S7 Ep12: Can contact between groups reduce prejudice?

We might have thought that prejudice would be something that would just go away with development or with new generations replacing all the generations, but that's not really what we see. Welcome to Vox Dev Talks, my name is Tim Phillips. The idea that contact between members of different groups under the right conditions produces prejudice dates back on 70 years now. At one time it was considered one of the best supported results in social psychology, but is that what the evidence really shows? A recent review suggests that the effects of intergroup contact might have been wildly overstated. Matt Lowe with the Vancouver School of Economics wrote that review. He joins me now. Welcome to Vox Dev Talks, Matt. Thanks for having me, Tim. Matt, what do we actually mean by intergroup contact? Usually we think of intergroup contact as being some form of interaction between members of different social groups, so these could be people of a different gender or a different caste or a different race, and often we'll be thinking of interactions that are face-to-face and in person, although the work on intergroup contact also considers interactions that are online or even just over text messaging. So the important thing is that really we're thinking of not just proximity to an out-of-group member, but some kind of meaningful interaction between an angry plan out-of-group member. And this theory, if I'm right, dates back to Gordon Allport's work in the 1950s. So what was the origin of that work? Why was it so relevant at that time? Yes, so Gordon Allport wrote this book called The Nature of Prejudice in 1954. You know, this is the book that everyone sides in modern papers on the contact practices. One interesting thing is that only one chapter in that book is on the contact practices, and all the other chapters are on all these different dimensions of thinking about prejudice, like the personality origins of prejudice and these different things. And so in that one chapter, just once more slice of the book, the tries to rationalize some existing findings by forming it in this idea of a contact hypothesis. And so the existing findings are really just a handful of studies that US researchers have done in the context of race relations in the 1930s and the 1940s. One example of this is this study by Ayra Brophy in 1946, what he does is he surveys about 400 white sailors, and then he correlates their prejudicial attitudes towards black people with different variables, and then one of the things he finds is that the white sailors that have been shipped more times with black sailors are less prejudiced. So that's kind of an early correlation that hints at contact hypothesis. But then at the same time, there were other studies that would find opposite results. Like there was a study on white students at university that found their longer they interacted with black students at that university than actually the more racist they became. What Allport then does in his book in this one chapter of his? He says, okay, to kind of rationalize these different correlations, I'm going to conjecture that contact works only one certain conditions are met. And then Allport had a graduate student, Thomas Patigrew, who later refined these conditions into the four conditions that everyone now talks about, and those conditions are equal status. So you know, the group should have equal status in a situation, common goals, in group cooperation to achieve those goals, and the support of authority, norms, and custom. And so I think the way to think about it is that because at that time there wasn't even that much evidence. It's really true that it's like a contact hypothesis rather than actually a law of nature or really reflecting what we know about the world. It's a conjecture or a hypothesis that what seems likely to be true. But by establishing these conditions for the contact hypothesis, then this becomes a situation in which we can intervene. Is this the way people were thinking about it in the 60s and 70s? We could actually reduce the amount of racism in society. Yeah, absolutely. I think, especially in the context of posts, you know, World War II, when white and black men had fought alongside each other, people are really thinking about now positively. What can you do about improving race relations? And the contact hypothesis gives a guide of how to do this, essentially. Of course. And at that time, it was the time of the civil rights movement as well. So this was a very live issue in the United States and around well exactly. The evidence base at that time, I guess that was the sort of evidence that people used, but it doesn't sound particularly strong. People weren't doing any kind of experiments, controlled experiments. This was observational. It was surveys. Yeah, exactly. I mean, I went back and read this profile paper from 1946. And what's interesting is that the whole paper is actually just cross-taps. It's just correlating, you know, one variable with another time, table after table. And so, you know, that time there wasn't even controlling for other variables. There is the odd example of a study that comes close to being what we would now call the quasi-experiment. There are some studies that were a bit more ahead of their time, but mostly these observational correlations, no real experiment. However, there have been a lot of policies, a lot of policy interventions based on this hypothesis. What sort of policies have been based on it in the last, well, 70 years now? I think you see a lot of policies that have, as part of the idea of the policy, the contact hypothesis. And then there's a lot of policies that are just about trying to improve the situation of disadvantaged groups that have a contact is potentially a side benefit. One place I think you see this a lot is in terms of schooling. The most famous example in the US would be trying to desegregate