There’s little doubt Cambridge Analytica’s behavior was scandalous, but was it a scam?
If any of what is alleged is true, the firm and its leadership behaved unethically, and possibly illegally.
Of course, neither they, nor I, are privy to the inner workings of the company, so we can’t say definitely whether or not they performed the work they claimed, nor whether or not the work itself meets any particular standard of professionalism.
Most of the criticism therefore has been directed at the theory and empirics behind the work Cambridge purports to do.
And that resolves to two basic questions:
1. Can information about an individual’s personality be derived from their online behaviors?
2. Is that information politically useful?
Personality itself is a fraught concept in psychology. Some scholars essentially deny the existence of the kind of stable traits we identify with the term personality.
But many psychologists adhere to what is called the Five Factor Model, which suggests five core traits are the building blocks of personality.
The word model rears its head in the paragraph above and throughout this and other discussions of targeting. So, it’s useful to keep in mind the admonition of distinguished statistician George Box, “All models are wrong, but some are useful.”
So it is with the Five Factor Model: perfect, no; useful for some purposes, yes.
And so it is with the tests that measure the traits in the model: far from perfect, but useful in classifying people.
Cambridge allegedly employed these tests with some Facebook users and then modeled the relationship between individuals’ digital footprints (e.g. Facebook likes, language employed in posts, etc.) and these personality traits.
The model is far from perfect in assigning people to the “correct” trait buckets, but academic research suggests a good model (which Cambridge’s may or may not be) can make more accurate personality assessments than friends, and even family, with only spouses coming close to matching the algorithms accuracy.
Of course, neither friends and family, nor the model are wholly correct, but they seem to provide useful information and the data it provides is more useful than the assessments of acquaintances.
So, to some meaningful extent, some aspects of personality can be inferred from properly constructed models based on digital footprints.
Is that information useful in a political context? Here some of the commentary has been ill-informed on two levels. First, what constitutes “useful” has to be subject to some criteria, and, second, one has to ask useful for what purpose.
A critic from Ted Cruz’s presidential campaign complained that almost half the people Cambridge said were likely to be Cruz supporters proved not to be when they were called. Dismal.
But since Cruz was rarely above 25 percent in any poll, the fact that Cambridge increased the hit-rate by 100 percent renders the model useful, at least for some purposes. If you are expending resources making phone calls and you can cut your wasted effort in half, you have a model that’s wrong, but useful.
Which begs the questions, useful for what?
Academics have argued that modeling issue positions doesn’t work. See my previous point—doesn’t work compared to what?
Others argue that personality more broadly is unrelated to politics.
But being able to predict political attitudes is far from the only use of psychologically-based models.
The intellectual progenitors of Cambridge Analytica found that advertisements designed for different psychological types generated 40-50 percent more purchases than psychologically mismatched ads.
Buying make-up, the subject of these ads, is not the same as choosing a candidate. This may not make a whit of difference for political ads.
But there is at least some possibility that psychographically appropriate advertising could be more effective. That has nothing to do with being able to model political choices or issue positions. But if it helps create more effective advertising, it’s useful.
I’m not defending Cambridge Analytica in any way.
I’m suggesting that we continue thinking about, and researching, ethically appropriate methods of improving our craft.
Mellman is president of The Mellman Group and has helped elect 30 U.S. Senators, 12 Governors and dozens of House Members. Mellman served as pollster to Senate Democratic Leaders for over 20 years and as President of the American Association of Political Consultants.