This episode is the first in a mini-series of weekly short episodes featuring young scholars entering the academic job market who discuss their latest research. In this episode, Shruti speaks with Arkadev Ghosh about his job market paper titled “Religious Divisions and Production Technology: Experimental Evidence from India.” They discuss the effects of inter-religious work groups on team productivity, how wider political tensions can affect the workplace and much more. Ghosh is a Ph.D. candidate in economics at the University of British Columbia. He obtained his master’s in economics at the London School of Economics and his bachelor’s degree at the University of Edinburgh.
SHRUTI RAJAGOPALAN: Welcome to Ideas of India, a podcast where we examine academic ideas that can propel India forward. My name is Shruti Rajagopalan, and we are kicking off the 2021 job market series, where I speak with young scholars entering the academic job market about their latest research on India. Our first scholar in the series is Arkadev Ghosh, who is a Ph.D. candidate in economics at the University of British Columbia. He obtained his master’s in economics at the London School of Economics and his bachelor’s degree at the University of Edinburgh. We discuss his job market paper titled “Religious Divisions and Production Technology: Experimental Evidence from India.”
We talked about his field experiment on worker productivity in religiously homogenous versus mixed groups in both high dependency and low dependency tasks, what this experiment reveals about Hindus and Muslims at the workplace, and his other research on political business cycles in India in conflict areas.
For a full transcript of this conversation, including helpful links of all the references mentioned, click the link in the show notes or visit mercatus.org/podcasts.
Hi, Arkadev. Thank you so much for coming on the show. It’s a pleasure to have you here.
ARKADEV GHOSH: Thank you very much for having me, Shruti.
Ethnic Diversity and Private-Sector Output
RAJAGOPALAN: I want to go straight to your job market paper. This is a fascinating experiment that you’ve conducted in West Bengal, where you randomly assign Hindu and Muslim workers at a manufacturing plant to either a religiously homogenous work group or a mixed-religion work group. And you find that mixed teams are less productive than homogenous teams, but only when it comes to a very high dependency [HD] task. You don’t find that result for low dependency [LD] tasks.
At the same time, while there’s this negative impact on productivity when there are inter-religious groups, that effect fades away over a period of time and in a few months. Can you describe this experiment and the mechanism at work here that you observe that leads to this productivity shock and then comes back to normal?
GHOSH: Absolutely. The broad motivation for this paper is really this recent evidence that ethnic diversity can lower private-sector output. We know there’s a larger literature looking at this in the public sector, looking at coordination failure amongst groups, but really there’s been this recent evidence that due to poor social ties or caste-based discrimination, mixed groups of workers can perform worse in the private sector.
It’s natural to think that maybe firms will react to these costs. Maybe they will stay small to minimize interactions, or that as they grow, they may segregate workers, which would only perpetuate discrimination in these societies. But really, to fully understand these effects, we need to know two things, basically: the long-run effects of contact in firms, which we don’t have evidence on—maybe things improve over time.
Also, maybe we don’t need to segregate workers. Maybe we just need the right technology as we grow. That’s exactly what my job market paper tries to answer. I look at both the short- and the long-run effects of intergroup contact in firms and see how these effects depend on the production technology. Like you summarized, I randomized workers into either mixed or religiously homogenous teams and kept these teams intact for a period of four months.
Really the key aspect of this is how the effects vary by the production technology or the production function. Like you said, a high dependency work here—just to elaborate a little bit more for the listeners what a high dependency work is—is the classic example would be work on a conveyor belt. There are multiple workers standing, and then each person is responsible for picking up every second or third piece of the product. There’s this instantaneous contact and coordination required with your teammates.
A low dependency task, in contrast, and since this is a food manufacturing plant, let me talk about a mixing room where raw materials are mixed. One worker is responsible for just weighing the materials properly; the other workers are arranging the flour buckets, where someone else is actually mixing into the machine. It’s not like you don’t need contact, you still need coordination, but contact is more loose in a way, if I’m to use that word, and it’s more intermittent.
