Forgive me the title, won’t you? It is my own creation, I admit, but numbered promises are eminently clickable. Plus they are the norm in business writing.
Anyway, the real post of the same name is at Quirks, the trade journal of marketers. In it, I outline the most common blunders users of statistics make, at least when it comes to marketing.
I won’t repeat any of the details, but here are the main points:
1. Asking too many questions Questionnaire fatigue is rarely considered but results in beleaguered respondents clicking the same answer for all queries just to get the thing over. Also, Big Data won’t save us all, though it will have successes, which will be just as transient as all sociological findings.
2. Failing to appreciate limitations Many feel that the answer can always be had if only sufficient sweat were expended or enough data collected. If the future were certain, then we’d all be climatologists.
3. Not understanding regression If you read only one point, make it this one. I am always amazed at how many people who routinely use statistical models have no idea of their purpose. We statisticians are probably responsible for this state of affairs because of the way we cherish parameters. Parameters become the reason for models, even though they are entirely invisible, metaphysical creatures.
4. Falling for the latest gee-whiz approach “Can you make my data Big Data?” Yes, actually asked. But more usually, “I just read about technique X. We need to use it on our data.” Even though technique X is no relevance to the question at hand, it sure does sound sexy. As I say, the best analysis often is no analysis at all: simple counts, tables, and pictures give a good feel of the situation and are less likely to lead to over-certainty.
5. Not coming to a statistician (soon enough) A lot of folks have their statistical training from well-meaning, kind, morally upright, entirely praiseworthy people who themselves are not statisticians and who have in turn learned their stats from people like themselves, and so on. Fine for dabbling, but not a course likely to lead to an appreciation of pitfalls and subtleties. The situation is like learning quantum mechanics from garage mechanics because both subjects study movement: it can work—there are lots of capable garage mechanics—but perhaps there is a better way.