Digital Psychometrics and its Discontents: Online Behavior for Psychological Analytics
Occasionally we hear about predictive data modeling aimed at gaining psychological insight that at first blush seems – even to tech sophisticates who pooh-pooh AI alarmism – like a primitive version of something unsettling to come, something too penetrating for comfort, something with the potential to compromise our privacy, even our sense of independence.
This week in Miami at the Linux Foundation’s Apache: Big Data conference, a researcher from Cambridge University discussed project work that builds insights into people based on their social media profiles that, combined with data collected from their smart phones, has the ability to know you better than your friends do, better than your siblings, spouse or partner – maybe better than you know yourself. As for the persona you try to present to the outside world, it can pierce that with ease.
This raises the question: If we can do this in 2017, one can only imagine the psychometric implications AI and high performance data analytics hold for consumer advertising and political campaigns, to name two, in future decades.
To be sure, a major theme of the presentation from Sandra Matz, computational social scientist / psychometrician at Cambridge, is the potential for good offered by the field of digital psychometrics, which is the assessment of psychological characteristics using our digital footprints, at scale. She also discussed the need for transparency in psychometric project work as the best defense against worst-case urban legends, which could undermine otherwise legitimate and well-intentioned research.
But she also acknowledged the potential for abuse, exploitation and, if nothing else, lowering the bank balances of consumers targeted by more psychologically acute digital messaging. In fact, she said, Facebook ad messaging based on digital psychometrics has already proven its effectiveness with improved conversion rates and higher average revenue-per-sale results.
The writer of this article admits his vulnerability to the charge of psychometric alarmism. In fact the research presented by Matz does not involve particularly complex algorithms or advanced big data analytics techniques. But this, in part, is what’s disconcerting about the research she shared. For now, it amounts to merely scraping data from Facebook accounts, with permission from the Facebook user, and then tabulating our Likes and the words we most frequently use in messages to Friends. Taken together, our Likes and our words are then associated with psychological traits – our degree of extroversion or introversion, our “agreeableness” or lack thereof.
Still, the insights derived from these relatively primitive analytics techniques point to a future in which advanced algorithms and big data are applied to our digital footprints. Coupled with increasingly powerful predictive models and the amassing of more and richer personal data, is it going too far to say we can foresee a time – for good or for ill – when, to rework the famous Orwell quote, “Nothing was your own, (not even) the few cubic centimetres inside your skull”?
Yes, that’s probably going too far.
On the other hand, Matz openly discussed the double-sided nature of psychometrics.
“I think on the one hand this is extremely exciting that we know so much about peoples’ lives now,” she said, “but with that great power comes a great responsibility. And I really mean that, because I think for a lot of people this is extremely scary. Because they don’t know what’s happening, they don’t know how this stuff works, they don’t know what’s possible. I think that’s true for all of us in the room and in academia, we need to get a lot better at openly communicating, and transparently communicating, about what’s possible and also about the technologies that are being developed.”
Matz herself emphasized the potential for good that psychometrics offers. For example: the ability to detect teenage depression and anxiety, which in its acute forms can lead to social alienation, substance abuse and suicide.
Psychometrics techniques, she said, make it possible to predict vulnerability to depression from data passively collected by a smart phone. This is because most teenagers have their smart phones 24/7, and phone data reveals where they spend their time (at home or in after-school activities?), their level of physical activity (sedate in front of the TV/computer or as a member of a sports team?), how much sleep they get (a normal amount or too much/too little?) – most importantly, how much do they interact with others?
More isolated and less active teenagers are more prone to depression. Using the smart phone to detect depression is important, Matz explained, because one symptom of the illness is a disinclination to seek help.
Matz acknowledged an alarmist scenario, in which teenagers identified with depression via their online life are vulnerable to abuse or exploitation. In fact, Facebook has been accused of doing this kind of research – and some groups have demanded that Facebook turn over the results of the alleged work.
On the other hand, she said, “if you think about it, this is an amazing chance, Facebook doing research trying to identify people who are at risk, if that information was being used to help teenagers who probably aren’t going to go to their parents because they’re having a bad time… This is a news story that’s very interesting, the positives, the potentials, and also the dangers for abuse.”
Matz said the work of her and her project team at the Centre dates back to 2007, when a Cambridge researcher developed a Facebook app called myPersonality, a self-perception test. The app was a hit, and the university reports its database contains more than 6 million test results together with more than 4 million individual profiles.
That data was crucial to the inception about five years ago of digital psychometrics, which grew out of a desire by researchers to find new ways of gaining psychological insight into people without requiring them to complete lengthy self-perception questionnaires. They wanted a means of collecting massive amounts of personal, psychological data – that is, at scale.
