Malcom Gladwell has a tremendous ability to detect trends and patterns in places where most people see chaos. He dispassionately analyzes his chosen subjects deeply, and presents insights which are often challenging but unimpeachable. But something must have rattled him during his recent investigation of personality tests, because he turned out anuncharacteristically flimsy and defensivepiece of rhetoric more suited to grade school assignment or a hastily-crafted blog post.
His basic argument is that personality tests are not useful, because they cannot predict certain specific actions that specific individuals might take, as in the example of a mild-mannered soldier who acts with great valor in the heat of battle. But this is setting up a strawman. It is absurd to think that profiling is designed to predict at the level of distinct individuals, or at the level of specific actions like ?valor under fire?. And it is absurd to assert that profiling which does not achieve such a level of detail is somehow deficient. This is like saying that heart medication is useless, since many people who do not take itnevertheless survive, and people who do take it sometimes die anyway.
In fact, Jung said as much, in a quote that Gladwell cites, but Gladwell apparently fails to make the connection. Gladwell cites very little actual data in his analysis; it’s almost entirely rhetoric. About 50% strawman and 50% personal. He claims that MBTI is black/white, when personality is gradient (he is either ignorant or deceptive, since MBTI uses a graded scale on all 4 dimensions). He tries to argue that, ?some friends and I invented a personality test as a joke, and ours is just as good as anyone’s?, then he argues that, ?Meyers-Briggs was invented by busybody housewives with no formal education, anyway?.
But soon we find out what’s really bothering him. We get the impression that he didn’t like his own test results very much. For people who read his writings, his test results as he relates them to us seem eerily accurate. Incredibly, the one person who doesn’t see the test results as being obvious is himself. He spends much of the column trying to propose alternate theories about his test results, rationalizing that the psychologist would need to do many more refined tests to get a more ?accurate? reading, results only apply in a business situation, and so on. He eventually succeeds in convincing himself that his personality profile is only slightly accurate, open to broad interpretation, and deeply dependent on context. It’s a gripping example of denial and self-delusion that plays out before your very eyes.
To be fair, personality tests are not magic. Some are good for certain things, and none are all that good at predicting individual behavior with much certainty. But they can be incredibly useful, and reveal some totally unexpected correlations. The key is to look at them as a statistical tool like any other: ?people who score highly on these areas of MMPI are 40% more likely to get into physical confrontation than average?, or ?people which this specific MBTI type are 80% more likely than the general population to be republicans?. One could argue that a ?personality test? could just ask ?Are you democrat or republican?, since that test would be an even better predictor of political affiliation, but this would be to fall into Gladwell’s original strawman test. As a predictor for any specific standalone attribute, the indirection if a personality test clearly yields poorerpredictive power. If you want to know how a particular individual is going to act, your best bet is to just watch him. But the indirection is also the power. If you know that a certain MBTI type is more likely to support your opponent; and you also happen to know that this MBTI type is more responsive to a certain style of advertisement, you can tweak your political advertising to take advantage of this. In a sense, it’s no different than any clustering/decision-tree algorithm used in data mining. In the end, the labels put on the categories don’t matter, and the attributes used to partition the sample into clusters need not be intuitive, as long as the categories are useful predictors. And in fact, some of the most useful mining models used today are built on categories which are purely ?discovered? from the actual data and have no relation to actual conceptual categories that humans use on a daily basis to describe the same data.