Institutions, Norms and Worker Decision Making

Individual decision making is shaped by a combination of internal and external constraints. While a lot of research has focused on external constraints (such as institutions and social norms) on an individual’s behavior, there has been relatively little focus on the importance of internal constraints. Such internal constraints may be ideas of morality, self-image and guilt that reflect some combination of internalized institutional and social norms as well as moral concerns. In this research paper, we examine the relationship between internal constraints and external constraints, in order to throw light on some of the behavioral roots of corruptibility.

A central focus of our investigation is the examination of whether internal constraints differ from and are independent of the quality of institutional enforcement and prevailing social norms. Accordingly, we will examine the impact of such constraints on the productivity, corruptibility and decision making of individuals in the workplace. We will conduct a couple of related (but conceptually distinct) experimental interventions in the same setting. Our plan is to describe these results in possibly two papers. Whether the results are presented as one or many papers depends on a number of factors, but we have in mind that intervention (iv) could be distinct from the rest if the findings. We could equally see combining the results into one paper with a broader scope.

In late Summer 2019 and early Fall, the SNE International Organization helped set-up a data entry firm in Arusha Tanzania to run several pilots and conduct surveys. Using the information collected, over the next several months SNE has/will recruit individuals as temporary day workers to complete data entry work. Each worker completes a data entry task that is assigned to them by a supervisor. Together, the supervisor and the worker form a team. The supervisor belongs to two different teams (i.e. assigns works to two different workers). In each team the worker completes a series of data entry tasks under very different institutional environments. The aim of the experiment is to examine productivity, reporting and accuracy as well as the possibility of side payments under alternative payment and monitoring conditions. As part of this ongoing data collection effort we are also conducting a Pope and DellaVigna (2018) component that assesses the ability of other non-participants to forecast the impact of alternative treatments on worker behavior in Arusha, Canada and (possibly) Burundi. Given Covid-19 related delays, we expect (hope) the data collection for the field work (including data collection / data entry) to be completed by December 2020.

Within each team we randomly vary working conditions to examine whether workers and supervisors would collude to earn more money from the project. In particular, the main things we varied was the institutional environment (i.e. whether workers expected monitoring or not). In addition, we also vary the incentive structure on other dimensions that have the potential to weaken internal constraints on worker-supervisor performance and behavior.

We supplement this information with two surveys. First, is that on agreeing to be temporarily employed by the firm, all supervisors and workers filled out a survey to elicit background information as well as a measure that provides us with a measure to gauge willingness to engage in corruption as an individual (i.e. weak internal constraints). We also collect a similar lab-based altruism measure from the supervisors. As part of the experiment, all subjects then complete their forms. Finally, after the work is completed, they fill out an exit survey that also elicits their beliefs about corruption.

We hypothesize that there will be heterogeneity in the extent to which workers and supervisors comply with the instructions that they have been given. We examine the hypothesis of whether more trust strengthens internal norms against corruption. Accordingly, we examine whether or not teams who feel trusted engage in less corruption. At the same time, we would like to examine whether there is a corruption contagion effect, where being teamed together with a more dishonest individual may weaken internal constraints and increase the level of corruption. We hypothesize that the pay structure likely matters for corruption as well, in the spirit of Olken (2010).

Environment: All prospective temporary day workers will be randomly assigned to work as either a supervisor or a worker. Each team consists of one supervisor and one worker, though each supervisor belongs to two (randomly selected) workers. All performance metrics (including monitoring and compensation) are conducted by manager/enumerators who are helping implement the interventions listed below.

All workers are supposed to complete a data-entry task that pays them at different rates: low, medium or high. Some teams get either only the low and medium tasks, while other teams are allocated all three tasks. The supervisor allocates the tasks to workers using a pre-specified rule that uses a randomization device. Supervisors are compensated on a fixed rate basis and workers on a piece rate basis.

We have the following interventions.

(i) Imperfect Enforcement of Contract:
All workers and supervisors are given a contract that includes a reward and bonus structure. All worker-supervisor teams are informed that performance will be monitored, subject to the time constraints of the managers. Any supervisor or worker team observed to be not following instructions will be penalized. However, in practice we will influence perceptions of all worker-supervisor teams that the enforcement of contract will be imperfect and there will be low monitoring.

(ii) Low monitoring/high trust intervention.
One group of subjects are informed that they are trusted to follow instructions and do their work without any monitoring. Managers will take actions to affect perceptions of all supervisor-worker teams accordingly. For instance, managers will be visibly absent and not undertake any monitoring of the teams.

