UC Berkeley

Prosociality is a core element of human behavior. A variety of explanations have been proposed for prosocial behavior, such as outcome-based altruism, giving for warm glow, and reciprocity. While these models can explain basic patterns, they have trouble explaining cases of avoidance of costless information, as documented in the moral wiggle room experiments (e.g. Dana et al., 2007; Grossman and van der Weele, 2016).

To explain these and other findings, Bénabou and Tirole (2006) introduce a model of prosocial behavior in which agents want to signal to others that they are prosocial. In this model agents behave prosocially not only out of intrinsic care, but also out of reputational concerns. That is, agents take costly actions to signal to others or to themselves that they are the kind of person that cares about others. Agents might then anticipate that reputational concerns will compel them to behave more prosocially, and therefore actively avoid situations in which their contribution is more visible.

In chapter 1 of this dissertation, I propose a new approach to distinguish signaling models from traditional social preference models. The main idea is to study a setting in which agents contribute to a prosocial cause while also receiving a personal bonus for reaching a threshold level of contribution. If agents are motivated by neoclassical incentives, outcome-based altruism, warm glow giving or social norms, then such a bonus incentive induces bunching at the bonus threshold. In contrast, if agents want to signal their prosociality, then no bunching occurs in equilibrium. This prediction arises because with any bunching, the most intrinsically motivated buncher can marginally increase their contribution to separate themselves from less intrinsically motivated bunchers, and thereby receive a discrete signaling benefit at a marginal cost.

Moreover, signaling models predict that increasing the bonus amount induces anti-bunching, that is an increase in the contribution level strictly above the bonus threshold. Anti-bunching arises because increasing the bonus amount lowers the intrinsic motivation of the marginal buncher. Since all agents obtaining the bonus still want to distinguish themselves from lower types, they need to respond by increasing their contribution. This prediction distinguishes signaling models from several alternative models, that predict a zero response from inframarginal agents.

In chapter 2 of this dissertation, I build on these insights to design and conduct a real-effort online experiment. In the control group, participants choose to complete up to 38 transcription tasks with a return to charity of 8 cents per task, but without any personal benefit. In the 40c-bonus group, the return to charity is the same, but participants also earn a personal 40c-bonus for completing 15 or more tasks. In the $1.20-bonus group the bonus for completing 15 or more tasks is $1.20. The comparison across these bonus groups allows me to test for the key no-bunching and anti-bunching predictions.

I combine the bonus component of the design with a second, visibility component. Making effort more visible to others should amplify the motivation to signal. Thus, in a crossed randomization, half of the participants are asked to create a personal Badge. The badge displays tasks completed, total donation amount raised, total personal gain and the bonus incentive scheme, together with a picture of the participant that they take using their webcam. After completing the experiment, each participant’s badge is shown to at least one other participant, who is then asked to judge the participant’s generosity. Participants are made aware of this when creating the badge as well as during the transcription task.

The experiment provides evidence for signaling motives: If a participant’s effort is private, then introducing a 40c-bonus incentive for completing 15 prosocial tasks increases the share of participants completing 15 or more tasks from 19.8% to 51.7%. This 31.9 pp increase is accompanied by a 3.9 pp increase from 18.3% to 22.2% in the share of participants completing 17 or more tasks. This is the baseline effect, that captures a participant’s motivation to signal to themselves or the experimenter.

If a participant’s effort is visible to other participants, then introducing the same 40c-bonus incentive increases the share of participants completing 15 or more tasks from 22.1% to 54.9%. This 32.8 pp increase is now accompanied by a 9.4 pp increase from 21.0% to 30.4% in the share of participants completing 17 or more tasks. Since 1.1% complete 15 or 16 tasks without a bonus, this implies that at least (9.4−1.1)/32.8 ≈ 25% of those responding to the bonus incentive exhibit signaling motives strong enough to complete at least 2 additional tasks. I expand on this basic finding with additional tests. In sum, the chapter provides a proof of concept for anti-bunching as a test for signaling motives.

In chapter 3 of this dissertation, in joint work with Garret Christensen, Zenan Wang, Elizabeth Paluck, Nicholas Swanson, Edward Miguel, and Rebecca Littman, we investigate an example of prosocial behavior in the field. Practicing open science is an inherently prosocial act as it allows fellow researchers to learn from and build upon existing research.

We conduct an incentivized survey of active social scientists to study the adoption of open science practices (posting data, code and study materials online, pre-registering studies, hypotheses, and analysis prior to conducting a study). We find that as of 2017, over 80% of scholars had used at least one open science practice, rising from one quarter a decade earlier. We also find similar attitudes toward research transparency between older and younger scholars, but the pace of change differs by field and methodology. Patterns are consistent with most scholars underestimating the trend toward open science in their discipline.