It’s real­ly a huge hon­or to be here amongst all of these real­ly incred­i­ble ideas and amaz­ing peo­ple, so thank you very much to to PopTech. What I want to talk to you about today are two things, behav­ioral con­ta­gions, and causality. 

And the rea­son that I am inter­est­ed in behav­ioral con­ta­gions is that I firm­ly believe that if we can under­stand how behav­iors spread in a social net­work and thus in a pop­u­la­tion from per­son to per­son to per­son to per­son, that we could poten­tial­ly pro­mote behav­iors like these, con­dom use, or tolerance.

And then we can also poten­tial­ly con­tain behav­iors like like these, smok­ing or dirty nee­dle sharing. 

So, what I do is I mine mas­sive social net­work data to under­stand how behav­ioral con­ta­gions spread in human social net­works, how peers influ­ence one anoth­er to do things like quit smok­ing or buy a cer­tain prod­uct or vote for a dif­fer­ent polit­i­cal can­di­date. And I focus on strate­gies, incen­tive strate­gies, to cre­ate cas­cades of behav­ior through a pop­u­la­tion. And one of the most impor­tant keys to all of these endeav­ors is causal sta­tis­ti­cal estimation. 

So, I have this car­toon on my door at my office. It’s two friends talk­ing, and one friend says to the oth­er, I used to think cor­re­la­tion implied cau­sa­tion. Then I took a sta­tis­tics class and now I don’t.” And the friend says, It sounds like the class helped.” And the guy goes, Well, maybe.”

And the idea is that maybe this guy has a pro­cliv­i­ty to under­stand sta­tis­tics and so select­ed into the class, and that this cor­re­la­tion between tak­ing the class and under­stand­ing this dif­fer­ence isn’t causal. The class isn’t teach­ing him about sta­tis­tics at all.

And in net­work sci­ence this is known as the reflec­tion prob­lem. Human behav­iors tend to clus­ter in net­work space and in time. We have abun­dant evi­dence on this now. But is this because of peer influ­ence, or alter­na­tive explanations?

Let me give you an exam­ple. James Fowler came and gave a talk here last year about his study on obe­si­ty being con­ta­gious with Nicholas Christakis. This is a great study that shows basi­cal­ly that Body Mass Index increas­es are cor­re­lat­ed over time amongst friends. And this was picked up by the col­lec­tive uncon­scious and by the media, and you got news sto­ries like this one that said are your friends mak­ing you fat?” And so, the basic idea here is that there could be alter­na­tive expla­na­tions for this. Like for exam­ple, birds of a feath­er may flock togeth­er. We may choose friends who are more like our­selves and thus cre­ate cor­re­la­tions in the behav­iors amongst peo­ple that are con­nect­ed in a social network. 

And this is a long line of the­o­ry. This quote was orig­i­nal­ly attrib­uted to Robert Burton in the 1500s. But it’s even old­er than that, because before Robert Burton it was Aristotle that was say­ing that peo­ple love those who are like them­selves. And before that it was Plato who said that sim­i­lar­i­ty begets friend­ship. And just to prove that it was a long line of wor­thy schol­ars who said this argu­ment, my mom said to me, Hanging out with a bad crowd will get you into trou­ble.” She was none too pleased when I told her that she might might have got­ten the causal struc­ture of that argu­ment wrong.

And so Slate ran an arti­cle that was titled Everything Is Contagious” [part 2]. It’s not just about obe­si­ty. Apparently hap­pi­ness is con­ta­gious, and prod­uct of option is con­ta­gious, and coop­er­a­tion is con­ta­gious, and lone­li­ness is con­ta­gious.

But it’s also not just about birds of a feath­er. There are many alter­na­tive expla­na­tions they can explain these cor­re­la­tions. Like, friends might be exposed to the same exter­nal stim­uli, or have a greater prob­a­bil­i­ty. This is impor­tant for two rea­sons, because the causal struc­ture of the under­ly­ing dynam­ic process of the spread implies dif­fer­ent dif­fu­sion prop­er­ties for the behav­ior. Where’s it going to go next, so who should we tar­get? And dif­fer­ent opti­mal con­tain­ment and pro­mo­tion poli­cies. How should we design sys­tems that pre­vent or pro­mote these behaviors. 

Let me give you two exam­ples. We stud­ied which types of mes­sages spread a con­ta­gion best. And we stud­ied the dif­fer­ence between per­son­al invi­ta­tions and pas­sive aware­ness. So imag­ine if I gave some peo­ple the oppor­tu­ni­ty to per­son­al­ly invite their friends to PopTech next year, and then I gave oth­er peo­ple in the room the oppor­tu­ni­ty to wear a shirt that said that they went to PopTech last year.

Which of these would be more effec­tive? Well it turns out that as you would expect, the per­son­al invi­ta­tion is more per­sua­sive. More of those peo­ple came to the behav­ior as a result of the per­son­al invi­ta­tion. And in terms of glob­al dif­fu­sion, it near­ly dou­bled the behav­ior in the glob­al net­work for which this per­son­al invi­ta­tion was released. But in fact it was the pas­sive aware­ness that gen­er­at­ed much more glob­al dif­fu­sion, because more peo­ple were exposed to it.

And I’ll leave you with this final result. We did a ran­dom­ized tri­al of sus­cep­ti­bil­i­ty to influ­ence on Facebook amongst two mil­lion peo­ple. And ini­tial results look like this for what your rela­tion­ship sta­tus is; sin­gle, in a rela­tion­ship, engaged, mar­ried, or it’s com­pli­cat­ed. And sin­gle peo­ple are are more sus­cep­ti­ble to influ­ence than those that don’t report their rela­tion­ship sta­tus. If you’re in a rela­tion­ship you’re even more sus­cep­ti­ble to peer influ­ence from your peers. If you’re engaged, you’re even more sus­cep­ti­ble to peer influ­ence. And if you’re mar­ried, you’re not sus­cep­ti­ble at all to peer influ­ence, appar­ent­ly. And if it’s com­pli­cat­ed, you’re the most sus­cep­ti­ble. So my col­leagues and I debate what is going on here. And if you’d like to debate that with me or talk about any of these results, I encour­age you to come talk to me. And more impor­tant­ly, I encour­age you to tell your friends to come talk to me.

Thank you.

Further Reference

This pre­sen­ta­tion at the PopTech site.