Virtual social networks and entrepreneurship in low-income countries: A pan-African RCT
Last registered on April 17, 2019


Trial Information
General Information
Virtual social networks and entrepreneurship in low-income countries: A pan-African RCT
Initial registration date
April 15, 2019
Last updated
April 17, 2019 8:31 PM EDT
Primary Investigator
Bocconi University
Other Primary Investigator(s)
PI Affiliation
Bocconi University
PI Affiliation
Bocconi University
Additional Trial Information
Start date
End date
Secondary IDs
We conducted a large RCT at the pan-African level to shed light on the impact of peer effects on innovation and entrepreneurship. The experiment involved around 5000 entrepreneurs (some established, others just aspiring) from 45 African countries. All those entrepreneurs completed an online business course; the treated individuals had the additional possibility of interacting with peers, within groups of sixty and according to one of three different setups: (a) face-to-face, (b) virtually “within” (where interaction was conducted through an Internet platform in groups of the same country), (c) virtually “across” (where the virtually connected groups displayed a balanced heterogeneity across countries). After two and a half months, all participants were asked to submit business proposals, the ones actually submitted being then graded by business professionals and senior investors. The latter also assessed them for possible funding. The resulting 1-5 grades – conceived as measuring entrepreneurship quality – defined one of our main outcome variables, i.e. the quality of submitted proposals; on top of this, the decision itself of whether to submit a business proposal in the first place determined another outcome variable of interest.
In the paper, first we outline our main results concerning the effect of the treatment on the two aforementioned outcomes, submission and quality (intensive margin).
As a second step in the analysis, we construct a social network in each group by defining a weighted directed link between two entrepreneurs as the amount of information (overall size of messages) written by one of them for which we can confirm that the other has been actively exposed to (and hence provided feedback on with some positive probability). Then, exploiting the network structure for the purpose of identification, we estimate the induced peer effects.
A combined consideration of the results reveals a rich interplay of treatment and peer effects. On the one hand, we may conclude that some group homogeneity – or face-to-face contact– bring about positive treatment effects while the group heterogeneity displayed under virtual-across interaction fails to deliver significant such effects on all three dimensions. In contrast, network-based peer effects deliver a quite different pattern. For, in particular, we find that under virtual-within interaction, entrepreneurs’ peers exert a significantly positive influence on submission and the extensive margin but not so on quality per se, i.e. on the intensive margin. A somewhat polar behavior arises in small countries, whose treated entrepreneurs only “interact across” in heterogeneous groups: they also enjoy positive and significant peer effects on extensive quality, but this is to be understood as a reflection of the impact on intensive quality, not submission. In this case, however, there is the third possibility of entrepreneurs of large countries interacting within heterogeneous groups. In a sense, these entrepreneurs appear to combine, in a dominant manner, the features of the other two contexts that, in each case, render peer effects non-significant on, respectively, submission and intensive quality. This, in the end, yields peer effects on extensive quality non-significant as well. The contrast between the nature and implications of the treatment effects and the network peer effects is interesting and deserves further investigation. A possible explanation could hinge upon the positive role that (national) homogeneity/familiarity may play as a source of encouragement (and hence participation), as opposed to the negative impact it could have in reducing novelty of ideas and/or highlighting the fear of competition (thus dis-incentivizing information sharing). To gain a good understanding of these issues, however, one needs the help of theory as well as a detailed investigation of on how communication actually unfolds in our context.
Both the development of a suitable theoretical framework and a a systematic semantic analysis of the vast flow of information exchanged by entrepreneurs (over 140,000 sentences) are part of ongoing research.
Registration Citation
Garbin, Francesca, Fernando Vega-Redondo and Fernando Vega-Redondo. 2019. "Virtual social networks and entrepreneurship in low-income countries: A pan-African RCT ." AEA RCT Registry. April 17.
