How Assigned Teammates Become Future Connections
This chapter focuses on the workplace mechanism behind the series’ central idea: can an assigned starting position change the likelihood that a later professional tie forms? The ConsultCo study is useful because it separates an initial condition created by the organization from later collaboration shaped by employee and manager choices. It can therefore show how a first project changes access to people without claiming that it determines pay, promotion, or a person’s eventual career.
Where assignment ends and choice begins
ConsultCo is a global professional-services firm with more than 50,000 employees. Its work is organized around projects, and employees move among teams that vary in size and duration. In this setting, a workplace tie is not an abstract feeling of collegiality. It is an observed relationship: two employees worked on the same project during the same month. That definition is narrow, but it gives the researchers a repeatable way to watch a network form and change.
The timing matters. New hires receive their first project-team assignments through a centralized human-resources process, randomly within their office. They do not choose the initial arrangement that gives them their first set of coworkers. Later staffing works differently. Junior employees seek projects—often hoping to prove their value on high-priority or client-facing work—while more senior colleagues have discretion over whom to recruit or select. Later collaboration is thus a bilateral process: a newcomer wants access to a project, and someone on the project decides whether to work with them.
That transition creates the study’s causal leverage. If people who later form ties also had larger or more useful initial networks, an ordinary observational analysis could not tell whether the network mattered or whether those people differed from the start. Perhaps they were more confident, more talented, more similar to the colleagues they approached, or already likely to be selected. Random assignment of starting teams changes the initial network position independently of those personal characteristics, at least within the limits of the design. The later tie is then observed after the initial exposure has had time to operate.
The study does not randomize friendship, trust, or a particular indirect tie directly. It randomizes the placement from which those features are calculated. In the paper’s formal language, an initial network is followed by a later network, and the researchers ask whether a selected feature of the first network changes the probability of a focal tie in the second. This is a useful distinction. The intervention is the starting arrangement; the network feature is the mechanism being tested; later co-membership on a project is the outcome.
Repeated observation also avoids a problem that appears in static network snapshots. A single finished network cannot reveal which relationship came first. If three people eventually form a triangle, the final picture does not tell us whether one person introduced the other two, whether the ties formed independently, or whether all three responded to some unobserved similarity. Observing the starting arrangement and the later relationship gives the analysis a temporal order.
There are two further conditions behind the causal interpretation. The exclusion restriction requires a reason to believe that the chosen initial feature is the relevant route by which the starting network affects the later focal tie, rather than a proxy for every possible characteristic of the network. The support condition requires enough variation: across feasible assignments, the relevant network feature must sometimes take different values for comparable potential ties. Without that variation, no method can learn the effect of the feature from this design.
This sequence is why the workplace case speaks directly to the first-project question. The organization controls the initial exposure, but the later tie is produced through a process that includes employee initiative and senior discretion. A starting project does not mechanically assign someone to every future opportunity. It changes the relationships and information through which later choices are made.
Three ways a starting network can matter
The researchers examine three features of the initial network: indirect ties, degree, and local network density. These are related, but they represent different mechanisms. Treating them as one generic measure of “being connected” would obscure the question the study is trying to answer.
Indirect ties: a route through a mutual contact
Suppose a new hire, A, works with coworker B. B has previously worked with employee C, but A and C have not yet worked together. C is an indirect tie for A: a connection through a mutual contact. The relationship can create an opportunity without requiring A to approach a complete stranger. B may make C visible, provide context about C’s work, or create a setting in which A and C can meet.
This is the intuition behind triadic closure, the tendency for two people with a shared contact to become connected. A mutual colleague can increase opportunities for interaction and reduce the perceived risk of a new relationship. They may provide trust or accountability, even if no formal introduction occurs. In project-based work, the mechanism is especially plausible because the shared coworker can carry knowledge of how people work, what they are good at, and which project needs are emerging.
The mechanism does not require the indirect contact to be a close friend, and the data do not establish that trust was the psychological reason a tie formed. They show a causal effect of the initial network feature under the study’s design. The interpretation is best kept at that level: a route through a mutual contact makes a later collaboration more likely in this setting. It is not evidence that every contact-of-contact relationship becomes useful or that a referral guarantees selection.
