Prof. Sinan Aral, New York University
Chris - Are you a leader or are you a follower? An intriguing new study shows how you can identify influential people from their activities on Facebook. Sinan Aral is at the New York University Stern School of Business. Heís with us now. Hello, Sinan.
Sinan - Hi.
Chris - So first of all, what were you trying to prove with this study?
Sinan - Well, in essence, finding influencers or as you said influential people is sort of all the rage today. Companies like Klout are trying to measure influence scores for people on social networks like Facebook and Twitter, but beyond marketers, managers and policy makers are more generally interested in how behaviours spread through society. In this paper, we present a general method for measuring influence susceptibility in networks and the main contribution of the method is that it avoids known biases in current methods such as homophily bias. Homophily means that we tend to make friends with people like ourselves. For example, if two friends adopt a product or a behaviour one right after the other, current methods have a hard time distinguishing whether itís because of pure influence; one friend influencing the other, or if friends simply have similar preferences and thus, behaves similarly.
Chris - And obviously, the world doesnít work like that because an influential person isn't just influencing their friends. They have the ability to influence a range of different demographics.
Sinan - Exactly, so what we did was we applied this general method to measure influence and susceptibility in the adoption of a commercial product on Facebook among 1.3 million users and we were able to recover influence and susceptibility scores that arenít subject to these known biases like homophily bias.
Chris - Okay, so can you talk us through what you actually did? How did you recruit the people and then what actually happened to discover this?
Sinan - We worked with a company that developed a commercial movie application where you can rate movies and buy movie tickets, and read about directors and actors. And as people adopted this application, we randomly assigned them to send messages to their friends in a random manner. So, every time you did something on the application like rate a movie or talk about a celebrity or something like that, it would randomly select a subset of your Facebook friends to send a message to. And this randomization removed the selection bias of people selecting friends with similar preferences or selecting people who they knew would be specifically susceptible to influence. And with this randomization, we were able to measure, for example, how your characteristics or your traits, your age, your gender, your relationship status on Facebook or anything that we could observe about you on Facebook, was correlated with your likelihood of responding positively and adopting this application upon receiving this influence mediating message. And because the messages were randomised, we could make causal inferences about whether this message was causing you to adopt or not.
Chris - What about the other way around because that's looking at people how they respond to receiving the message? What about in terms of the people who actually send the message? Are you inferring whether they're influential or not based on what the response of the recipients is?
Sinan - Exactly, so we estimated a statistical model that estimated both influence and susceptibility simultaneously while these random messages were being sent to people from their friends.
Chris - So, spill the beans then. What makes someone highly influential? Is it just that they're very well connected or is there something special? Is there some special recipe that means that if they say something on Facebook, everyoneís going to be talking about it?
Sinan - We found that itís not just how many people you're connected to and lots of people have been focused on that in the past, how many followers you have, but more importantly, itís whether you persuade your followers to change their behaviour. What we found was that in the context of this particular movie application when we applied this method, that men were more influential than women, that women influence men more than they influence other women, that older people are more influential and less susceptible to influence than younger people. Married people are the least susceptible to influence and influence and susceptibility trade off. Meaning, people who are more influential tend not to be susceptible and people who are susceptible tend not to be influential.
Chris - Isn't this just what we see in politics though? If you take a look at the Houses of Parliament here in the UK or you look at Congress in America, do you not already see this playing out weíre just basically proving what we already know?
Sinan - Not exactly. So, itís not clear whether or not influential people should be more susceptible or whether older people should be more influential than younger people, and my intuition is, that as we begin to apply this method across different behaviours and products, that weíre going to see different types of influence emerging in different contexts. In a different context, it could be that women are more influential than men or that younger people are more influential than older people Ė the opposite of what we find here, and the value of this paper is it provides a method to measure this in any context. I'm really excited to see what we might find for other types of products or behaviours.
Chris - If we could just look at the question of the men and the women, do you think the fact that it was movies might have led you to conclude that men were more influential than women in this context? Do you think if you've done something on a subject that women are regarded as more authority figures in, you'd have seen the flip side of the argument?
Sinan - It very well could be true, absolutely. So, you could imagine other contexts in which women might be more potentially influential than men, but this, in every context is an empirical question and the benefit of this measure is that we can now talk more scientifically, more rigorously about influence and susceptibility in a causal way.
Chris - And do you see this being applied to job interviews any time soon in a sense that you come from your job interview and someone decides they want management material or they want someone whoíll be well-trained and toe the line and it subjects you to this sort of analysis and you can put people into those sorts of categories?
Sinan - Yes, I think that it could certainly be applied to those types of situations, but I actually think itís much broader and itís interesting for other types of question as well. For instance, itís not only about targeted advertising or jobs. Weíre also now working on applying these same methods and the same science to promote HIV testing in Africa by trying to understand how we can use peer to peer influence to spread positive behaviours in society Ė diet, exercise, political awareness and like I said, HIV testing in South Africa.