This is the seventh in this year’s series of posts by PhD students on the job market
Digital technologies are increasingly accessible in lower-income countries, potentially offering substantial productivity gains, but there’s a catch. They often come bundled with a level of transparency, or “observability,” that can be a double-edged sword. For example, digital payments—often the entry points to digitalization—allow business owners to track transactions more easily, which reduces moral hazard and may improve efficiency. But for employees, digital payments have mixed effects. On one hand, it reduces the hassle and risks of handling cash. On the other hand, it involves monitoring, which may reduce some employees’ ability to avoid work and lead them to resist adoption. This dilemma extends to any technology embedding observability as a byproduct, like employee-tracking softwares or mobile apps raising data privacy concerns from users, and could contribute to the “digital divide” across businesses.
Despite how common this issue is, we know little about how much this tradeoff between observability and adoption really matters. In my Job Market Paper, I study two questions:
1) How do digital technologies impact information frictions, contracting, and firm performance?
2) To what extent might these impacts hinder technology adoption by agents?
I built a partnership with the largest mobile money and payment provider in Senegal, Wave, to design a digital payment technology for the taxi industry and conducted two field experiments with over 2,700 taxi owners and drivers.
Digital payments significantly reduce cash-related costs, while transaction observability improves firm efficiency…
The Senegalese taxi industry exemplifies common principal-agent challenges faced by small firms in lower-income countries. The typical arrangement involves a car owner (employer) and a single driver (employee) linked by a relational contract. The driver keeps any revenue exceeding a rental fee paid weekly and sometimes receives an upfront payment from the owner. Due to limited liability, drivers can default on the rent by claiming low earnings for the week. Importantly, the owner has no way to observe whether this is due to bad luck, lack of effort from the driver, or whether the driver misreports revenue. This creates scope for moral hazard in both driver's effort and reported output and allows drivers to capture informational rents, contributing to inefficiencies common in informal arrangements. Default may lead to the (costly) termination of the relationship to mitigate moral hazard.
I randomly vary the observability of drivers' digital transactions to taxi owners, allowing me to quantify its impact on the contractual frictions and employees' adoption decisions. In the Impact Experiment, I identified drivers willing to adopt and varied two dimensions: (i) access to digital payment technologies among 1,891 drivers and (ii) transaction observability for taxi owners among 613 owner-driver pairs. Taxi owners were randomly assigned to one of three observability levels:
- Granular Observability (20%): the owner has access to the driver’s digital transaction history, including timestamps and transaction values. This aims to provide a signal of effort and output.
- Coarse Observability (20%): the owner receives a daily SMS summary of digital collections up to a set limit. This aims to let the driver signal low-output periods while retaining some informational rent by not revealing the full transaction history.
- No Observability (20%): this aims to estimate benefits of digital payments unrelated to observability.
- Pure control (40%): No technology was provided to drivers that were placed on a waitlist.
The observability treatments aim to test core contract theory predictions: that imperfect information on the agent’s work can reduce moral hazard problems, increase agent effort, and improve contract efficiency.
I used multiple data sources to measure outcomes. First, I conducted five rounds of surveys with owners and drivers over two years, achieving a 95% follow-up rate among pairs. The key innovation of these surveys is that they measure informal contracts and how they change, providing a rare opportunity to study contracts in developing economies, where such contracts are typically verbal. Second, I measured driver effort through mystery passenger audits, during which surveyors stopped nearly 8,000 taxis across Dakar, thereby observing when drivers were actually on the road. Third, I used daily transaction data from the payment company to measure driver-level transaction patterns.
Figure 1. Impact Experiment: Transaction Observability Improves Efficiency
I find that digital payments are a valuable technology as it significantly cuts drivers' cash-related costs (e.g., small-change shortages) by half and serves as effective monitoring tools even with partial digitalization of transactions (about 13% of revenue). Under Granular Observability, drivers exert more effort: they are seen on the road 34% more often in mystery audits, process more digital transactions, and report working more hours. Drivers are 31% less likely to default, and owners adjust contract terms by being 18% more likely to provide an upfront “salary”, on top of the rental contract, to compensate for drivers' increased effort. Owners were 16% more likely to remain with their driver after nine months (see Figure 1). Coarse Observability had positive but insignificant effects on worker retention after two years.
Overall, Granular Observability raises owners' profits by 8%, though this effect is not statistically significant, and slightly increases drivers' profits. These results, in line with the predictions of a simple theoretical framework I developed, are concentrated under Granular Observability, which emphasizes the importance of moral hazard in effort, now alleviated with the technology.
… but observability acts as a barrier to digital technology adoption
In the Adoption Experiment, I measure how observability might impede technology adoption by drivers. Taxi drivers needed to provide their employers' contact information to adopt the technology. I followed up with drivers who refused to share this information to quantify the role of observability. I re-offered the technology to drivers and randomly varied whether the taxi owner could observe their digital transactions. Unlike the first experiment, drivers received this information before making the adoption decision.
Figure 2. Adoption Experiment: Observability is a Barrier to Technology Adoption
I find that observability is an important barrier to technology adoption, especially for the worst-performing and poorest workers. Initially, 50% of drivers did not want to adopt the technology, citing various privacy concerns for not sharing owners' information. I compare drivers in the ``Impact Experiment'' to reluctant drivers in the ``Adoption Experiment.'' Reluctant drivers are 83% more stressed at work about making rental payments, perform significantly worse on the job, and are poorer and less likely to have attended primary school. When I randomly removed the feature that allows owners to observe transactions, adoption doubled—with effects more than twice as large for the worst-performing and poorest—revealing the importance of observability in driver's adoption decisions (see Figure 2).
Implications for Policymakers and Technology Designers
To interpret these findings, I developed a theoretical framework within a relational contract setting where the principal (employer) cannot commit to contract terms once the agent (employee) adopts the technology. Using this framework and experimental effects, I quantify drivers' disutilities of work and estimate two key policy counterfactuals. First, a policy mandating adoption would increase overall welfare but decrease the welfare of low-ability drivers by 12%, who are induced to exert higher effort upon adoption, thereby increasing welfare inequality across and within firms. Second, redesigning the technology to remove transaction observability for employers would lead to Pareto improvements: all drivers could then adopt the technology and overall welfare would increase, even though the information frictions would remain.
These findings show that while digital technologies can drive business growth and reduce information frictions, their adverse effects on contracts may deter adoption. Since observability hinders adoption for many workers, a real-world impact of this research was the payment company’s decision to set non-observable transactions as the default to boost adoption in the taxi industry, though this decision also maintains contract inefficiencies. These findings highlight the crucial role of information embedded in digital technologies, as it magnifies gains for adopting firms and reshapes organizational structures but can deter initial adoption.
Deivy Houeix is a PhD student at MIT.
Join the Conversation