There is a monthly “Skeptic Salon” run by the Chicago Skeptics, aka a book club. This month’s selection is How To Lie With Statistics by Darrell Huff, 142 pages, written in 1954, copyright renewed in 1982. What follows is a short summary.
He talks about the difference between mean (average) and median. He talks about sample response (Yale salaries: Perhaps those who were poor did not respond, thus lifting the reported average to be greater than the reported average), is the sample big enough, changing the axis of charts (starting at 90 instead of 0 for changes between 95 and 100 can make the changes look much larger than they are) to mislead (intentionally or otherwise), using figures to distort: if b is two times as big as A, then the image for B should only be twice as tall or twice as wide. If you make it twice as tall AND twice as wide, you are making B appear about 4 to 6 times larger, and the post hoc ergo proctor hoc fallacy.
The last chapter he gives hints on how to spot bad or misleading statistics. It is almost a small skeptic primer.
- Who Says So: look for bias from whoever gives you the stat: conscious bias, unconscious bias
- How Does He Know: A survey was sent to a large number of companies, but only 14% responded. The survey was trying to determine if the firms were price gouging. A sample could be biased. It could be too small.
- Did Somebody Change the Subject: Learn to distinguish between the raw figure and the conclusion. More cases of a disease does not mean more people are getting it. It may simply have been misdiagnosed in the past, or people had it but died of something else
- Does It Make Sense: Social Security makes no sense. It is set up to give benefits when people reach the age of 65, but the average life expectency (at the time) was only 63. Nobody will live long enough to get the benefits. Also: Trends will not continue forever as they have in the past: TV ownership was increasing 10,000% from 1947 to 1952. That cannot continue forever.
Image from Wikimedia