Source · aikyam school

Peer Networks and Entrepreneurship: a Pan-African RCT

Original source

Fernando Vega-Redondo; Paolo Pin; Diego Ubfal; Cristiana Benedetti-Fasil; Charles Brummitt; Gaia Rubera; Dirk Hovy; Tommaso Fornaciari

Peer Networks and Entrepreneurship: a Pan-African RCT (2019)

working paper

Search for this source ↗

Permission for School’s use is unestablished.

Used in these cards

Decompose conditional quality metrics using the Better Always Submitters framework when program treatments induce differential submission or completion rates between treated and control groups.

"Therefore, if the estimated treatment effect on quality conditional, of course, on submission is positive and significant, this is in effect an unambiguous indication of a positive unconditional effect as well."

Utilize face-to-face group meetings when maximizing participant retention and completion is required, but transition to virtual within-country platforms when scaling globally to improve proposal quality.

"For the UgS we find that whereas the treatment effect on submission is positive and very significant for the face-to-face treatment, the positive effect arising under virtual-within interaction is not statistically significant."

Structure online entrepreneurship platforms with separate dedicated channels for formal business iteration versus informal peer encouragement to prevent diluting analytical feedback.

"Messages are quite polarized in either the business or sentiment dimension... showing an inverse dependence in the (strong) FOSD sense between the respective business-based and sentiment-based distributions."

Issue periodic peer-reward tokens linked to high-value performance-contingent prizes to stimulate active peer review and feedback in online learning networks.

"Every two weeks... individuals were provided with 10 tokens. They were advised to hand these tokens to those peers whom they found most helpful... the tokens given by the 50 highest-ranked entrepreneurs... acted as lottery tickets for one of the 30 (attractive) rewards."

Assign entrepreneurs from smaller or less saturated markets to international heterogeneous networks to boost idea innovation, while keeping entrepreneurs from large domestic markets in nationally homogeneous groups to drive engagement.

"In the set of small countries, where only virtual-across interaction is possible, there are positive and significant peer effects on both extensive and intensive quality but not on submission... Instead, when entrepreneurs of large countries are exposed to virtual-across interaction, no significant peer effects arise in any of the three outcomes."

Form peer interaction groups matched by baseline skill levels rather than placing novice entrepreneurs alongside highly experienced peers to prevent intimidation and participant churn.

"However, the average baseline quality of group peers have a negative composition effect on intensive quality. In fact, a similarly negative composition effect is also induced by peers' average experience level."

Set communication lag parameters based on intermediate message count ($\tau \ge 80$) rather than elapsed clock time when constructing adjacency matrices from online chat logs.

"For a value of $\tau = 80$... the estimated peer coefficient reaches in every case a value very close to the coefficient estimated for $\tau = \infty$."

Instrument direct peer influences using exogenous baseline traits of two-step network neighbors when estimating social spillover effects in non-saturated social networks.

"The approach we pursue here to tackle the reflection problem follows Bramoullé et al. (2009) and De Giorgi et al. (2010)... It relies on using the variables in G²X (the total influence weights associated to the peers lying two links away) as respective instruments for Gy."

Design virtual networking environments to permit flexible, multidimensional interaction, allowing informal social sentiment to co-exist alongside structured business discussions.

"Tables 20 and 21 show that, either through business-weighted or sentiment-weighted links alike, peers can exert significant influence... even though business-oriented messages are often weak in sentiment content and vice versa, individuals tend to use a quite balanced set of the two types in the overall communication they establish with any given peer."

Apply stability selection with $l_1$ LASSO penalization across repeated random bootstrap sub-samples to identify stable text regressors in high-dimensional communication data.

"We fit from 200 to 1,000 independent logistic-regression models on random subsets of the data... The regressors that have a score larger than 0.5 are declared to be informative."

Implement a two-tier evaluation framework using local ecosystem experts to filter business plans before presenting top-tier proposals to senior venture capitalists for funding consideration.

"A total of 93 proposals actually moved into a concrete follow-up by the investor. Up to 30 of them had been funded or were still followed up as of November 2018."

Collecting detailed baseline data ensures that participant backgrounds and business stages are thoroughly documented.

Physical face-to-face contact builds social norms and mutual familiarity among participants.

Linking tokens from top-performing entrepreneurs to valuable lottery prizes motivates participants to search for promising peers.

Second-degree neighbors influence an individual only indirectly through intermediate direct peers, satisfying the exclusion restriction.

Comparing full cumulative distribution functions avoids reliance on parametric normality assumptions.

Combining digital reach with social incentives allows programs to rapidly scale across national borders while maintaining local trust.

Exposure to cross-country peer diversity provides novel perspectives and non-redundant market information that enhances the intensive business quality of proposals.

National homogeneity provides a shared cultural foundation and familiar social context that fosters trust and encouragement.

Pre-recorded video lectures and automated testing allow consistent quality delivery across large cohorts without geographic limits.

Counting intermediate messages reflects how far down a user's screen an earlier message moves before a response occurs.

Random forest algorithms aggregate predictions across multiple decision trees to prevent overfitting on sparse text data.

When always-submitters have stronger baseline ability than conditional submitters, selective entry into submission can only pull down average treated scores.

Projecting network links onto specific semantic themes reveals whether peer influence operates through analytical business advice or emotional support.

L1 LASSO regularization forces coefficients of non-essential terms to zero, creating sparse models.

Local ecosystem experts effectively filter out low-quality proposals using standardized criteria.