schools through busing, which is quite controversial. In schools, even once the schools are mixed, people within the school might self-segregate. And so, there are other policies that you can think of trying to reduce self-segregation within schools. One example is that you can have these things called house systems. And at 10, did you grow up in a house system and you're schooling? Yes, my school did have houses, yeah. Okay, so there you go. We're actually evaluating the house system in India right now with schools around the mice to get it, not. And so, the house system is a way of, you know, randomly assigning people to groups that are going to cross-cut existing group identities. I think some of the other interesting examples are these nation-building policies that involve sending young adults to different parts of the country to live and work. I guess one really nice example of this is Nigeria has this national youth service core, which Ayurbola, up and up there has this nice AJ applied paper on. I guess another example would be that there are just many NGOs that are involved with conflict resolution, like search for common ground. And often they'll have some bundled intervention, but a component of that bundle will be trying to actually mix people across group lives. Hi there. Oliver Hannie here, managing editor of VoxDev and the host of our new podcast Ideas in Development. Over February and March, Dina Musa and I are hosting a series of conversations about the economics of AI. We've talked to Rose Matiso, Raghuram Rajan, Anton Koronek, Josh Lerner and other leading thinkers and researchers across 10 episodes. To listen to this series on AI, go to the separate Ideas in Development podcast feed by searching Ideas in Development wherever you get your podcasts. So when did the basis of contact theory, when did it begin to be seriously questioned? Around 10 years ago, so this is when I was in grad school, people started to run quite ambitious field experiments. So taking contact very seriously, going to a situation where there are real intergroup relations problems, randomly signing people to either have contact with the outgroup or contact with the ingroup. A few examples of these that found quite disappointing effects, I think, given what people felt they knew about the theory. The first big example was this paper by Alexandra Sceco and Shayna Warrens. This was mixing Christians and Muslims in Nigeria through vocational training, extremely well done study, but quite mixed effects of contact. And then there was another influential study sound moves this paper on mixing Christians and Muslims, which you might know of as well. Yes, through football teams in Iraq, she finds positive effects on behaviors related to the football. So things like training together, people continue to train with the outgroup, but not really any generalization to off the field behaviors. So things that are more related to intergroup relations outside of the context of football. So those two were just fantastic studies, like well-published, well-executed, find pretty mixed limited effects. A few years later, there were enough experiments now to have a metro analysis just of the experiments. And so that's what Betsy Palak did with Seth Reid and Don Green in 2019. In their metro analysis, they actually find that on average, the effects are pretty large of contact about point four standard deviations, but they start to see these signs of publication and reporting biases. They look at the biggest studies relative to the smaller studies, the bigger the study, the more people in it than the smaller with the treatment effects. And then if they restrict to the studies that had pre-analysis plans, then there's no treatment effect at all. So I think it was this new wave of experimental work that started to make people think, this is not as unambiguous as we thought it was in terms of the effects of contact. Yes, you could say that this new wave of experiments were part of the credibility revolution that we have seen, where we're much more rigorous about where we consider strong evidence comes from, explain a little bit more about the methodological problems of the historical research into this. Like you say, the first thing is kind of an identification problem, which is precisely what the credibility revolution was trying to solve. And so I think in the context of contact it's particularly severe. because we would really worry about reverse causation. So, you know, we would really worry that if you see correlation between contact and prejudice, it could be because the less prejudice people are more likely to select into having more contact with the output of goal. And so that's obviously gonna be a problem if you just do it by cross tab, like in a 1946 paper, we should still be trying to their research set of works on these things. People have really tried to deal with those issues pre-experiments through potentially adding controls, but I think it probably just turns out with contact that it's very difficult to completely purge that bias without actually having a clean natural experiment or a clean field experiment. And then there's this second problem which I think is taking fields across the social sciences a bit longer to solve, which is the reporting and the file draw a tight problem. So the problem that even if people are running experiments, we might not see the experiments run up that find null effects or find unexpected effects. Or even if we see them run up, we might see the paper focuses on the one outcome that had a positive effect rather than the other outcome. So I think these are the draw problems, identification and reporting. And then to the extent that we can solve both of those, I think we start to see a picture of what really are the typical effects of intergroup contact. - To your point, this isn't just something that applies to intergroup contact. This has been across the social