A Scientific Approach to Entrepreneurship and Innovation: Evidence from a Field Experiment in Tanzania

An increasingly complex socio-economic landscape has recently highlighted the importance of decision-making under uncertainty. Making decisions under conditions of uncertainty is crucial to technological innovation, social change, and the creation of new ideas, and it is at the core of entrepreneurship. However, evidence suggests that entrepreneurs lack a method to make decisions under conditions of uncertainty. This project extends prior work that shows that entrepreneurs can improve their ability to make these decisions by adopting a scientific approach based on the formulation of models tested with data, such as scientists do.
It proposes to conduct a randomized controlled trial (RCT) to provide solid evidence and identify elements of greatest importance in application of the scientific approach. By scientific approach to entrepreneurial decision making, we mean that entrepreneurs should behave in a fashion similar to scientists -- accurately framing and identifying the problem they wish to address, articulating theories, defining clear hypotheses, conducting rigorous tests to prove or disprove these hypotheses, measuring the results of the tests, and making decisions based on these tools.
Building on prior research conducted in Italy, the United Kingdom and India, this works focuses on the particularly important setting of uncertain decision making in the context of the food and agricultural sectors in Tanzania. Effective entrepreneurial behavior is critically important to sustained economic development in developing nations such as Tanzania, where the food and agricultural sector typically is the largest sector in terms of the number of workers employed.

External Link(s)

Registration Citation

Citation

Agarwal, Rajshree et al. 2022. "A Scientific Approach to Entrepreneurship and Innovation: Evidence from a Field Experiment in Tanzania ." AEA RCT Registry. May 31. https://doi.org/10.1257/rct.7560-1.1

Sponsors & Partners

Experimental Details

Interventions

Intervention(s)

Intervention Start Date

2021-06-01

Intervention End Date

2022-06-01

Primary Outcomes

Primary Outcomes (end points)

1. Performance: our main dependent variable is the amount of revenue/total sales generated by the businesses over time. We also record an alternative measure for performance by recording considering the number of employees and related changes over time.
2. Dropout: this is a binary variable that takes value 0 until the firm drops out (they abandon the program and cease the startup), 1 in the period in which the firm drops out, and it is a missing value thereafter. To avoid attrition biases, we will check that the firms which leave the program will actually cease entrepreneurial activity.
3. Pivot: this is the cumulative number of times in which a firm makes a major change to its business model. We defined a change to be major by analyzing whether the firm moved from the original idea to another idea that changed the core value proposition of the product or service.

The research team will also examine potential mechanisms behind these outcomes by collecting variables related to how and when decisions are made.
The key mechanism being tested is precision and improvement in the precision of predictions. To capture such mechanism, the team will collect through survey instruments the participants’ own perceptions about future revenues and likelihood of failing.

Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)

The research team is also examining the interaction of the treatments with several covariates of interest, including:
a) Gender
b) Psychological and cognitive traits
c) Perception of competition and information sharing
d) Management practices adopted by the entrepreneur and the entrepreneurial team
e) Accumulation of knowledge
The research team will also measure the level of adoption of the methodology taught in the intervention by using a semi-structured interview protocol. Such measure allows the team to test not only the “intention-to-treat” of the intervention, but also the extent to which entrepreneurs apply the concepts they are taught.
The research team will also analyze the effect of the intervention on entrepreneurs’ perceptions of internal and external challenges to entrepreneurship in emerging economies and their perceived ability to respond to these challenges. We expect the training to push entrepreneurs to focus more on internal challenges, with such focus shifting towards external ones over time. As for the perception on abilities, we have competing hypotheses about the direction of the effect of the intervention.
The research team will study not only the moderation effects of cognitive and psychological traits, but also the direct effects that the intervention could have over such traits. In particular, the research team will study the effects of cognitive traits related to overconfidence and learning orientation.

Secondary Outcomes (explanation)

Experimental Design

The RCT focuses on entrepreneurial individual who intend to start a new business or pursue a new business idea within an established entity. We advertise the program through digital channels as a general course overseeing all the important aspects of entrepreneurship. The course is free of charge to ensure the participation of individuals with limited financial resources. The same amount of training is offered to both groups receiving training to ensure comparability. Multiple instructors/coaches will be trained to deliver the training using standard materials, purposely designed. Each instructor/coach teaches two classes (one class employing the theory-driven approach and the second class using the experimentation-driven approach). The research team designs, coordinates, and oversees all activities, ensuring that the learning modules and coaching activities are appropriately carried out by the instructors/coaches.
The research team will collect various measures related to both participants and their business projects before, during, and after the training using surveys and interviews. Individual-level measures include participant demographic information, education and work background, prior experience with entrepreneurship, and some cognitive and psychological traits. Project-level measures include past and current performance, team composition and perceptions of future performance. We will also collect specific measures related to participants’ perception of internal and external challenges to entrepreneurship as well as their perceived ability to respond to them.

Experimental Design Details

Randomization Method

Pure parallel randomization, done in office by a computer using STATA. After being randomly assigned into one of the three groups (theory-driven, experimentation-driven and pure control), multiple subgroups of participants (each with about 15 members) will be created and randomly matched with the instructors/coaches so that each instructor/coach teaches two subgroups. Randomization checks: mean values comparison and t-tests across groups.

Randomization Unit

Organization Project

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

Not clustered

Sample size: planned number of observations

225 participants X 10 (points in time) = 2250 observations

Sample size (or number of clusters) by treatment arms

225 startups (75 in the theory-driven group, 75 in the experimentation-driven group, and 75 in the pure control group)

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

We report here the results of a sensitivity analysis, given a fixed sample size, in a conservative scenario with no attrition and only 5 data points collected. Calculations were done using the free GPower software. We set the study power and alpha at standard levels (0.8 and 0.05). We assume the treatment has a small effect (f = 0.1). F tests - ANOVA: Repeated measures, within-between interaction Analysis:Sensitivity: Compute required effect size. Input::α err prob =0.05; Power (1-β err prob) = 0.8; Total sample size=225; Number of groups=3; Number of measurements=5;Corr among rep measures=0.5;Nonsphericity correction ε=1; Output: Noncentrality parameter λ =15.1535971; Critical F=1.9488129;Numerator df=8.0000000;Denominator df=888;Effect size f=0.0820666 We also run other two simulated scenarios considering large attrition. In the first exercise, we consider a large 35% attrition (N = 146) and 5 collected datapoints; in the latter, we consider 6 repeated measurements. All the other settings (power, alpha, groups, correlation, nonsphericity correction) are kept as in the above simulation. The computed MDE for the first additional scenario is of 0.102 while it is of 0.09 for the latter. Thus, the experimental design should allow capturing a small treatment effect (f = 0.1), even in the case of large attrition.

Supporting Documents and Materials

IRB