScript.”

And then it struck me, that for data analytics, and in particular when developing dashboards; “Any metric that can be visualised as a time series, will eventually be visualised as a time series”

What do I mean by this? Well, modern data setups have access to data over time, almost every metrics table I’ve ever worked with in a database has a time column (often a date). And then some other dimensions and measures.

If you need to present these measures to stakeholders through a dashboard, or any other analysis tool really (looking at you PowerPoint), it is easy to draw the measures as a time series. In fact, in many tools, the time series is the only visualisation that works out of the box, as others need additional more-or-less complex filters to ensure they only display the last data and not an amalgamation of the measures over time.

Here’s an example, it’s common that an analyst creates a beautiful heat map of some metric broken down per dimension. Perhaps it’s per country, and you get a nice colourful world-map visualisation. Or they present a histogram of the distribution of the metric e.g. per some end-user characteristic.

Inevitably, someone will ask “So, how does this compare to a year ago?”, in fact that question might be the only question you ever get, as the graph does a great job answering any other question. It’s surely possible to add a time-filter to the chart, but seeing trends and doing a side-by-side comparison is hard. Hence, the goto solution is to change the graph to a line chart. Instead of a colour per market, you get a line per market.

It’s a mess, there are so many countries, and the story of geography mattering to the measure is lost. However, since the graph now plots all data, for all countries and all dates, no one can ask questions where the answer is not in the graph.

The line chart time series is great at highlighting trends, but many times they introduce way too much additional information, increasing the cognitive load of the consumer. The original point may get lost to that noise, an insight hidden due to the scale of your Y-axis making some lines more or less disappear. A histogram with an interesting distribution may be boiled down to lines for mean, median and quartiles — which are way harder to relate to.

A worrying trend I’ve observed is as data analysts mature, they tend to default to the time series instead of exploring better visualisation options. It’s just easier to throw up all data on the screen, than to tell an opinionated story, in particular when you suspect someone will ask for the time-series anyway. Why do both? We all have a bunch of other things we need to get to.