sciences. We have seen evidence of this in the last few years. - Yeah, exactly. Things are improving very fast in terms of people running experiments to do pre-registration. - Yes, and explain pre-registration to me and what's going on here. - Pre-registration, I guess we usually think of as people before they run their experiment, they post online publicly a description of the experiment they're about to run and some minimal details like the sample size, the primary outcomes that they want to look at, what that pre-registration helps with is that when you see the paper afterwards, you could check that people didn't have 100 outcomes and then they write the paper just about one outcome, which was the one that happened to be statistically significant. And so in the mattress analysis that I've done on contact, that was one of the key things was to not only restrict to experiments where we think we have good identification, but to also restrict to experiments that had been pre-registered, that way you can look and track through the effects on the outcomes that they pre-registered as primary rather than the outcomes that they decide to emphasize in the paper. - Were there enough studies that were pre-registered for you to do a meaningful meta-analysis? - This was the part that was a bit surprising when we first started searching for these papers and I should say I had a lot of help here from an amazing RA Catalina Valenzuela. So when we started searching for these papers, the first thing that made me realize this was worth doing was that there was just a surprising number now of pre-registered experiments on contact. We end up with 41 experiments. And so a lot of this is just because there's been a real boom in running these types of experiments even in the last five years, I think the majority of these are in the last five years. The 41 experiments that I end up with are almost completely non-overlapping with previous meta-analyses. So it's almost completely new experiments and they cover about 40,000 people. So that's large enough to get fairly precise confidence intervals around these estimates of contact. - Yes, that is surprising. It's far more than I would have expected. It's also very encouraging. But what size of effect are you seeing for these pre-registered experiments compared to the sort of evidence that we've relied on in the past? - I guess for context, there was a famous meta-analyses in 2006, covered 515 contact studies and they found effects of about 0.4 standard deviations. That meta-analyses cited around 14,000 times. So it's very influential. Oh yeah, mine is cited about 14 times. So they're still, it's early days, yeah, Matt. It's early days. - It's early days. - And then there was Betsy Palaknikov authors in 2019. They found also 0.4 standard deviations and then Khlosha more recently finds 0.2 standard deviations. When I bring in these newer experiments and I restrict a pre-registered primary outcomes, I'm finding something that's roughly 0.1 standard deviations. So about a tenth of a standard deviation. The trajectory is down. Hopefully now we're going to stabilize 0.1 and not keep going. For people who don't understand exactly what that means in real life, can we interpret what 0.1 standard deviations might mean? Yes, I think the easiest way is to think of a binary outcome. Suppose we care about whether a participant shares their money with an out-group member or doesn't share it. A tenth of a standard deviation effect is an increase from 50% of people sharing to 55% of people sharing. There's a glass half full and a glass half empty perspective, which is that's not a transformative effect. But for a modular intervention that involves 10 hours of contact, I think that's pretty meaningful actually and still kind of worth doing. These previous match analyses, the effects that people were thinking of were much bigger. They were more like 50 up to 60% or 50 up to 70%. And so you can think of those as being a view of contact as being much more transformative is what I find when I use these 41 experiments. I would imagine a lot of people would be invested in contact theory and the promise of it. And they won't enjoy hearing the news that you're delivering for them. Does the credibility revolution and the fact that we can do these meta-analyses now? Does it mean that we really have to confront this? Definitely. I just see this as becoming more realistic about what these types of interventions can do. In this case, we're trying to reduce prejudice. It's unsurprising, I think, that a few hours of interaction with the outgroup is not going to have these big transformative 20% of point effects. When you think about it that way, then it really does make sense, yes. Yeah, we think that this should be something that's quite hard to shift in people because these are people's attitudes and beliefs that they might want to defend. So I think it's very healthy to try and really get accurate realistic estimates of these policies to be able to think about how we should do policy. But I think the other important thing here is to realize that the standard that I've applied to this meta-analyses for contact is extremely stringent in a way that I think has not yet been applied to other classes of prejudice reduction intervention. Even though this meta-analyses says, maybe we should revise our beliefs about contact downwards. It's not clear yet, I think, if we should think of contact as being less effective than the alternatives, like a media-based intervention. What would be nice is if in the next 10 years or so, we can see that a similar standard applied across all these different classes of prejudice reduction intervention, and then I think all these estimates will become more realistic, and then we might have a sense of whether one is still better than another. And it still could be the case that contact is actually still better than all the alternatives that we have out there. To