Sponsors & Partners

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
Experimental Details
The aim of this work is to study how learning, cooperation and competition shape peers' interaction and affect innovation and entrepreneurship. In order to do so, we recruited entrepreneurs from all over Africa (via social media, viral campaigns, information events on the ground) and offered them an online business course; these participants represented the whole population (treated and controls). Prior to the intervention, all the prospective students had to fill in a 30-minute survey providing some information on their demographics, business profile, network measures, personality traits. The treatment involved exposure to peer interaction; hence, while taking the business course, the treated subpopulation had the chance to interact in groups of 60 entrepreneurs in three different subtreatment designs. The first treatment arm consisted of face-to-face interactions, in Uganda. The other two used an online platform were people could chat; these virtual arms differed in the composition, as one group was homogeneous by nationality (hereafter called the virtual-within group) while the other had members from different countries (virtual-across group). The online course lasted for ten weeks, from end of May to end of July 2017; while the peer interaction happened from beginning of June to mid August. At the end of the course, the aspiring entrepreneurs were asked to submit their individual business proposals by August 15th, which were then redirected to our African panel of investment professionals to be evaluated and possibly funded by our senior fi nancial partners (VCs, investment funds, angel and institutional investors). The evaluation in fact took place in two stages; the first one, from mid September to mid October, was conducted by experienced African professionals, and the second one from mid November 2017 to end of February 2018 was carried out by the partner investors on the subset of projects that had received the highest grades in the previous step.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
The key outcome variable is the evaluation obtained in the final project (business proposal) submitted at the end of the course. This evaluation delivers valuable information; we first investigate the effect on attrition, i.e. the probability of not submitting a proposal. Then, on a second stage, we study the effect on the quality of the business proposals, as evaluated, first, by a panel of experts, then by a panel of investors. The proposals have been graded from 1 to 5, in order of increasing quality in each stage.
Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
The construction of the network on the basis of the recorded online interaction is crucial to determine peer influence. This is one additional outcome of the experiment. Moreover, the messages sent on the chatting platform have been semantically analysed at sentence level in order to label the communication as more or less business-focused, as providing concrete feedback, encouragement or not, etc.. This is the third outcome of the experiment. All three outcomes (the primary outcome and the two secondary ones), are studied jointly, analysing their interplay.
Secondary Outcomes (explanation)
While for the submission and the evaluation of the business proposal the unit of measure should be clear, the construction of the network is hardly straightforward. In the virtual groups (both across and within countries) we use the data from the chats in order to define the existence of a link between two entrepreneurs. The specification of the connections has been conducted in many alternative ways. In principle, following a path of progressively refi ned formulations, we could posit that a (directed) link from i to j is formed when, at some time t along the interaction process, one of the following (nested) events happens.
(a) Individual i posts a message at some t and j logs in the platform after i's posting.
(b) As in (a) and, in addition, j writes back at some at some subsequent time in the same channel as i's original post.
(c) As in (b) and, in addition, for some given natual number x (a parameter of the construction), there are no more than x intermediate posts between i's original post and j's subsequent post.
(d) As in (c) and j's later post has some term(s) in it that relate to the content of i's original post.
The above alternative formulations require increasingly stringent evidence of awareness in the interaction of i and j. Our leading formulation has been (c), but we also checked the robustness of our results to the consideration of the weaker notions (a) and (b), and also outlined briefly what the use of (d) entail.
Experimental Design
Experimental Design
RCT involving a large population of African entrepreneurs from 49 countries all over the continent.
Before the start of the intervention, all the recruited participants had to fill in a 92-question survey that took approximately 30 minutes to complete; questions involved demographic information, the business profile, some measures of pre-existing networks and communication, personality traits.
The full population of treated and controls took an online business course; at the same time, the treated individuals have been assigned to different subgroups in order to favour interaction but in different frameworks. The first treatment arm consisted of face-to-face interactions, only in Uganda. The other two used an online platform to chat; these virtual arms differed in the composition, as one group was homogeneous by nationality (hereafter called the virtual-within group) while the other had members from different countries (virtual-across group). To stimulate interaction, an incentive mechanism was implemented every two weeks in all the three subtreatments to induce entrepreneurs to look for their best match; participants had to hand some tokens to the peers whom they found most helpful, both as a reward and as a means of keeping them involved. The tokens acted as lottery tickets for the best entrepreneurs for one of the 30 prizes offered by Bocconi. At the end of the course, the aspiring entrepreneurs were asked to submit their individual business proposals, which were then redirected to our African panel of investment professionals. These professionals assessed and ranked those proposals in two stages. First, a group of 15 junior members of investment firms, entrepreneurship hubs, or accelerators evaluated all proposals and selected the 600 best. Then, in a second phase, those selected business proposals were again evaluated, ranked, and some funded by our senior fi nancial partners (VCs, investment funds, angel and institutional investors).
