Monitoring in Small Firms: Experimental Evidence from Kenyan Public Transit2024

I hated this. This is some Big Brother shit, I am not crazy about RCTs to begin with but this is incredibly dystopic imo. L RCT. All the focus is on the firm outcomes, why should I care that rentier prinicpals can't grow this specific type of business? I am of the idea that the mechanism they describe and the environment it necessitates is a very specific sector of the economy. Transfers between principal and agents are welfare neutral, and the increase of the principal contract value while the agent's contract value decreases is framed as efficiency and not distribution. I cannot see how the intervention's gains, if scaled, would drive any sort of growth or justify the former relocation I describe. Monitoring constraints are a problem for firm expansion, but these firms are all in the public transportation sector, where market concentration is particularly undesirable, unless provided for free by the state.

The model is of course mathematically consistent. Two risk neutral individuals, principal and agent, interact repeatedly through a daily relational contract. The driver chooses effort \(e\) and risky effort \(r\), where effort is assumed to increase revenue with no cost in repairs, and risk damages the vehicle. Here already disaster: everybody knows increasing effort can absolutely conduce to damages that could imply repair costs. Agents suffer twice continuously differentiable disutility \(\psi(e,r)\). Stochastic revenue is drawn \(\tilde{y} \sim G(\cdot \mid e, r)\), where this distribution is bounded from below by minimum living income \(w\), and \(G\) is assumed to be twice continuously differentiable with respect to all arguments with density \(g(y \mid e, r)\). Increasing risk or effort increases expected revenue. Revenue available for transfer to the nefarious principal is \(y = \tilde{y} - w \in [0, \bar{y}]\). Here \(w\) is exogenous and already we have embeded ideology as lack of ideology, but everybody knows this false neutrality is capitalist ontology. Risk induces repair cost \(c \sim F(\cdot \mid r)\). This implies \[\frac{\partial \mathbb{E}[c]}{\partial e} = 0\] Because of this assumption repair cost condition FOC differ only in the risk dimension. Would you get into the matatu of a driver that has been driving for 24 hours? I doubt it. They impose an assumption that the owner's marginal willingness to trade effort for risk is always higher than the driver's over the relevant region. Defining the following incentive compatible set: \[\mathcal{I} = \left\{ (e, r) \in \mathcal{S} : \frac{\int_{0}^{\bar{y}} y\, g_e(y \mid e, r)\, dy}{\int_{0}^{\bar{y}} y\, g_r(y \mid e, r)\, dy} = \frac{\psi_e(e,r)}{\psi_r(e,r)} \right\}\] are the optimal bundles of effort and risk that agents chooses for whatever disutility they have. They then define the bliss point to be \[(e^D, r^D) = \arg\max_{(e,r) \in \mathcal{S}} \mathbb{E}[y \mid e, r] - \psi(e,r)\] at which the assumption satisfies the following \[\frac{\frac{\partial \Pi}{\partial e}}{\frac{\partial \Pi}{\partial r}} > \frac{\frac{\partial (\mathbb{E}[y]-\psi)}{\partial e}}{\frac{\partial (\mathbb{E}[y]-\psi)}{\partial r}}\] and thus by construction excesssive risk (compared to a planner's social optimum) is guaranteed, that is \(r^D > r^*\). Then, the owner owner maximizes expected value \(V\) \[V = \max_{e,r,t(y),\,p(t)} \mathbb{E}\!\left[ t(y) - c + \delta V - [1 - p(t(y))] \, h \mid e,r \right]\] subject to some acceptable constraints. This is the sum of expected transfers and continuation value \(\delta V\) minus expected cost of firing and risk. Transfer meaning Principal does not observe \(y\), and recieves transfer \(t(y)\), while agent keeps income \(y-t(y)\). After observing \(t\), principal can choose to re-hire with probability \(p(t)\). Replacing the agent costs \(h\). Driver has an outside option \(u\). The optimal transfer rule is then \(t(y) = \min\{y, T\}\). Agent transfers up to \(T\), the amount that guarantees re-employment. The optimal rehiring rule is \[p(t) = 1 - \frac{T - t}{\delta U - u}\] where \(\delta U\) is agent continuation value, which is destroyed if they lose their job. In the baseline case principal cannot punish risky behavior because they cannot observe risk, only transfers. With the introduction of Palantir style technology the principal can recondition the rehiring rule based on ton only effort but also risk. Thus the optimal rehiring rule given under monitoring becomes \[p(t,e,r)=\begin{cases} 1-\dfrac{T-t}{\delta \,\overline{u}_M - u}, & \text{if } e=e_M \text{ and } r=r_M \\ 0, & \text{otherwise} \end{cases}\] with owner mandated \((e_M, r_M)\). This reduces risk. The model predicts higher efforts, profits, and lower repair costs. We will discuss welfare implications shortly.

The following regression is run on daily panel data \[y_{id} = \alpha_d + \tau_{r(i)} + \sum_{m}^{6} D_{im}\beta_{m} + \mathbb{X}_i' \gamma + \varepsilon_{id}\] I will copy and paste the main results: By month 6, treatment drivers spend 1.45 (9.8 percent) more hours on the road per day. This is a significant increase in a context where drivers already work ­ 14-hour days. Treatment drivers are approximately 400 meters closer to the licensed route than control drivers throughout the study. We also see the number of kilometers increase by 13 kilometers per day on average. Taking the average gains over the study period and extrapolating to the full year (assuming matatus operate 25 days a month), owners can expect a $1,200 increase in annual firm profits. We find weak statistical evidence that treatment firms are more likely to grow their business than control firms. The effect of the tracking device on driver ­ take-home pay is ambiguous. The impact depends on how revenue changes and how much the owner adjusts the target. End of copy paste. They then estimate the model with GMM, and find that They estimate that the driver's contract value decreases by $19.7 (SE $13.8). Meanwhile, owner contract value increases by $82 (SE $133), implying a total welfare increase of $63 (SE $137). These calculations are uniformative.

The normative implications rely on the fact that whatever welfare increases from robbing workers then some consumer benefit from risk reduction has to apply but this is a particularity of public transport and if this RCT would be applied to a different sector the monitoring welfare effects are incredibly worrying for the working class. In fact, the fact that the effort increases so much could actually be concerning, and the authors do mention possible safety externalities. They say in the appendix that a motivation was to see if this technology could improve safety. If workers end up overworking the effect is not clear. The authors are incredibly defensive about welfare dimensionality for drivers, as if relationship with the principal was comparable with your wage, give me a break. The survey that they run and its results imply self selection, and possible coercion is of course ignored. This is hyperreal experimentation with the subjugated, and once more capital exerts its tentacles through the intelligentsia to justify exploitation. The defense is that in this particular case monitoring reduced risky behavior, but insofar as the powers that be are concerned, this is green light to do things NSA style.

The model codes the driver's preference for risky driving as a wedge to be corrected, as if the bliss point were pathology rather rational adaptation to subsistence. The social planner welfare benchmark is only normatively appealing if the actual relevant externality is repair costs. If a companion study found that there was no reduction in accidents, it is hard to see how this is anything other than a formal restatement of the owner's objective function dressed in utilitarian clothing.