Experimental Design
Research questions:
(1) Which incentive schemes do managers and workers perceive to work best and why?
(2) What is the extent of disagreement between managers and workers about their preferred contracts?
(3) What incentives work best in practice?
(4) How well do managers' and workers' predictions align with actual outcomes? What leads individuals to be able to better predict what contracts will work well?
Our study is composed of two main parts. The baseline survey will provide answers to the first two research questions: managers' and agents' preferred contracts and the extent to which they agree with each other. The RCT will provide answers to the last two research questions: the actual impact of various contracts and whether managers are sophisticated about these impacts.
To implement both the survey and RCT feasibly, we will focus on a small set of bonus schemes. These schemes represent key structures emphasized in the literature and are widely used in practice. Additionally, we have confirmed with the service provider that these schemes are logistically implementable (we have verified this through a technical pilot). Our selected schemes include:
• Individual Performance-Based Schemes:
(1) Simple Linear Contract (Status Quo): Workers receive bonuses for every unit of output.
(2) Threshold Contract: Workers receive a flat payment upon reaching a predetermined threshold.
(3) Pure Franchising Scheme: Akin to two-Part Tariff, workers receive a boost in the per-unit commission in exchange for an upfront fee.
• Relative Comparison Schemes:
(4) Tournament Scheme: Workers compete in tournaments where top performers in each locality receive an additional bonus.
• Control Scheme:
(5) Flat Bonus/week Contract: This contract is included (i) to check whether respondents pay attention and understand the options presented and (ii) to evaluate the implications of the provider MTN switching to non-performance-based compensation schemes.
#### Perceptions of incentive contracts [RQ1-2]
To answer the first two research questions, we will first conduct baseline surveys with managers and agents to elicit their most preferred incentive scheme and rank all incentive schemes in terms of revenue maximization.
#### Performance under incentive contracts [RQ3]
For the third research question of estimating the causal effects of various contract schemes, we will conduct a large-scale nationwide randomized controlled trial (RCT) using the same study sample. The RCT will involve randomizing the implementation of different contract schemes, implemented strictly according to managers’ rankings or preferences – i.e. how managers perceive the various contracts to work, from worst to best.
The randomization will occur at the local market (community) level, meaning every agent (and manager assigned) within a community receives the same treatment.
Specifically, the communities will be randomly assigned to the various contracts in proportions that reflect the managerial rankings. For example, if 30% of the time managers rank the Threshold Scheme to maximize revenues, then 30% of communities will be assigned to the Threshold. Similarly, if 10% of the time managers rank the Flat Bonus/week Scheme to maximize revenues, then 10% of communities will be assigned to the Flat Bonus, etc. However, for statistical power, in practice we only preserve the managerial rankings and then unequally assign communities according to 25.0% (for the best or top ranked contract), 22.5%, 20.0%, 17.5%, and 15.0% (for the worst or bottom ranked contract).
For fair comparisons, the contracts will be designed to ensure expenditure neutrality for the firm, such that the average total bonus payment is constant across schemes.
#### Connecting perceptions and performance under incentive contracts [RQ4]
By linking the administrative firm data to survey responses, we can answer the fourth research question: whether managers and workers accurately assess the effects of different contract structures.
###Analysis #1: Predicting Optimal Contracts: Managers’ vs Agents’ Preferences and Rankings?
We will combine `descriptive analysis’ and `machine learning’ techniques to examine views about (i) the value of monetary and non-monetary incentives, (ii) adequacy of the current agent compensation scheme, (iii) their ‘most preferred’ compensation scheme if they had to choose, (iv) the compensation schemes that would ‘increase revenues = profit for MTN’ if they had to predict. We will use this to assess whether managers preferred scheme ‘is always’ the profit-maximizing to MTN. This exercise will enable us to identify the most important factors (demographic and environmental) associated (i) with managers’ or workers’ preferences, and (ii) with the disagreement between managers and workers about their preferred contracts, while controlling for market and firm-level characteristics.
###Analysis #2: Which Contracts are `Best’? How Do they Compare with Managerial Predictions?
In assessing which contracts are best, and in connecting this to managerial (and agent) predictions, we will focus on two conditions: IR (Individual Rationality) and IC (Incentive Compatibility).
• IR: We will look at (i) uptake vs decline to participate rates across contracts (extensive margin), (ii) stayers vs droppers and how long agents stayed across contracts conditional on taking up the scheme (intensive margin), and (iii) the reasons for declining/dropping off.
• IC: We examine (i) worker output, (ii) worker inputs, and (iii) worker noncompliance outcomes overall and for only agents who stayed throughout the intervention’s period, while accounting for differential attrition.
• IR + IC jointly: We will look at (i) worker output, (ii) worker inputs, and (iii) worker noncompliance outcomes overall and for only agents stayed throughout the intervention’s period, while accounting for attrition.
###Analysis #3: Heterogeneity (Mediation Analysis) and Nature of Selection?
(1) Assess heterogeneity of IC outcomes based on:
*Market conditions: (i) level of past economic activity and output/payoffs both at worker level and community level; (ii) variance in past economic activity and output/payoffs (at worker and community level); (iii) rural/urban; (iv) number of agents in a community and within the agent firm; (v) whether the retail business is operated by owner/non-owner.
*Manager characteristics: (i) managerial level, (ii) experience.
*Agent characteristics: (i) gender, (ii) experience, (iii) household income.
(2) Assess selection based on differences in uptake/staying in assigned contracts (IR outcomes) and treatment effects separately for:
*Managers (if they were assigned their preference vs not),
*Agents (if they assigned their preference vs not), and
*Both mangers and agents (if both assigned their preference vs not).
Since the number of experimental communities relative to the total number of communities covered by the “typical” manager is small, we expect selection on effects on the manager side to minimal.
We will analyze whether uptake and staying are related to the market conditions and agent characteristics in (i)
(3) Test whether managers can predict which contracts will work best for agents within vs outside their territories, and see if it relates to manager characteristics
(1)-(3) help understand what leads managers and workers to be able to better predict what contracts will work well, while revealing potential correlation in preferences and information gaps between agents, lower-level and senior managers.
###Analysis #4: How Far Away are We from the IR Constraint?
As a follow-up, we plan to implement a willingness-to-accept (WTA) exercise that will elicit, from the agents that declined to participate or dropped off from their assigned contracts during the intervention, (i) why they declined/dropped and (ii) how much they need to be paid to be moved back to their assigned schemes.
###Analysis #5: Changes in Fraud of Agents and Consumer Protection Outcomes?
Look at whether there is differences in noncompliance (invalid transactions), service quality outcomes, misconduct or overcharging based on assigned contract (and demographics/market condition).
###Analysis #6: Changes in Psychological Wellbeing?
Evaluate whether there are differences in mental health or distress, subjective wellbeing, cognition, (domestic) violence, and risk/time preferences based on assigned contract (and demographics/market condition).
###Analysis #7 (Separate Work): Which Contracts Constrain/Promote Adaptation to Climate Shocks?
Evaluate whether there are differences in the effects of unexpected weather on firm outcomes (sales revenue, labor supply, the number of customers, etc) with and without weather forecasts based on assigned contract (and demographics/market condition).