Small Group Interactions and Social Networks

The intervention is an invitation to attend a structured, in-person quiz night during the first weeks of the MBA program at London Business School (LBS). Students are randomly assigned to treatment or control within prespecified strata. Quiz groups of six or seven students are then formed separately within each stream and experimental arm using a constrained randomized algorithm. Treated students are invited to the event and informed of their quiz-group assignment; control students are neither invited nor informed of their assignment. During the event, quiz-group members collaborate on a quiz and compete against other groups. The event also provides time for food and informal interaction. The intervention is intended to reduce the cost of forming social connections early in the MBA program.

The study uses two complementary empirical designs. First, random assignment at the individual level identifies the intention-to-treat effect of the invitation by comparing students assigned to treatment and control. Second, among treated students, the randomized components of the grouping algorithm generate variation across quiz groups in the shares of peers who share each student’s gender or ethnicity.

For this second design, we will follow Borusyak and Hull (2023): each student’s realized quiz-group composition will be recentered by subtracting its expected value, estimated from counterfactual reruns of the grouping algorithm. The resulting recentered variation will be used as an instrument for realized composition.

2026-08-17

2026-09-09

Network connectivity. At the student level, connectivity will be measured by degree: the number of other students to whom the student is linked. We will measure degree in the full cohort and separately for connections with (i) assigned quiz-group peers, (ii) students in other quiz groups assigned to the same quiz night, and (iii) students assigned to a different quiz night. At the pair level, connectivity will be measured by an indicator for whether an unordered pair of students is linked. We will analyze this indicator for pairs sharing a quiz group, pairs in different groups assigned to the same quiz night, and pairs assigned to different quiz nights.

Inbreeding homophily. For gender and ethnicity, we will measure whether a student’s links disproportionately involve students who share the same trait, relative to the trait composition of the relevant comparison set. Homophily will be measured for the full cohort and for the same prespecified subsets listed above.

Network surveys will be administered in August 2026 before the intervention (Survey 1), December 2026 (Survey 2), May/June 2027 (Survey 3), and June/July 2028 (Survey 4). In each survey, students answer four sociometric questions and may nominate up to six members of their MBA cohort per question.

For each survey, we construct an undirected OR network. Two students are linked if either student nominates the other in at least one of the four sociometric questions. The pair-level connectivity outcome equals one when a pair is linked and zero otherwise. A student’s degree within a given set is the number of students in that set with whom the student is linked.

For each trait and comparison set, raw homophily is the share of a student’s links that are made with other students who share the trait. The random-mixing benchmark is the share of all students in that comparison set who share the trait. Inbreeding homophily is calculated as:
(raw homophily − random-mixing benchmark) / (1 − random-mixing benchmark).

The measure is defined when the set has at least one pair having the same trait, and the random-mixing benchmark is less than one. A value of zero denotes random mixing; positive values indicate overrepresentation of same-trait links, and negative values indicate underrepresentation.

Network search effort: time spent creating new personal and professional connections.

Event registration: the number of registrations by each student for social and professional events organized by student associations and for career events organized by LBS.

Network satisfaction: satisfaction with the network developed within the MBA cohort.

Well-being: separate measures of life satisfaction, anxiety, and stress.

Career outcomes: internship offers, job-search effort, job offers received at the end of the MBA program, and post-graduation earnings.

Organisational cohesiveness: belonging, inclusion, and commitment to the MBA cohort and LBS

Access to network opportunity: we may design a field experiment that will measure the access to opportunity that are seeded in the MBA 2028 cohort. The experiment will be conducted a few months after the intervention and before the end of the programme.

Network search effort will be measured in Surveys 2, 3, and 4. The exact survey question has not yet been finalized.

Event registration will be measured using registration records provided by student associations and LBS. We will separately measure the number of registrations by each student for social and professional events organized by student associations and for career events organized by LBS. Registration is required to attend these events.

Network satisfaction will be measured in Surveys 2, 3, and 4. The exact survey question has not yet been finalized.

Well-being outcomes will consist of separate measures of life satisfaction, anxiety, and stress included in all surveys.

Career outcomes will include internship offers received (Survey 3), job-search effort (Survey 4), job offers received at the end of the MBA program (Survey 4), and post-graduation earnings (LBS administrative data and possibly follow up surveys administered by us). We will seek to collect longer-run career outcomes in subsequent years after the cohort graduates. The precise availability and timing of these measures will depend on the survey and administrative data available after graduation.

Organisational cohesiveness: we may ask survey questions related to belonging, inclusion, and commitment to the MBA cohort and LBS. If we do so, we will elicit these in Surveys 2,3 and 4. The exact survey question has not yet been finalized.