(iii) Contagion:
This intervention (embedded in our research design) is the randomized allocation of workers and supervisors across teams helps throw light on whether an individual’s internal constraints on behavior (be it productivity or corruption) are “contagious”. In other words, we examine whether poor compliance with the instructions of the experiment (under different institutional-incentive environments) results in contagion across some workers or supervisors.

(iv) Rationalization of poor compliance:
These (very distinct) set of interventions examine whether providing workers with possibly irrelevant information, may weaken internal constraints on behavior and lower team compliance. Subjects may find it easier to rationalize their poor compliance – if we make salient (and/or increase ‘availability’) of reasons that allow them to not comply with instructions without hurting their self-image. Two distinct interventions are examined: (a) Make a set of workers aware that a subset of workers are being paid a different amount for a similar task (b) introduce all three tasks (with corresponding low/medium/high compensation rates).

As outlined previously, the above set of interventions are aimed at examining two distinct set of conceptual points. The Interventions (i)-(iii) will form the core of the first paper. We have not yet decided on whether (iv) should be a stand-alone project or be integrated with the first set of interventions.

2019-08-30

2020-09-09

In our main project, our primary interest in examining the likelihood that a worker (and worker-supervisor team) is corrupt. This likelihood of a team being corrupt can be inferred from the distribution of the number of low, medium and high pay forms that each worker complete. Using a variant of Fischbacher and Follmi-Heusi (2013), we can compute the ex-ante probability that an individual was honest. We do this do by examining the gap between the reported number of doubles each worker reports and the ex-ante likelihood of such an eventuality (if the worker were honest). This generates a main measure as a function of the number of rolls a worker completed, as well as the distribution of doubles reported.

Relatedly, in our ancillary project we examine whether individuals are more willing to engage in corrupt behavior if they find it easy to rationalize and justify their behavior. One way we assess this is by examining whether subjects prefer to engage in small corruption (opt for medium pay forms) rather than large corruption (high pay forms), even though the expected return from large corruption is much higher. A second way in which we assess this is by examining worker response to exposure to information about compensation rates earned by fellow workers.

We also plan on examining some secondary outcomes that are a function of different economic/institutional environments. The two main measures we will be able to examine are quality and productivity. Quality can be measured by the degree of inaccuracy in the data-entry task completed the workers. This low quality could be the result of being poorly motivated or due to an effort to maximize their own payoff (i.e. put in random answers to get a form completed quickly). Given both these possibilities, it is not ideal as a main outcome but could still be consistent with corruption.

Similarly, we can examine productivity. This would be a function of the count of the number of forms completed. As with the quality measure, this could either be due to more or less effort or cheating.

Other outcomes:

We do intend to investigate the balance of out treatments, and plan to run regressions that include the treatments as outcomes to determine whether our various treatments are correlated with observables.

We also plan to collect information on beliefs about corruption, in a post-experiment survey. One item of interest could be whether reported beliefs about corruption are motivated and change depending on whether people were more or less corrupt in the experiment. Accordingly, we plan to investigate whether our treatments are associated with differences in reported beliefs about corruption in the real world.

Finally, we are interested in a prediction exercise in the spirit of DellaVigna and Pope (2018). In our case we plan to ask people in Tanzania (and possibly Burundi) what they expect reactions to our interventions to be. We expect to implement an analogous survey in the West (i.e. Canada or England).

Our planed estimation strategy is straightforward. We plan to control for various observables in order to reduce residual variance, but otherwise the estimation is straightforward due to the randomized nature of the various treatments. In general, we plan the following as our main estimating equation:

Outcome_{it} = B_0 + B_1 T_i + Gamma X_{it} + epsilon_{it}

The outcomes and treatment (T) are as described above. We plan to include a number of controls in X, including how much the supervisor altruistically cheated in the pre-experiment survey (measured by number of doubles rolled out of 10). This controls for the possibility that they may collude with workers altruistically, since it would be very difficult to experimentally prevent side-payments being made outside of the workplace. We also use our pre-experiment survey measure of cheating to control for the propensity to cheat by a player; the worker’s employment status and yearly income; correctly answering raven questions, age, age squared. We also plan to control for village, ethnicity, religion of both the worker and supervisor, whether the session was a morning or afternoon session, as well as time and room fixed-effects.

Combination of random sequence invitations and within session dice rolls.

individual

No

500

aiming for 750 individuals: 500 workers; 250 supervisors

8 treatment arms: 90-100 per treatment arm (incl. control)

Appendix Describing The Bujumbura Experiment