The “Strength of Weak Ties” Submitted to a Causal Test
In his famous 1973 paper, Mark Granovetter offered a provocative hypothesis: Weak ties (i.e., socially distant relationships with whom people have infrequent interactions) might be more beneficial than strong ties (i.e., close relationships with whom people have frequent interactions). Because weak ties connect individuals to diverse and remote parts of the network, they give people access to novel and unique information, which can then translate into new professional opportunities.
However, this is a very hard hypothesis to test: While many papers have shown that weak ties indeed correlate with job opportunities, no causal relationship had been established… Until recently!
A team of researchers from LinkedIn, Harvard, Stanford, and MIT found a way to submit the “strength of weak ties” theory to a causal test… And while their results confirm Granovetter’s initial theory, they also bring more nuance to it!
What are the results?
They show that moderately weak ties are most effective. If a tie is very strong (i.e., you have a lot of friends in common and often interact with one another), it is likely that this person would help you… but unlikely that this person would have unique, novel, or valuable information to offer. On the contrary, if a tie is very weak (i.e., no friends in common, no interaction), it is likely that they have access to valuable information that you do not possess… but unlikely that they will share it with you. The sweet spot, therefore, is in the middle: Ties that are sufficiently strong to be willing to help you, and sufficiently weak to have access to novel information.
They also find that the “weak tie” advantage differs according to the industry considered. Weak ties might be more beneficial in digital industries (characterized by greater IT, software intensity, machine learning, artificial intelligence, remote work, higher degree of robotization) and strong ties more beneficial in industries that rely less on software and robotization.
How did they show that?
The team of researchers had three main challenges:
First, the sample size: To be able to detect an effect of network structure (for example, the proportions of weak or strong ties in a network) on an outcome (finding a job), you need a (very) large sample. Indeed, people change jobs infrequently, so you need a lot of “events” in your data to be able to capture any effect.
Second, network ties and job outcomes mutually affect each other. People’s social network determines (at least partly) their professional position, but the opposite is true as well: People’s job also determines their social network. It is therefore hard to determine if it is ties that affect job outcomes or the other way around.
Third, many unobserved factors play a role in people’s social network and in their job trajectory: People’s level of effort, their ability, their social skills will affect both their ability to find a job and the type of network they have. These factors make it more difficult to identify a causal link between weak ties and job opportunities.
To address these challenges, the researchers use multiple large-scale randomized experiments on LinkedIn: They randomly recommended new strong (vs. weak) connections to LinkedIn users. They measure a strong (vs. weak) tie by the number of friends two users have in common (mutual connections), and by the number of messages they exchanged on the platform (interaction intensity).
They then track job applications (the number of jobs a LinkedIn user applied to in the three months after the experiment) and job transmissions (when a LinkedIn user reports working at a certain company, after having been friend for at least a year with another LinkedIn user working at the same company).
In this design, the random assignment of users to the weak- or strong-tie condition acts as a “shock” allowing to estimate the impact of ties on job mobility. This treatment is called an instrumental variable: It changes the network structure (explanatory variable) of LinkedIn users without directly affecting their job mobility (dependent variable), while keeping constant (through random assignment) all other unobserved factors likely to affect both the explanatory and the dependent variable.
The first experiment was run in 2015 over 4 million LinkedIn users and created over 19 million new connections. The second experiment was run in 2019 on more than 16 million LinkedIn users around the world, and created roughly 2 billion new connections.
Moderately weak ties (that is people with whom you share some friends and have low interaction) might yield the most in your job search. Weak ties give you access to novel information, new pockets of knowledge, new social relationships. However, they must not be so weak that you have no common ground with the person: You need a minimal level of familiarity and common interests to be sure that the person will share useful information with you.