Monday, September 17, 2012

Chartology

Assistant Village Idiot had a recent post reminding me of one of my pet peeves - truncating either axis on a chart like that below giving a visually misleading impression.  This is what AVI has to say:

"One of my favorite deceptions to point out WRT graphs. As we are talking about millions of people and long-term trends, this drop in participation may indeed be alarming. It is a four-point drop in how many of us are gainfully employed, after all. But the zero-point of this graph is 63.5, and the hundred-point is 67.5, exactly the size of the change, not 0-100 percent. Visually, it tells us that nearly everyone worked twelve years ago and now, no one does.

Naughty, naughty."

The point isn't that this trend isn't alarming.  This is a labor participation rate that is less than we've had in more than half a century.  But you can play too many games when you truncate like this.

It's not just in the charts themselves that this tactic is employed.  I remember an article in the New York Times several years ago (in a recent search I've been unable to find it on the Interwebs!) about climate change which has stayed with me as a great example of the misuse of a time sequence of data.

One of the facts cited in the article to support a claim about the reality of global warming was that Alaska had seen a 6 degree F increase in temperature in the 30 years since 1975.  This is an accurate statement.   However, because of my interest in the subject I was familiar with the underlying data (unlike most of his readership) and knew this was nonetheless misleading.  What wasn't said in the article was that this entire change happened between 1975 and 1980 and that there had been no increase in temperature in Alaska since that time.

This is the actual data from the Alaska Climate Research Center at the Geophysical Institute, University of Alaska Fairbanks.  The chart is included in a paper interestingly enough entitled "The First Decade Of The New Century:  A Cooling Trend For Most Of Alaska.
So what's going on here?  What caused the huge change in the mid to late 1970s?  According to the Alaska Climate Research Center it was a change in the Pacific Decadal Oscillation (PDO), an ocean-current phenomenon which alternately brings warmer or colder water into the North Pacific with a big impact on Alaska's climate.  The PDO changes every 30 to 40 years.  By the way, according to the Center's website the slight cooling trend noted in the chart above has continued in 2010 and 2011 (As an aside, I've seen similar date-truncated charts used in the opposite way to "prove" there is no warming).  Once you start looking around you can find examples to this approach to data everywhere.  In fact, think how much fun you could have with the data on this chart if you could pick your start and finish dates!

My guess is that the Times writer never actually saw this data and did not intend to be purposefully misleading.  As a long-time Times reader (before I finally gave up on it a few years ago) it's clear most of its reporters are uncomfortable with numbers and statistics and the reporter was probably just fed some summary information by an advocacy group during his search for examples to support the thrust of his article.  He fell into the trap of thinking that data is like an ornament to hang on a wall, a decoration to make the house look nice; not something that deserves thought and examination in and of itself.

All of us are prone to grab onto and trumpet data that supports something we already believe in.  That's when you need to be very careful and ask what's the scale being used; why were the specific start and end dates selected; is it statistically significant?  Sometimes things ARE too good to be true.

A good new blog devoted to these issues is Bad, Data, Bad.

1 comment:

  1. Good point, now I'll take a closer look...no longer a graph trusting soul! dm

    ReplyDelete