Access to network opportunity: we may design a field experiment that will measure the access to opportunity that are seeded in the MBA 2028 cohort. In case of implementation the experiment will be conducted in the window between a few months after the intervention and before the end of the programme. Furthermore, we will pre-register the experiment.

This study uses a stratified, individual-level randomized controlled design among incoming MBA students. Before the intervention, students complete a baseline network survey. After administrative stream and study-group assignments are available, students are randomly assigned in approximately equal proportions to treatment and control within strata defined by stream, gender, and region of origin.

Treatment is assigned before quiz groups are formed. Quiz groups of six or seven students are constructed separately within each stream and experimental arm using a constrained randomized algorithm. The six streams are paired to form three quiz nights. Students assigned to treatment are invited to the quiz night allocated to their stream; stream allocation to quiz night is done by the program office depending on stream availability (which depends on core courses scheduling). Students assigned to control are not invited.

Follow-up surveys will measure the primary and secondary outcomes.

The study uses two complementary empirical designs. First, individual treatment randomization identifies the intention-to-treat effect of the invitation by comparing students assigned to treatment and control. Second, among treated students, the randomized components of the grouping algorithm generate variation across quiz groups in the shares of peers with a certain characteristic, e.g., the same student’s gender or ethnicity. Following Borusyak and Hull (2023), we will recenter each student’s realized quiz-group composition by subtracting its expected value under counterfactual reruns of the grouping algorithm. The resulting recentered variation will be used as an instrument for realized composition.

Not available

Randomization will be conducted by computer in the researchers’ office after the LBS MBA program team provides the administrative stream and study-group assignments.

Treatment and control. Students will be stratified by stream, gender, and region of origin, with regions classified as Europe and North America, Asia, or all other regions. Within each stratum, approximately half of the students will be randomly assigned to treatment; the remaining students will be assigned to control. Treatment assignment will be completed before quiz groups are formed and before invitations are sent.

Quiz-group formation. After treatment assignment, quiz groups will be formed separately within each stream and experimental arm using a constrained randomized computer algorithm.

Determine quiz-group capacities. For each stream and experimental arm, the algorithm will determine the number and capacities of the quiz groups so that every student is assigned to exactly one group and group sizes are as close as possible to seven, typically six or seven students.

Allocate study groups to quiz groups. The algorithm will construct a binary incidence matrix whose rows represent study groups and whose columns represent quiz groups. An entry of one indicates that the corresponding study group contributes one student to that quiz group. Row totals will equal the numbers of students in the study groups, and column totals will equal the predetermined quiz-group capacities. Because the entries are binary, no quiz group can contain more than one student from the same study group.

Fill the incidence matrix. The algorithm will iteratively select the study group with the largest number of unassigned students, breaking ties randomly. It will allocate slots for those students to distinct quiz groups with the greatest remaining capacities, again breaking ties randomly when necessary. This process will continue until all study-group counts and quiz-group capacities are satisfied.

Assign individual students. Once the incidence matrix has been completed, students will be randomly ordered within each study group and assigned to the corresponding quiz-group slots.

Improve gender balance. The algorithm will identify quiz groups with fewer than two men or fewer than two women, as well as potential donor groups containing at least three students of the required gender. Donor groups and candidate students will be considered in random order. A proposed swap will be accepted only if each receiving quiz group continues to contain no more than one student from any study group. Subject to feasibility, every quiz group will contain at least two men and two women.

All assignments and swaps will occur within the same stream and experimental arm.

The individual student is the unit of treatment randomization. Treatment is not randomized at the quiz-group level. Students in the same quiz group have the same treatment status because quiz groups are formed separately by experimental arm after individual treatment assignment.

There is an additional randomized design stage. Individual students are assigned to quiz groups within their stream and experimental arm through the constrained randomized grouping algorithm.

No

Approximately 500 individual students. Because treatment is randomized at the individual-student level, each student constitutes one treatment-assignment unit; there are no treatment-assignment clusters. The exact number will depend on final MBA enrollment.

Approximately 500 incoming MBA students. The primary student-level analyses use one observation per student in each survey. Pair-level analyses will use all unordered pairs of students in the cohort; the exact number of pairs is N(N-1)/2 per survey and will depend on final enrollment.

Treatment arm: approximately half of the final cohort, expected to be about 250 students, assigned to receive an invitation to the quiz night.

Control arm: the remaining approximately half of the cohort, expected to be about 250 students, not assigned to receive an invitation.

The exact arm sizes will depend on final enrollment and the sizes of the treatment-assignment strata. Within each stratum, the number assigned to treatment will be fixed at approximately half of the students in that stratum.