Monday, February 20, 2017

How to tell fake news from real news

Graph department that is.  The Internet is awash in graphs claiming to show the growth, or shrinkage of all sorts of things.  Net Worth, employment, GNP growth, income inequality, and on and on.  The graphs typically show a bunch of colored lines, rising dramatically, and implying that something is getting bigger, or better, or worse, or something.
   Lies, damn lies, and statistics.  How can you tell a real and true graph from fake data trying to convince you of something that isn't true?
   These tricks may not work all the time, but they will weed out a lot of fake data graphs.
1.  Are both axes labeled?  With what they represent and what the units are (gallons, pounds, feet, furlongs per fortnight, dollars, whatever).
2.  Are the scale divisions of the axes uniform?  A graph with scale divisions every 10% except for a few on the end scaled out to 2%,1%, and 0.1% is attempting to bend the plotted curve somewhere.  Any graph with non-uniform scale divisions is trying to lie to you.
3.   Does the vertical axis go all the way down to zero?  I can take a straight line and turn it into a jagged mountain range if I expand the vertical scale enough.  If the vertical scale doesn't go down to zero, the graph is trying to make bumpiness bigger than it really is.
4.  If its a graph of something versus time, does the time axis go back before 2007?   Great Depression 2.0 started in 2007 and just about everything went down the drain that year.  A graph that starts in 2009 will show a steady increase as we pulled out of Great Depression 2.0  Same graph restarted in 1997 will likely show a great dropoff in 2007 and may show that things have not recovered to where they were in 2006.  Two different messages.  
5. Do the numbers at the extremes of t he graph make sense?   For instance I saw a graph claiming that of the top 0.1% income individuals in the country, 40% of them had not completed college and were out of work.  Somehow I just don't believe that. When I find one unbelievable data point, then I figure there are more that I don't find.  Put that graph into the damn lies category.

    I'm picky.  If a graph fails any one of these tests, I put in into the "damn lies" category.

No comments: