Search costs, mobile phones, and agricultural investment: a telephone directory intervention in Tanzania

The primary intervention is a business directory with a description and contact information for approximately 10,000 businesses in 6 districts of the Kagera region. The directory takes two forms: a printed paper directory similar to a traditional print Yellow Pages, and a digital directory that is available via both USSD and smartphone app. Households are assigned to control, paper directory, or digital directory.

2022-11-21

2023-07-01

Our key outcomes are at the household level and fall into the following categories:

1. Communication and search activities, including phone usage, SMS usage, mobile money usage, travel outside the village, communication with firms, number of crop markets/buyers searched, phone-mediated price search, transportation costs, travel time, and ordering goods for delivery.

2. Quantity and price of purchases of agricultural inputs (crop and livestock), inputs for non-farm enterprises, and other consumer goods and services represented in the directory (mechanics, tailors, food away from home).

3. Farm activities and investment, including crop choice, factor usage, acres cultivated, receipt of extension and advice, quantity produced, quantity sold, location of sales, prices received.

4. Non-farm enterprise choices and outcomes, including number and location of suppliers, goods and services offered, price search, and exit/entry.

5. Directory usage (for the treated), sharing of directories, specific firms contacted and information sought.

Spillover treatment effects within and across small communities

This is a household-level randomized control trial involving a sample of approximately 3400 households in 120 villages.

Prior to launching the RCT we conducted a census of enterprises in a wide range of sectors, in 102 locations in 6 districts of the Kagera region, Tanzania. The information collected from these enterprises forms the basis of our primary intervention, a business directory. Villages were chosen for the census if they had a population of at least 4500 people in the most recent census.

To choose study villages, we first excluded central urban areas by dropping the fully urban district of Bukoba Municipal, and excluding any villages within 20km of the center of the three largest towns/cities in the study area. We then enrolled three types of villages: the 30 smallest villages from the enterprise census used to populate the directory (those are the smallest 30 villages in the study area that have populations of 4500 people or above); the 30 largest non-census villages (the 30 villages in the study area with less than 4500 people); and 60 small villages randomly selected from a pool of villages with a population below a certain threshold. We randomly assigned 30 of the 60 small villages to pure control.

Within the 90 villages not assigned to pure control, households will be randomly assigned to control, paper treatment, and digital treatment.

The experiment is designed for the following comparisons:

1. Paper treatment v. Control and Digital Treatment v. Control (household-level randomization)
2. Heterogeneous effects based on whether the village was included in the enterprise census, and therefore has local businesses represented in the directory (household-level randomization, leveraging the RD around village size threshold for inclusion in the census)
3. Heterogeneous effects based on village size (household-level randomization)
4. Spillovers to untreated households based on comparing households in pure control villages to control households in villages with some treated households (village-level randomization)

Randomization of households into treatment will be done upon reaching the end of the baseline survey, by the survey software. Treatment status will be unknown by the enumerator until they reach the final baseline module.

The primary unit of randomization is the household, except for the spillover analysis, for which the primary unit of randomization is the village.

Yes

For the spillover analysis, 30 villages (clusters) were assigned to pure control, and 30 were assigned to treatment. In the latter 30 villages, we randomly assign 16 households to control, 12 to paper treatment, and 15 to digital treatment. In the pure control villages, we randomly assign 16 households to control.

Otherwise, randomization is at the household level.

3400 households.

30 small population villages with treatment: 1290 households
30 small population villages pure control: 480 households
30 villages just below population threshold for inclusion in the directory: 840 households
30 villages just above population threshold for inclusion in the directory: 840 households