It’s perfectly possible to undermine or sabotage your teammate in this setting as well. A core idea that differentiates—the core idea in economics that differentiates between HD and LD tasks is the degree of complementarity of labor inputs. It’s high degree of complementarity in HD. I find three key things, like you said. I find that mixing reduces productivity, but this happens only in HD tasks, high dependency tasks.
Consistent with this, I find that there’s more frictions and lower team cohesion when I mix workers in high dependency tasks. There are more accusations and inter-group blame and, in general, a lower team cohesion. There are some frictions in LD tasks as well, but the production technology is such that it mediates these frictions and it doesn’t affect output.
The second result, like you mentioned, is that despite these negative effects, over time through repeated contact, they go away. This is something that past studies haven’t shown, typically, because they don’t look at the long-run effects. The institutional setting doesn’t allow them to look at the long-run effects of repeated contact. I find that over time with repeated contact, these effects go away.
The third key thing that I find is that—and this for me is the most interesting part—is that despite these negative output shocks, intergroup relations improve from mixing in high dependency. Whereas in low dependency, while there is no output shock, there are no improvements in productivity either over time. This really suggests that there’s this tension between the goals of maximizing productivity versus improving intergroup relations.
Contact that really forces people to learn to work together is important in reducing discrimination and prejudice, but it might be costly since there are these frictions. And maybe you just don’t like your teammate in a low dependency task, but there’s no real cost to output. Maybe that’s also why there’s no incentive to try to overcome your differences. That’s a big-picture takeaway from the paper.
Hindu and Muslim Beliefs About Each Other
RAJAGOPALAN: One of the mechanisms at play here, or at least as you find initially, is that Hindus as a group have very low priors regarding how hardworking their Muslim coworkers are, relative to their within-group Hindu coworkers. In contrast, Muslims don’t make a similar distinction between in-group Muslim coworkers versus Hindu coworkers. What are some of the reasons these priors exist and are reinforced over a period of time, and how do they exactly break down through interaction in these high dependency tasks? Can you just walk us through that process?
GHOSH: Sure. Yes. That’s a great question. Just to rationalize some of these core results that I told you, I developed a theoretical framework in the paper. All it helps me do is differentiate between competing mechanisms. There could be a multitude of explanations for what I just told you. A fundamental assumption of the theory or the model in the paper is that Hindus start off with a lower prior about Muslim teammates, about their effort or how hardworking they are, whereas Muslims do not make this distinction.
At first glance, you may think that a weird assumption to make, but this is really based on the empirical fact that Muslims are always working with Hindus, whereas Hindus are not, just because of the distribution of Hindus and Muslims in the population. There’s this fundamental asymmetry and baseline contact that Hindus and Muslims have with non-core religionists.
Of course, there’s residential segregation. People go to different schools, but in formal workplaces or factories, Hindus being the majority, they don’t come in as much contact with Muslims as Muslims do with Hindus. And together with the general evidence of labor market discrimination against Muslims and the fact that they’re deemed to have lower education, I assume that Hindus start off with this lower prior, whereas Muslims, having always worked with Hindus, do not make this distinction.
One of the key implications of this model is then the priors of Hindus would really matter. Whether they’ve had past contact, or what they believe about Muslims, would really determine the effect you have in mixed teams. This is heterogeneity by priors of Hindus, whereas Muslims would essentially be more forward-looking and trying to integrate in knowing that they’re discriminated against. They’re really used to this.
I find evidence that is consistent with this mechanism. In the paper, I actually find that while Hindus are more likely to blame initially low output on their Muslim teammates, Muslims actually work harder than would be optimal, having this long-term view in order to mitigate these negative effects. That’s essentially why we see this attenuation.
I think this general finding is applicable to many settings, even outside India. You can think of—there’s work on how African Americans have to work twice as hard to achieve similar career goals. Or women have to work twice as hard—or even Asian immigrants in the U.S., how they try to achieve a model minority status as opposed to trying to culturally assimilate. I think this minority group initiating the integration process is applicable to many other settings, even outside this Hindu-Muslim context.
RAJAGOPALAN: What is interesting about what you find is that the prior is not very strong; it’s a relatively weak prior, which can be overcome through this kind of initiation through the minority group. And it can be overcome sometimes in as little time as four months, right? I know four months is a fairly long period of time in a production cycle at a food manufacturing firm, but it’s a very, very short period of time when we’re talking about weakening of long-term held views and biases. Right?
