Experimental Design
Our sample is drawn from a census of small firms in Aburi township in the Eastern Region of Ghana as of May 2022. The census sampling strategy located (1) every permanent structure with signage of business, including mobile businesses and those within a household, and (2) every 5th household without signage, in order to identify owners of a business in operation in May 2022. This sampling strategy identified 1,595 firms. Census data collection included basic measures of firm performance and firm (and owner) characteristics, as well as descriptors of firm's main product category.
From the full census sample, we then retained firms that met the following inclusion criteria: (1) at least 3 other owners in the same product category, (2) an active mobile money account, (3) a valid contact phone number, and (4) consent to be included in future studies. 164 firms are dropped based on these criteria and our final sample is 1,431 firm owners.
We stratify our randomization based on product category, which is the main product of each owner's primary business. The reason for this stratification is that we want to match owners that compete for a similar customer base into pairs for the treatment, in order to induce potential consolidation in the trade. There are 29 effective strata, the smallest of which include four firms and the largest of which includes 284 firms.
This stratified randomization at the firm level assigns firms owners to treatment or control as follows. First, owners within the same strata are randomized into groups of four. Then within each group, two of the four owners form a treated pair, and the other two form a control pair. For strata of which the number of owners divided by four has remainder, if only one that owner is assigned to control, if two both are assigned to treatment, if three the first two are assigned to treatment and the final to control.
Each treated pair of owners are given the offer to collaborate with each other by providing a positive number of days of labour in their partner's business on a working day for a flexible number of days negotiated between the partners, and the research team will pay the partner providing labor a daily rate of compensation of GHC 20. The partner receiving labor is required to report to the research team the exact day, time and location for planned collaboration days for the coming week no later than 5pm on Friday. The program runs for six weeks and each pair have up to 30 collaboration days to claim a subsidy against. Control firm owners are not given their partner's contact information, nor are they informed of the program details.
All treatment assignments were communicated with treated pairs in August 2022 during the baseline survey. During the baseline survey and ahead of the treatment announcement, both treated and control owners are asked whether they know their partner.
In addition to the census data regarding May 2022 and baseline data immediately prior to the intervention, we plan to include multiple survey follow-up rounds that could be stacked for analysis, pending funding and logistics. We will track attrition rates at followup and use Lee bounds if attrition imbalance is found across treatment groups.
We are pre-specifying a limited number of outcome variables that attempt to directly measure basic characteristics of firm performance. Given this structure, we do not think multiple hypothesis testing adjustments are necessary. However, in some cases our survey uses multiple questions for a single measured outcome, where we will use an index instead of any individual questions (e.g. products, mental health).