be clear, what you are not saying is that contact theory is no good, a waste of time, wasted money. We are not saying that. Yeah, not at all. Two things. One is that contact is often cheap to weave into a program because there are many settings in which people need to do some collaboration in groups anyway. Yes. And so given that the effects, even in my meta-analyses, are typically positive, they may be small, but given that the costs can be quite low, then it will still often be cost-effective to use contact in a program. And then it's still hard, I think, to compare contact to the alternatives without holding the alternatives to the same standard. And what we do know about the alternatives from, again, you know, work by Vetsy Palak with Ronnie Pora, Chelsea Clark and Don Green. So they have an annual review of psychology article that looks at 400 prejudice reduction experiments across all these different domains and classes of intervention. They would say that outside of contact, we mostly just don't have a great evidence base yet. Mostly like touch interventions, small scale experiments. The experiments that are more ambitious and big tend to find smaller effects. And so it's still possible that contact will be better than these other interventions. I think we don't yet have the apples to apples comparison that we would need. It's one of the seductive things about contact theory as we understood it in the past was that it just had these four conditions, which anyone could understand. They were very intuitive. If it turns out that that isn't sufficient to produce the size of effect that we want, could we maybe refine this and think, OK, well, here's some more conditions under which it is effective. Do we have an idea what those might be yet? So what do we know about the awkward conditions now? There are experiments that have started to randomize the conditions themselves. Uh-huh. And so typically those experiments do find that if you turn a condition on, then the effects of contact are more positive. The conclusion now, I think, is that the awkward conditions are not enough to guarantee big effects, but they do tend to improve effects. Can we come up with more conditions that will make the effects bigger? People are starting to do that a bit, trying new things. There's something that I've done in the recent paper with my frequent co-authors. Arcadevegosh, Garanthnullis, and Agit Chakrabatines. And we have a paper where we randomize whether the contact is broad or deep. And so what we mean there is broad contact is you have contact with many different output members, but for a short amount of time each. And then deep contact is you have longer contact with one person. And in that paper, we find it just depends on the outcome you want. Brad is good for making beliefs about the output more accurate. Depth is good for creating social and economic ties with the output. Even there, these are not generating these big positive effects, so it doesn't solve the problem. What else should people look at? I mean, an obvious thing is the duration of contact, the intensity of contact. Surprisingly, people and experiments have not really played with varying just how long you have. A lot of experiments have quite short on contact. And so we might also think that we've just got into this mode of thinking that two hours of contact is going to be effective. But actually, maybe it really needs to be a long period of time very intensive. There's a huge amount of ideas and psychology of different conditions, you know, after all people continue to pitch conditions to the point that it's almost, I think, a running joke and psychology and out that, you know, there's a thousand conditions. How can we have exactly some thousand conditions? For the economists who are studying these things, this would be the place now to get these ideas is to go back and read psychology and think about the things that people have suggested as possibly maturing. And it really does matter, doesn't it at the moment, Matt, because look around us, the rise of conflict in the world, the rise of prejudice. We are seeing it openly displayed across Europe and North America. It is really important that we do manage to create effective policies to reduce prejudice. 100%. We might have thought that prejudice would be something that would just go away with development or with new generations replacing the old generations. But that's not really what we see. And so we see that prejudice remains a problem regardless of the state of development. There's a really important role for policy to try and reduce prejudice because it's not going away on its own. And I think the field of prejudice reduction is exciting right now because I think we're just on the cusp of now running these really ambitious interventions and moving away from the light touch. Very gradually, I think in the next maybe 10, 20 years, we're going to hopefully start to learn what actually can work and what can work people when we scale it up, which I think the scaling issue, I think we're not even there yet, that makes it a very exciting field to be a part of. As you say, glass half full or glass half empty, thanks for talking about it, Matt. Thank you, Tim. If you want to read this paper, it was published in the annual review of economics 2025 and the title is Has Intergroup Contact Delivered. Author of course, Mallow, you just heard from him. This has been a Vox Dev Talk. The best way to make sure you don't miss an episode, you can accompany me on this now is to say you subscribe wherever you get your podcasts, you find us there. Our best episodes are always a Vox Dev.org where you find articles about the topics we feature. Like what you're hearing, tell someone else about us, leave us a review as well. Vox Dev Talks is a talk normal production. The assistant producer is Megan Bieber and our editor is Andrei Zalgarian. Thanks to Oliver Hany and Iman Sidiq at Vox Dev.