GHOSH: Absolutely, and this gives me a lot of hope. If you can convince firms, or even think of policies or subsidies for firms that have HD production functions, that would really help integrate workers in these kinds of settings. Because as I was mentioning briefly, there’s really this tension. As a firm, if there’s output lost from mixing workers, even if it’s for 10 days, I have no incentive to do that.
Whereas, if there are really these positive effects that we’re finding of mixing in these groups where there can be negative effects on output initially, maybe there’s clearly a case for government intervention or subsidies to these firms in order to incentivize them to integrate in these settings.
RAJAGOPALAN: On the other hand, I would say that even for the firms, there might be an incentive in the longer run, because segregation of workers works out great when everything is normal. But the moment there are conflicts, actually, segregated workers lead to more violence and worse tensions and worse outcomes. For the long-term productive health of a firm, it’s actually better to spend the money in the short run to integrate and have mixed teams, and have that social cohesion within the firm, than risk some of the terrible consequences that come from long-term segregation.
GHOSH: I couldn’t agree more with you. This is a very, very key point. To really understand this better, I actually went ahead and surveyed more than 100 supervisors at different firms after my project. Part of this was to assess the external validity of my results, whether they can predict the results, and also to get recommendations for policy implications.
One of the things that I asked them was, “If you see these negative effects on output, would you segregate workers?” A fair share seemed to understand that these effects go away over time. They internalized this fact, and they say no. But the first concern that the people had was—like you just said—it’s impossible to segregate workers; we just cannot do that because that will actually raise tensions. Think of a line with only Muslims and only Hindus. We don’t do that; that’s an impossible HR policy to implement. It’s always in the interest of the firm to have workers integrated, especially because of the political instability in these areas.
RAJAGOPALAN: One interesting thing that I learned from reading your paper is that some of the work that you did, you made these observations during the CAA protests and in the backdrop of the riots that broke out. You find that the backdrop of riots, even though it’s not happening within the firm, it still impacts the productivity of tasks, especially in low dependency tasks, which previously were not that affected by the kinds of intergroup mingling that you were talking about.
What is the mechanism at play there? These are tasks where it’s harder to sabotage your coworkers, but in the midst of tensions outside of the firm, there is almost a greater incentive to sabotage workers. Did I interpret that result correctly?
GHOSH: Right. I look at two different events. There were the riots in New Delhi, but a fair bit of violence also broke out in West Bengal, and especially in the district where the factory is located. I use both of these events.
What I find is that during these periods of heightened tensions, there were negative effects in low dependency as well. This is important for two reasons. Firstly, it shows that the production technology is important, so things can go wrong. These are actually teams as well in low dependency, but we see that only during this heightened period of tension. It really tells you that if you wanted to sabotage or undermine the efforts of teammates, or you have a strong dislike or distaste towards your teammate, you really take actions that otherwise you wouldn’t.
People don’t come into the firm thinking of, “Oh, I definitely want to sabotage or undermine my own group member.” That’s not the mentality I saw generally amongst workers, but the political discourse or these extreme events can make them do that. That’s also why, in general, I gravitate towards this mechanism of priors rather than this strong distaste against our group workers, because if that were the case, we should always see these effects in LD—like you said—and we do not. That means that workers are really not trying to always sabotage—not coming in with that mentality.
Generalizability of the Findings
RAJAGOPALAN: How generalizable are some of these results outside of the specific factory or even outside of some of these districts and neighboring districts in West Bengal? The reason I ask is that the priors about other religious groups and the interaction with other religious groups, and inter-religious tensions, they are so specific to location, to urban versus rural setting, even specific blocks and neighborhoods in urban settings. How do you think about generalizing some of your findings from this particular factory firm to other kinds of settings beyond food manufacturing and beyond this district?
GHOSH: Yes, I know, that’s a great question. I think this is a general and fair criticism of works that look at a very specific setting, and generalizability is, of course, important. What I can tell you about this project is, in terms of the theory and in terms of the sample of workers—and especially in terms of the theory—I don’t make any assumptions, or the priors that I assume about Hindus and Muslims are super location-specific.
