Choosing reference levels

Reader Jordan G. submitted this infographics (via Business Insider) to Junk Charts, my other blog about graphical presentation of data. Graphically, the chart has nothing to commend itself but the most annoying failure is the awful choice of statistics.

Stating the Trivial

Regarding Pandora, the chart tells us 18.7 million hours of streamed music per day (across many millions of computers) is equal to one computer streaming music for more than 2011 years.

What the chart is really saying is that there are more than 18.7 million hours in 2011 years. (If you do the math, there are 17.6 million hours in 2011 years.) This is true whether or not anyone streams music on Pandora.

Making Non-sensical Comparisons

Regarding Netflix, the chart tells us 22 million hours of old TV shows and movies on Netflix in one day is equal to the number of hours of movies watched in theatres in 3 days.

It's unclear why TV viewing is counted on the one hand but not on the other. Besides, movies on Netflix and movies in theaters are more or less disjoint sets despite both being called movie viewing. Someone can create a Venn diagram similar to this user's amusing take on Netflix selection (link):

***

Picking inappropriate reference levels is an endemic problem. I cringe every time I hear how many people are dying every day from hunger (or substitute any disease).

So if disease A causes 36500 deaths a year and disease B causes 18250 deaths a year (half as much), disease A causes 100 deaths per day while B causes 50 deaths per day... it's still twice the number of deaths. The only new information added by changing the unit of measurement is the number of days in a year.

A statistic like the proportion of deaths among 15-35 due to hunger would be much more telling.

Comments

[Heather]

I work every day to improve my ability to communicate statistical ideas to policy makers and other generally non-technical people. In my experience, a proportion like the one you are suggesting above is not as clearly understandable to many people as X deaths per day. And that changing the unit of measurement from days to years makes it more relatable to the average funder or policy maker. I am confused why you think the proportion is much more telling. Could you elaborate? I would like to learn. Thanks.

[Nayel]

I agree with Heather. In my own work with seeking funding or volunteer support, I have found the single, annualized statistic to be much more persuasive than comparative statistics (X kids in the age-bracket of 0-5 getting polio every year). Even though the figure is the same, I think it dilutes the focus by allowing the listener's mind to think that because they haven't been given the "whole" picture (and the inverted commas are sarcastic, mind you), maybe, just maybe, the problem isn't so bad. Gives them a reason to think a little less of the statistic.

[Kaiser]

Let me explain this in a different way. If I'm told 10,000 people per day die of disease A, this statistic has no meaning to me unless I have a reference point. This reference point might be that 8,000 die per day from disease B, which is comparable in some way to disease A. But what this pair of death rates tell me is really that disease A causes 25% more deaths than disease B. The per-day part of the statistic cancels out and doesn't matter a bit.

A different reference point is the total number of deaths per day among the population being referenced. If 15,000 people die per day, then disease A is a killer disease. If 150,000 people die per day, it is probably not an important cause of death. Again, the per-day part of the statistic adds nothing to the conversation.