Digital Technology Adoption

In this project, I investigate the role of learning costs on digital technology adoption by women microentrepreneurs in the state of Karnataka in India. Over the last decade, digitalisation has slowly transformed how informal micro-firms operate in rural areas. This transformation involves a sharp rise in new and emerging digital technologies, however, the adoption still remains low. To help rationalize the low adoption of digital technology, I argue that high learning costs faced by a firm limits adoption of digital technologies. This study examines whether a temporary intervention that can subsidize learning costs can move firms from a low adoption equilibrium to a high adoption one. To guide the key predictions of the paper, I set up model that describes technology adoption choices made by a firm that optimizes on an infinite horizon in discrete time. I complement the model with a field experiment where I randomize firms to treatment arms where learning costs are subsidized to varying degrees, and study how this drives adoption choice. I also present findings on the firm level effects of sustained adoption. Further, to understand the comparative advantage of various firms from digital technology adoption, I study the heterogeneity in treatment effects by size of firm, level of human capital and behavioral parameters.

2022-06-20

2024-01-01

The two primary outcome variables are - (i) a binary measure of adoption and (ii) the extent of adoption. Data for the primary outcomes is collected using the MeraBills dashboard which collects real time data on when and how the app is used.

For the binary adoption dummy, real time data from the app is aggregated to the daily level. To study the extent of adoption, the real time data is aggregated to calculate the number of transactions entered weekly on the digital platform. Using this data, I create weekly measures that are then pooled for the study period.

The secondary outcomes of interest are under two broad categories: business practices and firm outcomes. Under business practices, I look at outcomes such as, managing trade credit, managing inventories, managing customers etc. For firm outcomes, I look at profits, revenues, costs, probability of shutting down, probability of starting new business, number of employees and time use. The study has other auxiliary outcomes of interest such as Women’s Economic Empowerment and time use.

The sample is comprised of 320 women-led microenterprises that are located across the Kolar, Tumkur, Chikkabalapura, and Ramanagar districts on Karnataka. The design randomizes participants into three groups: one control and two treatment groups corresponding to the three experimental treatments described below. All three groups are offered access to the app. To study whether learning costs drive adoption decisions, I vary the extent to which subsidize learning costs across the three arms.

Not available

Randomization done in office by a computer.

Individual level randomization

No

320 women microentrepreneurs

320 women microentrepreneurs

~100 microentrepreneurs in the Control group ; ~110 microentrepreneurs in the T1 group; ~110 microentrepreneurs in the T2 group.

For power calculations, I assume 2\% of the control group will use the app, 15\% of T1 will use the app and 30\% of T2 will use the app. I estimate the sample size for two-sample comparison of proportions (for T2 vs. T1 and T1 vs. C), where the proportion in each group are given by the proportion that will use the app. This shows that a sample of 288 participants is enough to detect the above change in proportions with a power of 85\%. In this project, I set the sample size 320 (slightly greater than 288). I also run power calculations by simulating the experiment data and obtain similar estimates. \cite{mckenzie2012beyond} provides useful insights on optimal design. He shows that using multiple rounds of data collection and averaging outcomes over these rounds, leads to substantial increases in power, especially when outcomes are poorly correlated over time (which is often the case for business outcomes). Using the app, I collect weekly data on outcome variables and pool the data over the rounds.