Using social media for personality profiling
We can accept that someone’s personality can give a fair indication of how good a fit they might be for a certain job: a complete introvert is unlikely to thrive as a radio presenter, for example. The controversy starts when we ask how these judgements are made. Personality tests are one thing, but what if companies were building a profile of us using metrics other than those we gave them in a personality test? What if they looked into our online activity as well? With us now is David Stillwell, Professor of Computational Social Science at the University of Cambridge, who is looking into this very possibility.
David - Personality tests have their problems. We've been looking at alternatives. Some companies do automated video interviewing, for example. This is where a computer does an interview with you, and then an algorithm tries to measure the quality of your answers. Other companies do gamified assessments, so you play games and then they use that to try to assess things. And what I've been looking into is using social media data. So you probably know when you apply for a job, quite often someone in HR will search you up on Google and see what information they can find. And there's actually data. From Ghent University, they found that those who have an attractive profile picture get 38% more job interviews than those with a less attractive profile picture. So that demonstrates the biases of humans again. So what me and my team looked into is, instead of asking all these questions on a personality test, or instead of a human looking at your social media data, maybe an algorithm can look at your social media data and try to assess your personality. So instead of saying, do you like going to parties, we just look at the data. How many parties do you actually go to? Do you talk a lot on social media?
James - I can see the value, but how does this all square with people's rights of privacy if you're snooping around their social media profiles?
David - Some people might say it's just public data, so you should be able to go ahead and use it. I don't agree with that. I think companies should ask for permission before doing this kind of thing. When you apply for a job, they should tell you the kind of information they're going to look at. The other thing that a company should do is they should share what they learned or concluded from their analysis when you ask. Under GDPR, you've got the right to get data about you and companies should share. I think what really matters is in what context it's being used. For example, SAP, the massive German multinational, they came to us, they were redoing their recruitment and they said, "well, maybe we can use social media data." And we came to the conclusion that people wouldn't like it if you use social media data to decide if they get a job. What we created instead was a job recommendation app. So you shared your data, it predicted your personality, and then it said, "well, here's a role for someone like you inside this big company, SAP." And that's much lower stakes - people still have control over what jobs they can apply for.
James - Do you have evidence for this being more effective than a big five personality test?
David - In terms of reliability, using big data is definitely less reliable than using a test which is made to measure personality because the data is more messy. On the other hand, this kind of technique has the advantage that it's based on real behaviour, so it's what people are actually doing rather than what they say they do on the test and therefore we found that it predicts future behaviour better.
James - We've been suggesting how these sorts of techniques might actually be removing bias from the recruitment process. But then there must be examples of people who perhaps these algorithms aren't designed for. To account for people with a disability, for example, where the model that these systems are working to means that they get caught in the net.
David - Yeah, and as you said at the top of this, professional personality tests are quite expensive and it's easy to slap it together, a few questions. But the reason why they're expensive is, to really create a good test, you've got to collect a lot of data and a lot of evidence around it. And part of that is the test publisher should get data on what groups it works with, what groups does it not work with, they should also provide advice to the people they're selling the test to. To say, these are the accommodations you need to use when you're using this test. Maybe it's give people more time or just, it doesn't apply to this group, you need to use some other method.
James - And how reliably do personality traits built up through this data translate into job performance? Is there a strong correlation there?
David - The answer is we don't really have evidence yet as far as I know. So when we predicted future behaviour, we found it was at least as good as a traditional test. But I'm not aware of evidence on job behaviour. So some startups are starting to provide this kind of technology, but I'd say we're more proof of concept stage.
James - And David, do you generally feel quite positive about the future of this technology? It strikes me as a bit of a can of worms we're opening up here. The science is the science, but if this were to fall into the wrong hands, could this be used to do more damage than good?
David - Yeah. There are measures in place to stop this bad behaviour when it comes to classic personality testing. In order to administer a personality test, you have to get a certificate of occupational test use. The British Psychological Society, it does reviews of tests, and you can read those reviews and see how good the test is. We are relying on those professionals to use it in a positive way. But what we always have to bear in mind is what's the alternative and what's being done right now. I mentioned, attractive people get more job offers, or more interview offers. I just think we can do so much better. It's not about getting perfection, it's about getting better than what we've got now.