The Cost of Partial Altruism: Oriana Bandiera at Princeton

Social incentives explain why some people work harder than others, but that’s not always a good thing.

As a new post-doc at Princeton in the Paluck Lab, I’m excited to be liveblogging talks at the Center for Behavioral Science & Public Policy.

Speaking today is Oriana Bandiera (@orianabandiera), professor of at the London School of Economics. A fellow at the British Academy, Bandiera isa leader in development economics and directs the STICERD centre for research and policy development. Today she talked about social incentives.

Oriana Bandiera at the World Economic Forum (CC BY-NC-SA 2.0)

Understanding Social Incentives

When economists try to theorize organizations, they typically propose the principal/agent model, a way of formalizing the key problem that bosses have: the boss (principal) would like the employee (agent) to work hard, while the agent would like to go to the pub. When you pay people, you give the agent something they want in exchange for something the principal wants. This is based on two things: that agents care about pay, and that whatever else they care about does not interact with pay.

What else do people care about? Organizations are not made by one agent; if you only have one agent, you don’t need an organization. The point of organizations are to bring together multiple people’s contributions to provide goods and services to other people. Agents might care about these other people, and the way they care might interact with other incentives.

Social incentives theorize the way that an agent’s preferences impact on others’ welfare and the extent to which those preferences affect her choices. These social incentives shape the agent’s choice of effort and also interact with financial incentives. Today, Bandiera promises to share a series of studies she recently recently reviewed with Nava Ashrav on Social Incentives in Organizations. The paper provides a taxonomy that is based on two things: the nature of social preferences and the the agent’s task.

People often cooperate with the people they care about, something that can hurt or benefit society

Generalized and Partial Altruism

Economists generally model social preferences as the value a person gives to others’ outcomes in a utility function, says Bandiera. When someone cares about other people equally, we call it generalized altruism, though many people exhibit partial altruism. In organizations, people often cooperate with the people they care about, something that can hurt or benefit a firm or society at large. When it hurts, we call it collusion. When it benefits society, we call it cooperation. Both stem from the same preferences, but depending on the context, they can have very different outcomes.

The nature of the task also plays an important role in incentives, says Bandiera. Many jobs require people to generate resources and allocate them between others. For example, managers can take some actions to improve everyone’s productivity (generalized altruism). They can also do things that target specific workers, allocating the “pie” differently between different people (partial altruism).

Social motivations do make people work harder. But not all altruism is general, says Bandiera. When people are partial in their altruism, they create a “targeting bias’’ that leads people to compete with others for those benefits. When we care about some people more than others, that can have negative effects on how much people work when they see those disparate outcomes.

Experiment Designs for Studying Social Preferences

How can we identify the bias that results from differences in allocation from managers to agents? Getting causal knowledge on people’s friends is hard because we can’t randomize who knows each other and who they like–that was what Bandiera thought was the case. She shares a new study where they collaborated with an NGO in Uganda to study the effect of social ties on the delivery of an agriculture extension program aimed at poor farmers. They found that the program is effective at reducing poverty, yet half of the communities didn’t adopt the program. Bandiera wondered if people’s social ties might explain this difference.

The initial agricultural training program was effective on average, but a large number of villages (on the left) didn’t adopt the training. Bandiera did a followup to understand why.

In a followup experiment, researchers asked people to propose two possible agents to provide the agricultural training. They then measure social connections between villagers and the two agents in each of the villages. Researchers then randomly chose one of the two agents as the person to deliver the agricultural support.

In some villages, agents share the same politics and religion with the community, but in other villages, only one of the agents does. Half of the villages received agents that shares their politics and religion, and the other half received an agent that differed in politics/religion. In the study, they observe outcomes for villages where both agents were had a similar identity to the community, compared to villages where the agents had different politics and religion. Comparing the two, they found that people were less likely to accept training from someone with different politics and religion, and that in the villages with different agents, even more people accepted training, but primarily from the person whose values agreed with theirs.

When the two agents delivering agricultural training didn’t share politics/religion with each other, the community adopted agricultural training in much higher numbers than when both agents were similar to the community, but only from the person whose politics/religion agreed with theirs.

What do we learn from this? Bandiera argues that their findings are consistent with other research about parochial altruism: dislike for the other group motivates people to take the training. Does this mean that NGOs should create political and religious rivalries to improve agricultural education? Bandiera points out that the agents were paid very little; perhaps they should just be paid more.

The Effects of Favoritism on Colonial Performance in the British Empire

Favoritism isn’t limited to NGOs or local agents in poor countries, says Bandiera, who describes a paper by Gao Xu on favoritism in the British Empire. From 1854–1966, the secretary of the state had full discretion over governors. Halfway through, that discretion was taken away. Xu collected data from Who’s Who to analyze the social networks of people who were related and who went to the same schools.

