Nassim Taleb says, “When someone says he’s busy, he means that he’s incompetent.” I think he also distrusts anyone wearing a tie. In college, I wrote an essay called “The Scientific _______” in which I argued that any writer who uses the term scientific without explaining what it means is incompetent and you should stop reading immediately.
I still believe that. Now, for the first time, I am going to update my list of incompetence giveaways: Plotting something on a raw scale that should be on a log scale. Size-versus-time data should usually have the size axis on a log scale.
This presentation by someone at Sequoia Capital, the Silicon Valley venture capital firm, is full of examples. The Dow Jones Industrial Average (from the 1960s to now) is on a raw scale (where the distance from 5 to 10 equals the distance from 10 to 15), should be on a log scale (where the distance from 5 to 10 equals the distance from 10 to 20). Same for an index of housing prices. Same for the Nikkei. Many other examples. You can still believe the data, of course; just don’t trust what’s concluded from the data. Given the ubiquity of this practice (plotting on a raw scale what should be on a log scale), especially among financial supposed-experts, Taleb and I are not far apart.
More Taleb makes a similar point in his online notebook. Writing about a debate with Charles Murray:
Finally I showed a graph of the rise of the US stock market since 1900, on a regular (non-log) plot. Without logarithmic scaling we see a huge move in the period after1982 —the bulk of the variation comes from that segment, which dwarfs the previous rises. It resembles Murray’s graph about the timeline of the quantitative contributions of civilization, which exhibits a marked jump in 1500. Geometric (i.e. multiplicative) growth overestimates the contribution of the ending portion of a graph.
Umm, can you explain why size-versus-time data should have the size axis on a log scale? Thanks.
The simple answer is: Because doubling in size in a year is equally impressive whether the absolute change (end point minus starting point) is a small number or a large number. The more complicated answer is: Because plotting size on a raw scale means that major changes in the rate of growth at smale scales (e.g., between 50 and 500) will be invisible if some of the growth takes place at large scales (e.g., between 5000 and 50000). Interesting structure, if present, will be invisible.
Seth — absolutely fantastic points. This is especially great as most people are unconsciously cowed by the authoritativeness of the “experts”, in this case, VCs.
thanks, Patrik
Shouldn’t log scales be reserved for series with geometric growth?
In finance this is usually the case, but there are lots of places where it would mislead, like anything with S-curve behavior that hasn’t reached the asymptote.
Andrew, maybe it helps that you understand the value of log scales.
Allow me to add to your list of incompetence giveaways:
Ordering (ostensibly) meaningful multidimensional data in alphabetical order*, especially in charts and graphs.
This makes me insane! All of the major media do it all the time. Once I noticed that The Economist even does this, I lost a lot of respect for it.
https://www.economist.com/markets/bigmac/displaystory.cfm?story_id=11793125
Look at the Big Mac Index, it would make a helluva lot more sense if it was ordered (largest to smallest or smallest to largest) by either the 3rd, 4th or 6th column items.
*Yes, alphabetical order makes sense for, say, a guest list or any other sort of one-dimensional list.
Patrik, I agree. I’ve heard Andrew Gelman make this point several times.
OneEyedMan, “reserved for series with geometric growth” — not at all. There are lots of other cases where log scaling is an improvement, such as power-law-like histograms. The power-law (Pareto distribution) similarity is nearly invisible unless you use log scales for both the x and y axes.
In fairness to The Economist, they produce countless charts of cross-national statistics. Keeping countries in the same order promotes a kind of global coherence – the presentation matches users’ expectations. Besides, the typical user won’t read the chart thoroughly; they`ll scan for the one or two countries they`re interested in.
On the other hand, I`d say there`s a case for them organizing countries by region rather than putting them in one alphabetical list.
Tie-wearing? It really depends on your work environment. The underlying factors here is (IMO) dress as an indicator of ability to read social cues.
Keeping countries in the same order promotes a kind of global coherence – the presentation matches users’ expectations.
@Chris
I think what you are saying is that since The Economist thinks its readers are lazy, they should be lazy too.
You could say the same about not using log scales, many readers don’t expect to see them, so why use them?
The average reader of the The Economist has an IQ that is more than enough to process such data presented in a meaningful manner. By organizing multidimensional data in such a lazy manner, that is, alphabetically, The Economist casts away immediate and intuitive meta data that surely its high-income, highly-educated readers value. This is not People Magazine we are talking about here.
Besides, the typical user won’t read the chart thoroughly; they`ll scan for the one or two countries they`re interested in.
I don’t think so. I think, at the very least, people will want to see their home country and then the greatest and least in the same category. And then, who they are comparable with. Alphabetization does not help them at all.
My stats professor always joked/stressed, presenting data alphabetically is criminally irresponsible. And after a few thousand corporate PPT presentations, I agree.