I just finished reading this 140 paperback about statistics, and I found it enlightening:
How to Lie with Statistics: Written by Darrell Huff, 1993 Edition, (Reissue) Publisher: W. W. Norton & Company [Paperback]
It was written back in 1954, but the topics it covers are timeless, and -not surprisingly- Mr Huff has few very useful tips applicable to nowadays business intelligence. You can only tell that the text is 60 years old when looking at the details: in one example there is the price of a good hotel room in San Francisco (an eye popping figure: 8$).
One of the first Chapters explains the “biased sample” problem, which affects so many scientific results; another one points the finger to an old trick used to convey distorted messages with line graphs. Many of the wrongdoings he exposes, sounds hard to believe, are still in use today, and you can find plenty of them in ads, magazines, and (sadly) in government polls. Not earlier than one month ago, one very famous cable news network was showing a chart picturing an “impressive” improvement in Spanish unemployment rate: from 26% down to 22% (ouch!). It was something like this:
What they failed to show was data from preceding years:
data source: Spain Unemployment Rate 1976-2015
One of my favorite part is when Mr Huff shows how unions or entrepreneurs give inconsistent (to say the least) results from the same set of data. Or how “experts” sum percentages in an mind bending attempt to rewrite the math rules: “buy any 20 products, wait one year, see their price increase by 5%, so overall the cost of life (20 x 5% = 100%) has doubled in one year…”.
Oh, one last point: don’t forget to read the exhilarating story about the rabbit burger…