Xu found that well-connected governors earned more, were poor performers, funds were misallocated, and that the effect persists; modern countries exposed longer to connected governors during the period of patronage exhibit lower fiscal capacity today. Favoritism of this kind is an example of partial altruism, says Bandiera. This ultimately hurts the British empire’s revenues and performance, she argues.

Modern countries exposed longer to connected governors during the patronage period have lower fiscal capacity today.

The Effects of Religious Preferences on Loans in India

Next Bandiera describes a paper by Raymond Fisman, Daniel Paravasini, Vikrant Vig, and colleagues on favoritism towards same religion groups in banking, based on a study of the state bank of India. They compare the performance of branches with loan officers who share a religion with people in the region. They find that own-group borrowers have better outcomes, and that own-group borrowers still perform better after the officer leaves. Why is this outcome different? Religious similarities aren’t friendships or social relations; they may be an example of local knowledge. These loan officers are paid, and their rotation depends on their performance, so they have incentives to do well.

Can Financial Incentives Crowd Out Social Incentives?

Next Bandiera talks about an experiment on favoritism and financial incentives that she led with managers in a fruit farm. Managers allocate resources to workers, making them more productive and increasing their pay. The same worker earns 10% more when managed by a manager she is connected to. Managers devote more resources to connected workers at the expense of high productivity workers. In an experiment, they look at the effect of giving people a bonus based on the performance of the workers they manage. In that context, favoritism disappears and managers start favoring high ability workers, “crowding out” this bad behavior. In this fruit farm, social preferences created a gap between the agent’s interest and the organization’s financial incentives; an increase in pay re-aligns those interests.

Bandiera and colleagues conducted further studies to better understand this re-alignment. In a study in Zambia, they worked with a public health organization to test the idea of a “female condom,” which was a regular condom sold to women to give to the men in their lives. The organization, which recruits retailers to sell condoms in their shops in Lusaka, wanted to know if the incentives reduced the effort of motivated agents.

To conduct this study, they measured each of the sellers’ motivation for the cause through a donation game invited them to donate to a public health organization. They then allocated the agents to 4 incentive treatments to test whether the agents who were more motivated in the first place work less when they get incentives. They found that (i) non-financial rewards are effective at improving performance; (ii) the effect of both types of rewards is stronger for pro-socially motivated agents; and (iii) both types of rewards are effective when their relative value is high.

Does Emphasizing Financial Incentives Reduce the Quality of Applicants for Community Health Work?

In another study, with the government with Zambia, they worked to recruit community health workers. Rural areas have serious shortages in health workers, because people trained in urban areas often don’t stay in rural positions. In this study, they worked to train people from rural areas and send them back to serve in those areas. The government was worried that if they offered a good salary to people, they would stop getting the do-gooders and get people who were less committed to the work. To test this idea, they randomized the way they recruited people in different districts. In the control group, they advertised the position as an opportunity to do something for the good of the community. Elsewhere, they focused on the career benefits.

The community health study compared people who applied to the career-oriented advertisement (left) with people who applied to the “do gooder” advertisement (right). They also compared village outcomes between these groups.

When policymakers discuss offering material benefits for public goods, they often fear that the median applicants will have better skills, more ambition and lower pro-sociality. This is what happened in this case. But in practice, nobody chooses the median applicant. In this study, they found that by more clearly advertising the resources and career advantages, the civil service gets higher skilled people who are also more pro-social at the upper end of applicants. Over the next four years, emphasizing career benefits attracted candidates who put in 30% more home health visits, brought childbirth at facilities up by 31%, vaccinations up by 20%, and reduced the share of malnourished children by 25%.

Do financial incentives ever undo positive social incentives? Research by Erika Deserrano shows that emphasizing financial interests in job ads can signal that the job is not pro-social and consequently discourage pro-socially-motivated applicants. In studies on team incentives, they have found that if group incentives are emphasized, people try to work with other high skilled people rather than mixed teams and bring down the quality of work overall.

How Do Cultural Differences Influence The Effects of Social and Financial Incentives?

Bandiera notes that most of the positive evidence on the effects of incentives have been done in countries where financial incentives are common. In countries like the United States, individual performance is important. At the other end of the spectrum, being the best can be embarrassing because you make other people look bad.

Bandiera shows us a chart of the correlation between these measures of individualism and how much performance-based pay is used. To unpack the cultural differences between these effects, Bandiera and Fischer ran the same incentives experiment over two days in three countries. In India, a more individualist country, they found results similar to the US and Europe. In Indonesia, they found that the effect was smaller. In Ghana, a highly collectivist country according to the measure they used, they found no effect. This is consistent with case studies finding that multinational companies often struggle to apply similar incentives models across multiple cultures.

To what extent are incentives shaped by culture, or how to incentives shape social preferences and the culture of an organization? “I have no answer to that question, and I might have it in a couple of years.”