It’s probably not ironic that I’ve changed and rewritten the overall title of this blog post probably a dozen or more times. It’s true that the theme for this blog is that each title poses a question, so that’s a very specific (and probably unnecessary) restriction. But I’ve had many other stabs at this title, some of which include:
- Should a visualisation’s title always state its findings?
- Should a visualisation’s title always be a statement?
- Is it OK to pose a question as a visualisation’s title instead of answering one?
- What should you include in a chart’s headline?
If it’s going to get in this blog, then the overall title always *asks* a question. But a standalone visualisation should always be titled. Even the most ingenious and impactful visualisations can be improved with the context and explanation that a title brings; many are useless without one.
In many ways this follows on from my last post and previous ones before it about stories and goals. The subject of this blog comes from a visualisation challenge set by Cole Knafic in her blog post where she states that she had a number of reasons for concern, but challenged the readers to submit their own makeovers before she published her own. Here’s the original visualisation from the Economist, with title included:
Hurricanes in America have become less frequent
My last post about the point of visualisations has spurred me to take my own advice of setting out my objectives for any given visualisation. So here’s my take on the original visualisation:
What works well:
- Concise, compact layout
- Using decades smooths out yearly random distribution
- Distinction between stronger and weaker hurricanes works well
What could be improved:
- Axis labels are a bit of a jumbled mess
- Vertical axes on the right is unconventional and axis lines are obtrusive
- “All hurricanes” trend is in the blue text designating category 1-2, which is confusing
- Is it valid for the two trends to overlap (would have more validity to compare categories 3-5 versus 1-2 rather than versus the whole)
- Stacked bars make it particularly hard to compare the lighter blue categories year to year
- There might be visual upward and downward trends, but can we state this with certainty or draw conclusions from it?
- The final decade is only half-present. While that makes sense, it’s a bit out of place
- There’s no explanatory text (although we don’t know what may have been included in an accompanying article, the viz is likely to be disseminated on its own, as indeed it is here!)
My goals for a makeover:
- Find a story in the data
- Remove gridlines
- Make horizontal axis more legible
- Discuss trends but don’t state them as fact
- Use a different visualisation type that allows for easier comparison of all hurricane types
- Include more explanatory text
- Give more consideration to the title and whether we can state findings there
- If I’m going to include the incomplete decade, remove it from findings or make it clear that it’s not incomplete.
Before I move onto my makeover, a word about the first point: “Find a story in the data” – wait, don’t I keep blogging about the fact that its not always necessary? In this case it’s specific goal. It doesn’t have to be, but we are making over a viz from a news outlet: its job is to tell a story. What I want to do is find whether that “story” is there and whether it’s true, but then consider how we can frame this. Can we assert our findings as fact?
Here’s my makeover:
(edit: minor change to image since first publish, iterating on feedback about orientation of axis labels and left-aligning title)
Mission accomplished in terms of my goals. In separating the two grouped categories out of the stacked bar chart form we can easier compare numbers of hurricanes in each category from decade to decade and see trends. I state the noticeable change in difference between strong and weaker hurricanes, and call this out by separating my chart with colour (yes, they are hues of red and green, but they are colourblind-safe!)
Andy Kriebel advises that it’s good to phrase a chart’s title as a question, and that’s my biggest takeaway from this exercise, perhaps the biggest improvement. I don’t say that we are getting fewer hurricanes, or that the hurricanes we are getting are becoming stronger by comparison, because I’m not comfortable stating that. I’m a statistician by training and there’s no rigour or analysis to back the assertions up. But what I can do is present the data in such a way that I make it easy for the reader to explore and come up with their own conclusion, and that conclusion will probably be similar to the conclusion as stated by the Economist, but if a reader wants to conclude that as fact, they’ll need to explore, experiment and research further.
In framing a title as a question we remove any liability in stating something that can’t be proven. My explanatory text mentions the words “it seems” or “appears to”, and perhaps these ploys are more the tactics of an exploratory, rather than an explanatory visualisation; the types of visualisation that rely less on a story, but still require their own points and goals in order to be viable.
Finally, this is only a remake and if I were to look at my version I would agree that my own could be improved. Is a dumbbell chart the most intuitive for users? Is my explanatory text too wordy/repetitive? Should my vertical axis text not be vertically aligned? Iteration and practice are still key, and expertise some distance away!
“Is it valid for the two trends to overlap (would have more validity to compare categories 3-5 versus 1-2 rather than versus the whole)”
Well, I’d massively overlooked this in my own critique, but seeing someone else make the point it really stands out and is nagging. The two cannot possibly overlap, so it seriously challenges the validity of the ‘trends’, and they absolutely cannot be described in a linear fashion.
As for how to use titles, I’ve long struggled and agonised over this. There’s a lot of good advice in this blog I will try out in the ongoing search to establish my own style.