Social Psychology Without Social Interaction

This is a story of how a last-minute idea (and the surprising result that emerged from the new codings) transformed a paper. It’s never too late to stop thinking!

Seeking to understand social psychology’s replication brouhaha, two colleagues and I reviewed all the published multi-lab replications in social psychology. We had finished the manuscript, polished it, revised it, and gone through the process of submitting it for journal publication. I was thinking, whew, that was a big job, glad it’s done for now. But then something else occurred to me. We ended up going back and coding all the studies again, running another analysis, and having to ask the editor to replace the submitted version with a revised version. To us, it was a bombshell.

Here’s what struck me. My own graduate training emphasized that the way to conduct a good experiment in social psychology involved setting up a situation that would be highly engaging with the participant, based on ongoing interaction between the experimenter and the participant (and sometimes a confederate). Live, carefully staged interpersonal interactions were the norm in social psychology for decades. Indeed, some of the best experimenters I knew had had theater training, because it helped them understand how to get the audience (the research subject) into having the intended experience and reaction. Good acting succeeds at sending the emotional message, and good social psychology experimenting thrived based on the same. Some graduate training programs (mine included) would encourage the Ph.D. students to take an acting lesson or two on their way to their psychology Ph.D.

More recently, however, researchers have abandoned this live-interaction style of research in favor of having research subjects sit at computers and make ratings. The gradual disappearance of behavior from behavioral science has been widely remarked (e.g., Baumeister, Vohs, & Funder, 2007). The pressure to run larger and larger samples has contributed greatly to this shift. If we think of Milgram’s obedience studies, for example — with an experimenter plus a paid confederate spending over an hour with each participant to get one data point — we can readily see how difficult it would be to do this with several hundred participants. In contrast, getting a couple hundred people to do an online survey is much easier and hence more efficient.

I wondered how much this affected the multi-site replications. So we went back and coded all 36 studies into three categories: A few did have live interactions, even if these were just that the experimenter interacted with the participant throughout the session. There was an intermediate category of computer-mediated or computer-simulated interpersonal interactions. And then the no-interaction category, which was by far the most common: Data collection involved the participant being alone or responding alone, at a computer or workstation, in a cubicle, in some cases in a large room with plenty of other participants silently and separately doing the same study.

It turns out that live interaction makes a huge difference. Most of the replication attempts were failures — yet none of the failures involved live interaction. Not one. The correlation between successful replication and having live interaction was a whopping r=.7. Indeed, the single most successful replication out of the 36 was also the only one to have live, unscripted social interaction. (It was most successful in that it not only yielded significant support for the original finding but also reproduced the original effect size; Ito et al., 2019).

To me, this underscores the conclusion that multi-site replications often get null results because they fail to engage the research participant as a human being. When talking and interacting with a live human being who is physically present, people respond — emotionally, motivationally, and in other ways. When sitting alone in a cubicle making ratings, their reactions are muted. And hence so are the data they furnish.

This is a bigger problem for social psychology than for other fields. The phenomena other areas study depend much less on getting the person responding fully as a social being.

There may be an important point about human nature here. Much of the human psyche is shaped by evolution for live social interaction. The influential French historian Philippe Aries famously asserted that until the 17th century, no one was ever left alone. (No doubt an exaggeration, but still broadly correct.) When you are dealing with real other people in the same room your emotions, cognitions, and motivations are all active. In sharp contrast, when you are alone with a computer, the emotions and motivations (at least) are muted. Social psychology experiments sharply lose power when they relinquish real social interaction.

John Bargh has talked about the change in social psychology. As he nicely put it, the field switched what it relied on for power in research. Early work, including the Milgram obedience studies, looked to the power of the situation. Modern studies have shifted to emphasize statistical power, which translates into getting data from lots and lots of people. Instead of trying to engineer a high-impact live interaction that will get people involved, we go for having large numbers of people furnish data. In practice, it means having lots of people sit at computers and make ratings.

There are some important reasons for making that shift. Still, it seems that there is a cost of minimizing the “social” part of social psychology. I suspect the roots of this go back to evolution. Humankind evolved to deal with live human beings, and in such situations mind and body come alive. In contrast, there is little evolutionary preparation for motivation and emotion to be galvanized when someone sits alone at a computer.

Psychology Essential Reads

Is a Texas Bitcoin Miner Making People Sick?

What Is a Scientific Model?

To forestall misunderstandings: I do not object to research that relies on people sitting alone at a computer making ratings. There is much to be learned that way. But when it becomes the main research method in social psychology, then something is lost, and many possibilities for scientific advancement are curtailed. In other words, I don’t have a problem with what social psychologists are doing — my problem is with what we are not doing. The glory years of social psychology were marked by a large diversity of methods. Crucially, the sense was that we should study every phenomenon with the best methods available for that topic. That’s what I still think is the best. But today’s psychologists feel pressure to use the same methods, and therefore many phenomena simply cannot be studied. Unfortunately, that includes most live interpersonal interaction. And without live interpersonal interaction, social psychology loses much of itself.

This sets up an intriguing dilemma for further multi-lab replications. Supposedly the shift toward getting large amounts of data from people sitting at computers will improve the validity of our findings, so they will replicate better. Yet the findings thus far suggest, on the contrary, that live social interaction produces more replicable results.