The asymmetry that I assume—that Hindus do not have enough contact with Muslims, whereas Muslims do—is also true for most of India. Maybe even start there because the proportion of Muslims in West Bengal is actually higher than the average across India. In terms of the results and the representativeness of the firm, what I actually do is I also study firms in and around my partner firms. I look at their hiring policies, the religious composition of the staff in these other factories in the area, and I argue in the paper that my partner firm is representative of the average factory in the district where I worked.
Furthermore, as I was briefly alluding to you before, to really see the generalizability of my results, whether people outside this firm can even predict these results, I actually surveyed more than 100 production supervisors across multiple different firms in West Bengal. I was fairly surprised to see that, not everyone, but a fair share of workers could actually predict these results, that religious division is more costly in high dependency versus low dependency.
This gives me more confidence that I’m not looking at something that’s super specific only to this firm. In terms of this literature, the first paper that comes to my mind is by Jonas [Hjort] in a Kenya flower plant. He finds similar effects, and he only finds negative effects. From then on, there have been a few papers looking at very different contexts, but looking at generally low-skilled jobs with ethnic diversity, and they’ve also found these effects, of course, in different contexts, different mechanisms.
I think, what this tells you is we need this setting to look at these studies. At least this effect of religious divisions in the private sector is becoming fairly generalizable, even though we need this specific setting to look at it.
Mixed-Case Versus Mixed-Religion Groups
RAJAGOPALAN: There’s some work also done on caste composition, compared to religious composition, as you’re studying. There’s of course work by Afridi, Dhillon and Sharma, which is in garment factories. There’s Matt Lowe’s now-famous paper on cricket leagues with inter-caste composition. What do you see as the similarity and differences between inter-religious mixed groups versus caste groups?
GHOSH: That’s an important question. I do look at the effects of caste as well. That’s not my unit of randomization, or that’s not the main question I’m looking at in this paper, but you get some natural variation in the caste composition of teams from randomization. I actually, I don’t find much going on, whether it would be just homogenously teams with different castes, or the interaction of Hindus who are with Muslims, whether they’re high caste or low caste.
I went in with this prior because, in contrast to Farzana Afridi and coauthors’ setting and even Matt’s setting, caste divisions are not super salient, at least in the workplace in West Bengal. Of course, it matters for residential segregation; in marriage markets they are important. But at least in a factory of workers, I do not find that to be a super important division. The local politics is now changing, so we never know. Maybe it would be interesting to look at in the future. The short answer is I looked at it, but I do not find any effects.
Mixed Groups in Work Versus Non-Work Settings
RAJAGOPALAN: How much do you think some of this changes when you study a factory floor versus, say, a sports team or a league? I ask because when you’re looking at a sports team and intermingling, a win or a loss is a joint win or a joint loss, unlike a factory floor where your wages are your own and evaluations for your promotions and so on are your own. You’re not the owner or the residual claimant. How does that change how we think about these mixed groups in work setting versus non-work settings like sports?
GHOSH: That’s a great question. Even in the work setting, these are still teams. While promotions would determine your individual effort, there’s still complementarities and teamwork that you have a set of common goals that you need to work towards, even at the workplace. I think it’s similar in that setting. Now, an important question is this difference in contact inside the firm versus outside the firm. So can I make sports teams with these workers outside and have the similar effects in the firm?
That would be super interesting to find, particularly because I find this tension. Because you have to bear this cost to integrate, but maybe you can play sports with these workers and have similar effects; maybe you can avoid those types of costs. But that’s not to say that organizing a cricket tournament or a soccer tournament with factory workers is not expensive; that itself is expensive. That also costs you productivity, but I think what is key is whether it’s just pure dislike, in this case, for your teammate, which can be overcome through sports, versus whether it’s really beliefs about productivity.
If it’s really beliefs about productivity, then the contact outside the firm might not be useful. My model suggests that it is beliefs about productivity, but I think that’s a super interesting direction for a future study, and I think I’ve talked to Matt about this. We maybe plan to do something like this too.