Degree: visibility through the number of existing ties
Degree is the number of a person’s existing connections. A newcomer with many initial coworkers may be visible to more people, receive more information, and have more chances to be considered for later work. This is the network version of cumulative advantage: a small early difference in visibility can attract additional opportunities, which then create still more visibility.
But degree is not the same as access to a particular person. A large network can consist of many weak or irrelevant connections. It also increases the number of possible paths simply because there are more starting points. If a study finds that high-degree people later form more ties, the result may reflect both greater visibility and the larger number of opportunities available to them. That is why the paper considers degree separately from the presence of a specific indirect route.
Local density: how closed the neighborhood is
Local density describes how many of a person’s contacts are connected to one another. Consider two new hires with the same number of initial coworkers. In one network, those coworkers have all worked closely together; in the other, they connect the newcomer to several groups that rarely overlap. The first neighborhood is denser. The second may expose the newcomer to more distinct parts of the organization.
High density can help maintain trust and coordination, but it may leave fewer openings for additional ties. Existing relationships absorb time and attention, and a tightly connected group may provide repeated contact with the same people rather than routes into new circles. This is why density can plausibly reduce the formation of an additional tie even when degree is high. The prediction is not that dense networks are bad. It is that network cohesion and network expansion can serve different purposes.
These mechanisms can overlap in real life. Someone may have many contacts, several mutual links to a prospective collaborator, and a highly cohesive local group. A causal analysis must therefore ask what remains when the features are considered together, rather than treating a result from one feature as the whole story.
What the estimates show
The empirical setting contains many more possible pairs than actual later collaborations. The final analysis sample includes 6,042 inexperienced new hires across 979 offices and more than 130 million potential new-hire/prior-employee pairs. In the full potential-pair sample, only about 0.43% form a connection within the relevant period. That low base rate is important: a change measured in percentage points can be large relative to the chance of a tie forming in the first place.
When the binary features are considered one at a time, the presence of an indirect tie raises the estimated probability of a later tie by about 0.11 percentage points from a 0.26% baseline—more than a 1.4-fold increase. Above-median degree raises the probability by about 0.09 percentage points, while above-median local density lowers it by about 0.09 percentage points. These are absolute percentage-point changes, not claims that the chance of collaboration rises by 11% or falls by 9%.
The more revealing comparison comes from the joint specification. When all three binary features enter together, the indirect-tie effect remains large, at about 0.10 percentage points. The degree estimate falls to about 0.02 percentage points, and the density estimate to about −0.03 percentage points. This does not make visibility or neighborhood structure irrelevant in every circumstance. It shows that their apparent separate effects overlap with other network features and are less robust when the mechanisms are considered together.
The continuous specifications sharpen the result further. Each additional indirect tie raises the probability of a later tie by about 0.056 percentage points; a one-standard-deviation increase in the number of indirect ties corresponds to roughly a 0.25-percentage-point change. By contrast, the estimated effects of degree and local density are essentially zero and far from statistically informative in the continuous model. The robust signal is therefore not simply “people with more connections do better.” It is more specifically that being connected through mutual contacts can change the probability of a subsequent workplace tie under the conditions studied.
The researchers also run placebo tests using characteristics that existed before the later collaboration, including gender, race, and whether a new hire attended a top-20 institution. These outcomes should not be changed by the randomized initial team placement. The placebo estimates generally show no meaningful effects, which supports the claim that the initial assignment was not systematically sorting people by those characteristics. That check cannot prove that every possible confounder is absent, but it makes the assignment assumption more credible.
The result has a practical implication for how a first project should be evaluated. A team is not only a collection of tasks and skills; it is also a set of possible paths into the organization. A well-connected starting team may expose a newcomer to future collaborators through mutual contacts. Yet increasing the number of names in a person’s address book is not the same as creating useful access. The joint and continuous results argue for attention to the structure of exposure, not just its volume.
The causal claim still stops at the observed outcome. The design tells us that initial network features can affect the probability of a later project tie in one professional-services firm, under a particular assignment process and definition of collaboration. It does not tell us that the new tie improves performance, leads to promotion, increases compensation, or changes lifetime earnings. A tie is an intermediate opportunity, not the final return.
That distinction matters when extending the model to scholarly mobility. The workplace case shows how an initial position can help create later relationships. The next question is how researchers’ existing individual and institutional connections appear in the destinations of their first academic moves—and why those pathways are not all the same.