Experimental Design Details
Randomization Method
The randomization across the subsamples was conducted by stratifying according to the following baseline characteristics: gender, having a prior business, and submitting the fi rst "milestone" of the course on time (i.e. a first business proposal submitted at the beginning of the online course). This strati fication was implemented in order to ensure control-treatment balance on those three characteristics, judged particularly important. However, our randomization also achieves a quite precise balance on the rest of individual characteristics elicited from the survey.
Randomization Unit
The population has been divided into three disjoint subsamples. In the Uganda sample, people living near Kampala have been randomly partitioned into three equal-sized groups: control, face-to-face interaction, virtual-within interaction. People from the large-country sample have been randomly divided into three equal-sized subsamples as well: control, virtual-within interaction, virtual-across interaction. Finally, the small-country sample has been divided only into control and virtual-across interaction groups. Randomization into treatments was conducted for each of these three subgroups.
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
The virtual-interaction groups count a total number of 44 clusters (total of virtual-across and virtual-within chatting platforms).
Sample size: planned number of observations
Almost 5'000 entrepreneurs from 49 countries in the African continent.
Sample size (or number of clusters) by treatment arms
Uganda sample (568 observations): control 189, face-to-face 189, virtual-within 190.
Large-country sample (3,333 observations): control 1,111, virtual-across 1,111, virtual-within 1,111.
Small-country sample (1,057 observations): control 529, virtual-across 528.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
(1) Virtual-within interaction has a positive and significant treatment effect on the three dimensions: submission, intensive quality, and extensive quality. Instead, when interaction is face-to-face (thus also “within”) only submission and the extensive quality margin are affected (positively so). (2) Virtual-across interaction yields no significant effect on any of the former three dimensions. (3) The baseline quality of entrepreneurs has a uniformly positive effect on performance. 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. (4) When effective, the treatment operates by shifting up, on average, the evaluation grade of business proposals from low levels (grades 1 and 2) to high ones (grades 4 and 5). (5) As a test of robustness, the core treatment effects described in (1)-(2) are confirmed to remain essentially unchanged under a full range of control (baseline) variables, while the composition effects identified in (4) are found to survive a standard placebo test. (6) In large countries (i.e. those for which we can form homogeneous groups), virtual-within interaction leads to positive and significant peer effects on submission and extensive quality (but not intensive quality). Instead, when entrepreneurs of large countries are exposed to virtual-across interaction, no significant peer effects arise. (7) In the complement set of small countries, virtual-across interaction (the only possible) has positive and significant peer effects on both extensive and intensive quality but not on submission. (8) Neither peers’ baseline quality, nor their experience, have any significant effect on performance. (9) The results in (6)-(7) are robust to redefining the network in two alternative ways: (a) requiring that links embody communication spanning a maximum (parametrized) lag; (b) assuming that links are fully two sided, their weight reflecting the amount of information flowing in both directions.
IRB Name
Ethics Committee of Università Commerciale Luigi Bocconi
IRB Approval Date
IRB Approval Number
Post Trial Information
Study Withdrawal
Is the intervention completed?
Intervention Completion Date
August 15, 2017, 12:00 AM +00:00
Is data collection complete?
Data Collection Completion Date
August 15, 2017, 12:00 AM +00:00
Final Sample Size: Number of Clusters (Unit of Randomization)
2940 individuals treated in 44 different chatting platforms (summing up the virtual within and virtual across interaction groups).
Was attrition correlated with treatment status?
Final Sample Size: Total Number of Observations
Uganda sample: 568 observations; Large-country sample: 3,333 observations; Small-country sample: 1,057 observations.
Total: 4958 observations (including control, face-to-face, virtual within, virtual across).
Final Sample Size (or Number of Clusters) by Treatment Arms
Uganda sample (568 observations): control 189, face-to-face 189, virtual-within 190. Large-country sample (3,333 observations): control 1,111, virtual-across 1,111, virtualwithin 1,111. Small-country sample (1,057 observations): control 529, virtual-across 528
Data Publication
Data Publication
Is public data available?
Program Files
Program Files
Reports and Papers
Preliminary Reports
Relevant Papers