Ethics of Field Experiments
RAJAGOPALAN: Do you at all worry about some of the ethical issues at play when it comes to these field experiments, where you’re randomizing and placing individuals—and you don’t intend to—but you could possibly be exposing them to physical or mental harm because of discrimination on the factory floor, or sometimes it’s abuse, sometimes it’s violence? How do we square these concerns as social scientists, while relying on field experiments that can potentially provide a lot of insight?
GHOSH: I take this issue very seriously, but of course, I’m not an expert on this. The project, of course, went through a rigorous ethics process before I actually implemented it, but I think the field gradually is moving towards making this ethics process more transparent and being open about it. I think that’s a great thing. Rather than just an ethics review board looking at it, authors like Dean Karlan and Christopher Udry have suggested that papers now have a structured ethics appendix. You really talk about the potential harm to participants, the conflicts of interest, the misuse of research results, typically the things you would have to submit to an ethics board before you get the approval for the study.
We can have general consensus building on what is acceptable and what is not. I think that’s a great direction that we’re moving towards. What I can tell you personally for this experiment is, I try to be as least intrusive as possible. I did not change anything for the workers that would fall outside the purview of the work contract. I did not make them work more or less. There were no changes to payments, no changes to timing of shifts or any of those things.
Even the teams that I formed, either mixed teams or Hindu-only teams, that’s the naturally occurring team structures at baseline. Again, there’s very few Muslim-only teams because there are few Muslims. I didn’t also want to change that structure, because you might be worried that you suddenly have different structures of team that are formed at baseline than in the experiment. I also made sure that I do not do that. Overall, there was this one-time change to these teams, keeping everything else about their work condition constant, and this is something we agree to in the work contract.
RAJAGOPALAN: That’s the relevant thing, that the workers themselves agree that they might be placed in an individual team or a mixed-group team. The only level of loss of control is that they are being randomized by an external entity, as opposed to choosing themselves to go into one mixed group versus a homogenous group.
GHOSH: Yes, and in the context of this firm, actually, the contract asked them to—they submit in the contract that they could be put in any team according to the requirements of the firm. That’s something that it’s not unusual or a big change that I made in terms of something that they’re not used to at all.
RAJAGOPALAN: Do these firms have—just out of curiosity—some kind of redressal mechanism when these kinds of tensions might crop up or when there might be a lot of discrimination faced on the factory shop floor, or India has not gotten quite there yet?
GHOSH: That’s a good question. In general, I think part of what they try to do is they are aware that there’s these external tensions that can happen and that have these effects internally. One of the facts is that they try to integrate workers. Or I’d say rather, they try not to separate workers. Apart from that, I don’t think there’s organized adjustment mechanisms that take place rather than those naturally occurring amongst people. I think there’s scope for that kind of intervention too, absolutely.
Mining and Political Business Cycles in Conflict Areas
RAJAGOPALAN: I want to switch gears a little bit and talk to you about some of the other research that you’ve been working on, which is also very fascinating. This is your work on political business cycles in India, especially in mining areas where there’s a lot of Naxal activity. Can you tell me a little bit more about that paper?
GHOSH: Absolutely. I studied this paper, which looks at electoral cycles in mineral extraction, so several different proxies of mineral extraction. There’s a large literature that dates back to the ’70s which looks at electoral cycles in economic activity. The first models were developed by William Nordhaus and Lindbeck, who essentially argued that opportunistic politicians would stimulate the economy and increase employment right before elections, and the inflationary effects would be felt after. Myopic voters would not be able to see that.
Since then, the literature has moved to younger and more nascent democracies looking at general infrastructure spending, rural credit expansion, health infrastructure, health spending, et cetera. Typically, the finding is that there is this increase in economic activity or economic spending before elections which then, again, falls. With respect to mining, what I find is the exact opposite pattern.
Whether you look at mineral extraction—so just output or licensing or even accidents and deaths in the fields, which is an important aspect of this economic activity—I find that there is this inverted U-shaped pattern rather, in the sense that after an election, it really goes up; then as this election cycle ends, the next set of elections, they go down again. That’s the stark pattern.
