Bad graph:
As is often the case in these situations, the good graph takes up less space, is easier to understand, and is easier to construct.
P.S. I think the “good graph” could be made even better by labeling the y-axis using round numbers. I don’t think the exact numbers as displayed there are so helpful. Also, I’d convert to a more recognizable scale. Instead of liters per year, perhaps ounces per day, or the equivalent of glasses of wine per week?
Excellent example! I also concur on your idea on using recognizable units. However, using ounces per day would really be recognizable only to Americans – Britons no longer use the imperial system – so the appropriate unit would be decilitres.
Hi –
It's only a bad graph if you aren't trying to show the absolute changes over time from a set period in time. Then it's a good graph. :-)
Of course, your graph is better to show the relative changes over time. Given that the level of alcohol consumption is so completely different between countries, the first does a bad job of showing what the levels were and have been.
So it's a bad graph since it doesn't show what the author was probably trying to show. But the chart type isn't per se, bad: I use them all the time to show the differences, say, between industries (setting index = 100 at the beginning of the chart), but given that these kinds of numbers are usually indices to begin with, that's not so much a problem…
Great blog: been reading for quite a while…
John
Am I reading it right that the liquid being measured is 100% alcohol? So they're taking beer/wine/liquor sales and controlling for the alcohol content of the beverages and assuming that all beverages bought in the time period are consumed in that same period?
If that's true, I doubt the actual unit values (litres or oz) are very informative: I really have no idea what a 8.4 litres of 100% alcohol will do to a person over a year. Percent change is probably more immediately meaningful to me, even if the reference point is arbitrary.
Plus, percent change retains the sexy "Blotto Brits" thesis by keeping the Brits on top of the chart. Yous buries them under their (supposedly) less blotto comparison countries.
But your graph sure belies their story. France is off the bottle? Now that's a lede!
Has anyone noticed that the two graphs use different figures? The top one has 8.4 litres for the UK in 2003, whereas the bottom has 11.2 litres in 2000. I suspect all the other figures are different too . . .
Different sources of data. The trends are the same. The Economist tends to suffer from a 'if the graph doesn't show the country we are talking about as the extreme, we can't use it' syndrome. Hence forced, artificial graphs like this one.
Different sources indeed. For both graphs the Italian figures appear more-or-less the same. In both graphs the trend and slope for UK and US are the same, but in "Good Graph" the actuals appear some 25% higher. The French figure for 1985 in "Good Graph" is some 15% higher than "Blotto", but for 2001 the figure is some 33% higher: the "Good Graph" absolute for 2001 is above the "Blotto" figure for 1985.
Hi all, the two graphs indeed use different data sources as I indicated in my writeup. I couldn't replicate the Economist data because that data is being sold commercially in a report. But even though the absolute levels are not compatible, the trends are no different in the two sets of data. My data came from an OECD report.
A bigger point I was trying to make was that indexing is not always a good idea. In this case, it obscures the fact that France is still higher than other countries on the graph.
Another point that I neglected to mention, which is shown well in the "good graph" is that the spread has significantly narrowed over time.
What's wrong with the top graph? The two plots illustrate different information. The `bad' graph shows relative change and the `good' graph absolute change. The two are not quite comparable as the `good' graph takes a 30+ year span and the `bad' graph a 15+ year span. In the `good' graph, the relative change from 1985 to 2000 is not easily appreciated. It probably depends on what information you are trying to get accross, changes in trends or absolute comsumption.
Eric,
In general I agree that there is no one perfect
graph, and that different graphs present different information.
But in this case, I think the second graph dominates the first graph. I see no advantage to seeing 1985-2000 without the perspective of the first fifteen years, and I see no advantage to losing the absolute level for each country. I would fix the second graph by labeling the axes using round numbers (and clearly going down to zero), but that's it.