How to lie with statistics by Darrell Huff

¶ … Lie With Statistics

Huff, Darrell. How to lie with statistics. New York: W.W. Norton & Co., 1993.

'There is terror in numbers.' Darrell Huff was not a statistician. However, he wrote his 1954 classic How to lie with statistics to help his math-intimidated readership better "look a phony statistic in the eye and face it down; and no less important, how to recognize sound and usable data in [the] wilderness of fraud" (Huff 124). Over the course of his work he attempted to expose the deliberate lies and faulty reasoning exhibited in the advertising and partisan political writing that the average American was exposed to on a daily basis.

Obvious examples of lying with statistics include graphical misrepresentations, like drawing a very large item next to a small item, even though the actual difference in the amounts are statistically insignificant in mathematical terms. Graphs can be used to make an emotional impact upon the reader, rather than to convey useful information. A big graph with a dramatic tilt upward because of carefully spaced out numbers looks much more impressive than a graph with closely spaced numbers and a moderate incline. Positioning irrelevant statistics attached to graphs or pictures can make the statistics seem significant even though they are not. The statistics given a 'feel' of logic, but really, the reader is reacting emotionally and viscerally, rather than logically.

Huff also details logical fallacies such as equating correlation with causation. A good example of this is the claim that because people who have a cold take antihistamine pills, a cold can be cured with antihistamine pills (Huff 10). While many individuals who suffer a cold may take the remedy and eventually their cold goes away, the cause of the cold going away is the natural duration of the illness, not taking the pills. It is simply that someone who has a cold is more likely to take relief pills because of the other symptoms associated with the stuffy nose of the cold. The pills temporarily alleviate the symptoms but do not cure the cold.

Samples with built-in biases are another good example of faulty reasoning. For example, many alumni publications cheerfully report how well their graduates are doing. Of course, this sampling bias is skewed given that individuals with more pleasant college experiences are likely to send back school-sponsored surveys. Also, these graduates are more likely to want to brag about their accomplishments on a survey if their experiences have been pleasant ones (Huff 16). Marketers can carefully cherry-pick the populations they survey when selecting populations to create the statistics they will use in their advertising. Huff points out that a magazine knows its target population and can survey members of this select group who say they prefer it to another publication, versus a grouping more representative of all magazine readers as a whole.

Perhaps the easiest way to distort survey results is simply to keep taking surveys of populations until the desired result is reached. The reason this result is produced is chance, however, rather than scientifically legitimate findings. The 'well-chosen' average is another example of this, whereby the survey sample is carefully selected to yield a figure desirable to 'prove' the contention of the reporter. Including or not including persons who would distort the average is another statistical lie. The presenter can also select the specific statistic that bests proves his thesis -- the mean, median or mode (the mean is the sum divided by the number of values, the mode is the number that occurs most often in the sample, the median is the 'middle' sampling of all listed numbers). For example, finding the 'average' American salary can produce wildly different results, given the discrepancies that can result between the median and the mean because of the high salaries of the small numbers of persons at the top, and other people who make very low salaries.

Words are powerful: calling something 'flimsy and cheap' sounds much worse than calling something 'light and economical,' and even the words 'practicing celibacy' can sound ominous, because of the association of the word 'practicing' with something nefarious (Huff 102-103). The language with which statistics are presented can also cause an unwitting reader to believe in them: for example, saying 'it is obvious that the pollution is killing all of the birds, because 100% of persons surveyed said they have not seen a single bird flying this year." (The persons may not have been paying attention, for example, to the birds). More seriously, Huff gives the example of a manager who wants to construct an anti-union survey. The manager collects any and all of the complaints that have arisen about the union, and uses these complaints to 'prove' that no one wants the union on the premises. However, it is very difficult to find an entity with no complaints about it at all, so the conclusion that is arrived at is fundamentally self-serving and misguided because the survey population did not say that it disliked the union (Huff 82).