There are basically two things that I find. One is that really, the electoral competition matters. If, as a politician, I’m sitting on a smaller victory margin, my cycle shape is even larger. I reallocate more mining to earlier years in my term, and I really, really do a sharp drop as we approach the next election. And also the conflict matters; the Naxalite conflict, or the intensity of Naxalite conflict in the district, is an important determinant of the shape of these cycles. If it’s a Naxal-affected district, again, the shape is starker in that it goes up more after elections and comes down starkly before elections.
RAJAGOPALAN: I had a follow-up question about the conflict in particular. There is some work on political business cycles in India. There’s the work by Sukhantar on sugar canes in Maharashtra or the work by Kapur and Vaishnav on cement and builder connections. What they find is pretty standard, which is there is a decrease in activity because the funds are basically being diverted to political activity.
Now, those are not in conflict areas. Whereas now there’s an additional question of not only is this a conflict area, but a very large part of the Naxal conflict is to stop the smooth functioning of elections. How does that interact with the political business cycle when it comes to mining?
GHOSH: That’s an important question. You’re right. There’s a paper by Kapur and Vaishnav: With respect to cement, what they find is that you’re right, that they have to allocate funds to politicians, and that’s why they face a liquidity crunch, and that’s how output goes down. With respect to the conflict, it’s established that Naxals want to stop the smooth function of elections, and it’s also established, in previous work, that mining forms a large resource or economic base. Extortion of mining revenues forms a large economic base for the Naxalites.
What I show is that this reduction in mining leading up to an election is strategic in that it’s a way to reduce this resource base for the Naxalites. In that sense, it could affect their funding, and that will, in turn, affect the level of violence. I compare mining districts to non-mining districts, and in non-mining districts the funding base of the Maoists are very different. It’s mostly small traders or businessmen that they extort from, and politicians have less leverage over these more disaggregated funding bases.
I do not find a fall in violence in these areas, typically, because politicians have less leverage over these funding sources of the Maoists. Whereas, mining is something which Sam Asher and Paul Novosad in a recent paper have shown has a large involvement of politicians. I argue that because of their large-scale involvement in this industry, they’re actually able to manipulate activity in a way that suits their electoral agenda, in that they try to reduce violence by minimizing the rebel group’s resource base.
RAJAGOPALAN: Actually, it’s a weird political business cycle which may actually enhance the smooth functioning of elections, while simultaneously dealing with the conflict. That may not be the intended effect, but it’s certainly a byproduct.
GHOSH: Actually, yes. The propensity of conflict if the economic activity increases there, we might see this sort of a pattern.
RAJAGOPALAN: What have you been up to during the pandemic?
GHOSH: The style of the pandemic was very stressful because I had to move my end line survey basically online. The pandemic struck in the middle of my job market paper experiment. I’ve just been watching a lot of shows since then, being at home. I really liked “Ted Lasso,” which I’ve been recently watching.
RAJAGOPALAN: I’ve been obsessed with “Ted Lasso” right now. I’ve completely caught up, which is rare for me, so I’m now waiting for the next episodes to come out as they release.
GHOSH: Every Friday I’m just waiting for the next episode.
RAJAGOPALAN: What else have you been binge-watching?
GHOSH: Just been watching some movies that come up. In terms of shows, I think I was just watching “Manifest,” which was a little dark but I enjoyed it.
RAJAGOPALAN: “Ted Lasso” was perfect pandemic watching. It just puts a smile on your face.
RAJAGOPALAN: Thank you so much for doing this, Arkadev. This was such a pleasure.
GHOSH: Thank you very much for having me, Shruti. It was just a wonderful experience.
RAJAGOPALAN: Thanks for listening to Ideas of India. If you enjoy this podcast, please help us grow by sharing with like-minded friends. You can connect with me on Twitter @srajagopalan. In the coming weeks, we will feature weekly short episodes with young scholars entering the academic job market discussing their latest research.
Also check out our new initiative commemorating 30 years of India’s market reforms at the1991project.com. The 1991 Project is an effort to revive the discourse on growth-centered economic reforms in India by focusing on the economic ideas that drove them. In the coming months, we will publish essays, data visualizations, oral histories, podcasts and policy papers demystifying the Indian economy and the 1991 reforms. You can see all the content and subscribe to our newsletter for updates at the1991project.com.