Bar chart races insights for 2021 with Abbott Katz

If bar chart races aren’t for everyone, then neither is every set of data right for the races. The races – or BCRS, as we’ve been calling them – commence their mad dashes to the finish line with multiple data series that dot themselves across time; and as such, one might assume that any data that feel at home in a line chart could be usably transposed to a BCR. We’ve already pointed to the line chart-BCR relationship in the first instalment, but we surmised then that some data might not be gladly attended by the former. Now, we can think conversely about data that might not – or might – lend themselves to a BCR.

Bar chart races insights for 2021

Guest post by Abbott Katz

It’s a concern to which BCR pathfinder John Burn-Murdoch has devoted some thought. He’s set forth three criteria for data aspiring to BCR entree, reproduced in the Stephen Gossett piece on BCRs we introduced in part one:

1. The values should change substantially over a given period of time.
2. The changes should include entities moving in or dropping out over time.
3. The data need to comport with the BCR’s marching order: “So anything where it’s ranking,” Burn-Murdoch adds.

Addressing the last criterion first, I’m not sure what numeric data series could not be ranked, as a matter of definition. Multiple data series bearing multiple data points – be they widget sales, population changes, goals scored, or yearly test scores, could all be ranked comparatively, it seems to me.

Points 1 and 2 are slightly more pause-giving and could be paired, as they address the two axes along with BCRs proceed – the horizontal (point 1), across which the bars unfurl, and the vertical (point 2), within which the bars rise and fall.

Burn-Murdoch likes BCR data series whose numbers change markedly, the better to spur the bars’ flight and foment the BCR frenzy we’ve come to know and love. No one, after all, is dismayed by the resolute stillness of a traditional bar chart; it is what it is, and there’s nothing more to say about it. But watch a BCR whose bars fail to zoom crazily across your display and the experience somehow disappoints, though I’m not sure why it should. We’re still learning something from the chart, aren’t we?

On the other hand, keep in mind that some bars that do exhibit Burn-Murdoch’s hyper zigs and zags may at the same time not uncork the thrills BCR aficionados associate with the viz. Consider two widget manufacturers, Acme and Archrival, and their ascending sales’ trajectories:

Note that in each year, and for purposes of illustration, Archrival always realizes 80% of the industry leader’s total. A resulting line chart would look something like this:

And what would the corresponding BCR look like? Good question; because, if it’s driven by the apparent industry standard, its chart bars will never move.

They’ll never move because the superordinate bar, the one bearing the highest total, is always immured in position, per our discussion in the first installment. And because runner-up Archrival always achieves the same fraction of sales as the market leader, its bar will necessarily hold fast as well, to retain its constant proportion to Acme’s.

We’ve thus returned to perhaps the most pause-giving element of BCRs as they’re typically conceived and executed – their structural insistence on a fixed-length, highest-total bar that needs to juggle two remits – the requirement to remain stationary while at the same time somehow reflecting the changes in the actual values it records.

And the workaround that tiptoes past that contradiction? Introduce motion in the value axis instead. We see in Burn-Murdoch’s BCR – along with every other one I’ve seen – that the axis continually recalibrates and repopulates its numeric markers in order to represent the bar values streaming beneath it – because if the values change, something has to move.

And while of course the subordinate bars do expand and contract, their respective spans are always expressed as a proportion of the bar on top, the one that never changes.

All of which begs the obvious follow-on question: must the BCR operate that way? I don’t see why. Again, I think the fixed upper-bar conceit has more to do with the coding predilections of the Tableaus and Flourishes who propel so many BCRs rather than some indispensable precondition of the charts.

I would thus submit by way of counterexample a demo BCR devised by your faithful correspondent fashioned in Excel 365, of all applications. It charts the counts of movie and television shoots in Paris across the 2016-19 period by the city’s arrondissements, or districts. Sparing you the technicalities, my BCR lets its bars happily plunge ahead as their numbers mount – even the top-ranked one (and by the way, Excel lets you easily modulate the speed at which the bars proceed, if you want to ratchet up the excitement).

Thus, for example, the totals for September 29, 2016 look like this, in excerpt (Paris comprises 20 arrondissements):

(The fourth and fifth digits in the values in the leftmost column denote the arrondissement). By April 24, 2018, the race has reached this stage:

Apart from leapfrogging the early-leader 16th, the 18th Arrondissement has actually stretched its bar, commensurate with its actual value. And why not?

Now as for Burn-Murdoch’s second proposition – that the bars (“entities”) should barge in and out of the chart even as they lurch en masse towards the finish – that caution that needn’t be taken literally. Remember that his BCR tracks city sizes across a 500-plus year swath, and as such, a good many of them could be expected to vanish or debut in mid-race, or even come and go (note that other BCRs plot city rankings over a range of several thousand years). But my BCR maintains its complement of arrondissements throughout; none emerge in mid-chart, and none disappear. But their totals and positions change just the same, and that should be good enough.

Now take the conjecture further. Imagine a BCR in which the bars pull along the horizontal dimension but never skitter up or down, sitting atop the same rung throughout the race instead. Are those data unworthy candidates for the glittery spectacle of a BCR?

Again, I’m not so sure. Does movement in only one direction – the horizontal – shrivel the “entertainment” quotient of the BCR? Is the perpendicular frenzy stoked by the leaps and nose-dives of the bars integral to the whole idea? If so, then we learn just how high our expectations for BCRs are. No one expects a conventional bar chart to “do” anything, and no one complains. Why, then, should we feel shortchanged when a BCR only scampers along one axis?

We’re asking a lot of these BCRs. Sounds like high-end entertainment to me.

A native New Yorker living in London, Abbott Katz is an Excel instructor and spreadsheet designer who has written two books on the application (Apress). His work with Excel has been cited by the Wall Street Journal, the Guardian and the Freakonomics website. He can be reached at abbott@abbottkatz.com.

See more stories here.