I have a paper that will be coming out in the upcoming Big Data special issue of Behavior Research Methods, a top methods journal in psychology. It’s called “The rippling dynamics of valenced messages in naturalistic youth chat” and it is out in an online preview version here:
https://link.springer.com/article/10.3758%2Fs13428-018-1140-6
The paper looked at hundreds of millions of chat in an online virtual world for youth. The popular pitch for the piece is about your words coming back to haunt you on social media. That’s one takeaway you might draw out from the work we did. We looked at social influence: how my words or actions affect you. Of course, a lot of people look at social influence. Some papers look at those influence over minutes, and that’s good to do because it might help us understand behavior in, say, online political discussion. Others look at social influence over years, and that’s also good to do because it tells us how our peers change us in the long term. But say you wanted the God’s eye view of specifically what kinds of small daily interactions have the smallest or largest effect on long-term influence. That would really get at the mechanisms of the emergence of identity and, in some sense, social change. But the same things that make that kind of conclusion exciting also make it hard to reach. Short-term social influence is a tangle of interactions, and long-scale influence is a tangle of tangles. We were able to untie the know just a bit. Specifically, we reconstructed the flow of time for chat messages as they rippled through a chat room and, reciprocally, as they rippled back to the original speaker. The finding was that, when I say something, that thing elicits responses in two seconds (predictably), and keeps eliciting responses for a minute, getting stronger in its effects quickly, and then slowly tapering off. The effect of the single chat event is to produce a wave of chat events stretched out over time. And each of those itself causes ripples that effect everyone else further, including the speaker. Putting it all together, your words’ effect on other ripples back to affect you, in a wave that starts around 8 seconds, and continues for several minutes, almost ten in you were being negative. We are able to count the amount of chat that occurred as a consequence of the original event, that wouldn’t have occurred if the chat hadn’t happened. By isolating your effect on yourself through others, and mapping that wave’s effects from 2 seconds to thirty minutes, we’re able to put a quantitative description on something we’ve always known but have rarely been able to study directly: the feedback, self-activating nature of conversation and influence. If chat rooms are echo chambers, we were able to capture not just others’ echoes of you, but your own echoes of yourself in the past.
Social scientists are very ecological in their understanding of causes and effects. If you stay close to the data, you are bound to see the world in terms of everything affecting everything. It’s what makes social science so hard to do. It’s also what makes virtual worlds so exciting. They are artificial places composed of real people. Stripped down, the social interactions they host can be seen more clearly, and you can pick tangles apart in a way you couldn’t do any other way. For this study, we were able to use a unique property of online youth chat to pry open an insight into how people’s words affect each other over time. To really do that, you’d have to piece out all the ways I affect myself: I heard myself say words, and that changes me. I anticipate others responding to my words and that changes me. Others actually respond and their responses change me. Those are all different ways that I can change me, and it seems impossible to separate them. The accomplishment of this project is that we were able to use the artificiality of the virtual world separate the third kind of change from the other two, to really zoom in on one specific channel for self influence.
This world is designed for kids, and kids need protection, so the system has a safety filter built in. The way the filter works is that if it finds something it doesn’t like, it won’t send it, but it also won’t tell you that it didn’t send. The result is that you think you sent a chat, but no one ever saw it. That situation never occurs in real life, but because it occurs online, we are able to look at the effect of turning off the effects of others hearing your words, without changing either your ability to hear your own words or your belief that others heard you. With this and other features of the system, we were able to compare similar messages that differed only on whether they were sent or were only thought to have been sent. By seeing how you are different a few seconds after, a few minutes after, when you did and didn’t actually reach others, we’re able to capture the rippling of influence over time.
This is a contribution to theory and method because we assumed for decades that this kind of rippling and tangling of overlapping influences is what drives conversation, but we’ve never been able to watch it in action, and actually see how influence over seconds translates into influence over minutes or tens of minutes. That’s a little academic for a popular audience, but there’s a popular angle as well. It turns out that these patterns are much different if the thing you said was positive or negative. That has implications for personally familiar online phenomena like rants and sniping. The feedback of your actions onto yourself through others corresponds to your rants and rages negatively affecting you through you effects on those you affected, and we’re able to show precisely how quickly your words can come back to haunt you.
