This chapter of Invisible Gorilla discusses what we might call “jumping to conclusions”. A jump might lead to an incorrect assessment of a root cause, an ineffective remedy, or the adoption of an ineffective process The authors provide three sources of error.
1) we perceive patterns in randomness,(that is the seeming patterns results from chance). See also Nassim Nicholas Taleb’s Fooled by Randomness
2) events that happen together are perceived to have a causal relationship,
3)in a chronological sequence of events, we assume that something preceding another is the cause (post hoc ergo promter hoc. at least when the connection is plausible.
These fit my own observations that people will “look for a cause”, latch onto something plausible, then cling to their belief. People remember stories, but consider statistics to be boring or untrustworthy.
Our challenge is to untangle cause-effect from coincidence. This is harder than it appears even when bring statistical methods to the fight, see the Ronald Fisher tobacco controversy in The Lady Tasting Tea for a more complete discussion.
The best approach to untangling cause is to conduct a controlled experiment. Best is a double blind randomized study, next best is randomized without blinding. The randomization will likely help you isolate a single factor. If you cannot randomize, large scale observational studies are next best. Scientifically, one needs to establish statistical significance (unlikely to result from chance) and practical significance (a economically important effect size).
One errs by using a single case and generalizing, or carefully selecting the cases for study. The book cites one of my favorite examples of the latter, Good to Great that focuses on what successful companies did, but never examined companies that did the same thing failed (a typical example of being “fooled by randomness” or “survivorship bias”)
In the process world, experiments are expensive and hard to generalize to the practical industrial world. Extensive observations are difficult and expensive. Moreover, there are countless factors obscuring the observations. The typical approach of pilot, roll out, leaves much to be desired from a scientific perspective. We are on firmest ground when process changes have a substantial effect. We also need reliable measurement. For example, inspections of code, is the “lab rat” of software engineering, because the effects are substantial (capturing 50-90% of defects) and it is easy to measure.
In some respects, the evidence for this chapter is weak. The narrative presented focuses on the MMR vaccine/autism controversy. The authors make valid points, but neglect general distrust of big pharm, the substantial increase in child immunizations at that time, and the plausible cause of high cumulative doses of mercury (used as a preservative).
Skepticism of medical science, where there is often a strong vested interest, has some foundation, see “Lies, Damnded Lies, and Medical Science” . This inability to trust the studies is also a major problem in the process community.
The Lady Tasting Tea: How Statistics Revolutionized Science in the Twentieth Century, David Salsburg, W. H. Freeman, ISBN-13: 978-0716741060
Invisible Gorilla, http://books.google.com/books?id=f8AN1DAud5sC&printsec=frontcover&dq=invisible+gorilla&hl=en&ei=qsy9TObNLoL-8Abo_Mn9Bg&sa=X&oi=book_result&ct=result&resnum=1&ved=0CCsQ6AEwAA#v=onepage&q&f=false
Fooled by Randomness: The Hidden Role of Chance in Life and in the Markets, Nassim Nicholas Taleb, New York, Random House 2005. ISBN: 1-58799-071-7
Lies, Damned Lies, and Medical Science, http://www.theatlantic.com/magazine/print/2010/11/lies-damned-lies-and-medical-science/8269