The idea, she said, is to try “to get away from giving people questionnaires about self-perception, which are time consuming and limited in scale, since it’s difficult to get millions of people to answer the same self-perception questionnaire. Instead, the idea is trying to look at what people actually do out there in the real world. What web sites do they visit, what do they do with their social media profile, and what kind of things do they buy with their credit cards” – along with capturing physical activity and geolocation data from smart phones.
“It’s not very surprising that everything you do out there tells us a lot about your personal preferences,” she said, “your habits, what you do with your everyday life and how that translates into something like…psychological traits.”
A key to the evolution of digital psychometrics has been the combination of data generated from the myPersonality app along with the 40 percent of those of who took that test who also voluntarily agreed to give access to their Facebook profile data and social network data.
According to the Psychometrics Centre’s web site, “Respondents come from various age groups, backgrounds and cultures. They have been highly motivated to answer honestly and carefully, as the only gratification they received for their participation was feedback on their results. Their scores are combined with unprecedented amounts of additional information from those who opted in to sharing it with us, including detailed demographic profiles, a record of their behaviour in Facebook environment, their interests, preferences, opinions, etc.”
“So for 3.5 million users we have their self-perception and also have their Facebook profile data,” Matz said. “…This is the interesting part: basically once you have this behavior on the one hand, and psychological characteristics on the other hand, you can start building models.”
This combination enabled the Cambridge researchers to develop Magic Sauce, an online application in which people give the application access to their Facebook profiles, and Magic Sauce rapidly generates a profile that includes your personal interests, leanings and IQ. This is the Holy Grail of psychometrics: generating massive amounts of psychological data about people without requiring them to take tedious self-perception tests.
“In a few seconds it scrapes your Facebook profile, takes your likes, your stated preferences and turns it in to a psychological profile,” she said.
Matz said her own test results show her to be “liberal, artistic, a bit more introverted and a bit more laid back than the average,” with intelligence level “a bit above average.”
“It takes a few seconds, you don’t have to sit down and complete questions for a half an hour.” And there’s no charge. The app is attractive because, she said, “people really like to learn about themselves, they like to know themselves.”
Take extroversion: these are people who are energized by being with others, they tend to be optimistic, enthusiastic and group-activity oriented. We see this, among other traits, in the words they tend to use:
Magic Sauce also tests for “agreeableness,” along with its opposite. Agreeable people tend to use words that indicate warm and trusting personal relationships, they talk about their families in highly favorable terms, and they tend to be generally grateful with high life satisfaction. On the other hand, words used by “low agreeable” people tend to show up on George Carlin’s list of the “Seven Words You Can’t Say on Television” (cropped from photo at left).
Matz downplayed the sophistication of Magic Sauce. “At a basic level, this is just counting words. It’s not rocket science, there are no complex algorithms.”
Having said that, the accuracy of the Magic Sauce test, correlating generalized personality traits to behavior and statements on Facebook, is more accurate than you might think. Accuracy is measured by comparing its insights to those of self-perception questionnaires. Using Facebook Likes as the basis for psychological understanding, Matz said it takes about 10 Facebook pages, on average for Magic Sauce to understand you better than your work colleagues, 65 Facebook pages to understand you better than your friends, 120 pages to understand you better than your family, and 250 pages to exceed the understanding of your spouse or partner.
Given that the study of the test’s accuracy was conducted using an average of 230 Facebook pages, Matz concluded that “the computer is basically smarter than everybody except just a little bit worse than our partner or spouse.”
Digital psychometrics has obvious and profound implications for digital marketing. The Psychometrics Centre ran experimental ad campaigns on Facebook for an online UK-based beauty products retailer, with ads designed to appeal to various psychological traits. Facebook users known to be extroverts were targeted with ads showing groups of people having a good time; introverts received ads with a singular person in a quiet setting.
Matz said the ads were run for seven days and generated about 6 million impressions, 10,000 clicks and 400 purchases on the retailer’s website.
“Basically what we saw is when we match the message that’s used to promote a product to people’s psychological characteristics we see higher conversion rates, people purchase more often and they also generate more revenue for the company,” she said. “This is really simple, right? It’s the same products. The only thing we changed is the way we communicate based on their psychological profile.”
Matz is fully aware that digital psychometrics runs counter to most peoples’ natural inclination to guard their privacy and individuality. She also acknowledged that most of the public only hears alarmist accounts – including, perhaps, this article – about this technology. This is why openness is so important, she said. Without it, digital psychometrics will be widely looked upon as a menace.
“They hear how these technologies are going to ruin their lives, make them prone to being manipulated and basically posing a huge threat to their lives,” she said. “We need to show them the other side. Show them that yes, these technologies exist, and we’re working on them to make them better, but there is so much potential to help you make better decisions, to help you lead healthier and happier lives. And we’re not doing that behind the scenes, that we’re actually openly communicating to the public about what’s possible and what we’re currently working on. I think that should be part of the Open Source movement, and is one of the movements where we can see (that ethic) is at least going in the right direction.”