Massachusetts Institute of Technology: The power of weak ties in gaining new employment

If you have a LinkedIn account, your connections probably consist of a core group of people you know well, and a larger set of people you know less well. The latter are what experts call “weak ties.” Now a unique, large-scale experiment co-directed by an MIT scholar shows that on LinkedIn, those weak ties are more likely to land you new employment, compared to your ties with people you know better.

“When we look at the experimental data, weak ties are better, on average, for job mobility than strong ties,” says Sinan Aral, a management professor at MIT and co-author of a new paper detailing the results of the study, which involved millions of LinkedIn users.

The experiment upholds the idea, first posited nearly 50 years ago, that weak ties have a value strong ties do not. The people you know best may have social networks that closely resemble your own and thus may not add much new job-seeking value for you. Your more casual acquaintances, on the other hand, have social networks that overlap less with yours and may provide connections or information you would not otherwise be able to access.

In recent years, however, some scholars have suggested there is a “paradox of weak ties,” in which strong ties actually are more useful in the job market. But the new experiment provides evidence to the contrary; weak ties are indeed more useful, a finding that particularly applies to more digitally oriented industries.

“Our experiment provides evidence in the opposite direction from the ‘paradox of weak ties,’” Aral says.

The paper, “A Causal Test of the Strength of Weak Ties,” appears today in Science. The authors are Karthik Rajkumar, a computational social scientist at LinkedIn; Guillaume Saint-Jacques PhD ’18, a senior manager at Apple who previously worked as a research scientist and manager at LinkedIn; Iavor Bojinov, an assistant professor at Harvard Business School and a former data scientist at LinkedIn; Erik Brynjolfsson, the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI, and director of the Stanford Digital Economy Lab; and Aral, the David Austin Professor of Management at the MIT Sloan School of Management and director of the MIT Initiative on the Digital Economy.

A novel test of weak ties

The notion that there is something especially useful about the more tenuous connections in your social network dates to a highly influential 1973 paper by Stanford sociologist Mark Granovetter, “The Strength of Weak Ties,” from The American Journal of Sociology. In it, Granovetter identified weak ties as a key source of “diffusion of influence and information, mobility opportunity, and community organization.”

Granovetter’s ideas have spread widely in academia and Silicon Valley, especially with the growth of online social networks, but have been tough to test. For instance, regarding job-hunting, it can be difficult to untangle the impact of someone’s social network from their networking skills. As the scholars also note in the paper, it is also challenging to find solid data sets linking social networks and job searches in the first place.

The current study gets traction on the issue in a unique way, as a five-year experiment involving LinkedIn’s “People You May Know” (PYMK) algorithm, which suggests new connections to site users. To conduct the experiment, from 2015 through 2019, LinkedIn adjusted the PYMK algorithm, so that some site users saw a higher concentration of PYMK suggestions to whom they had strong ties, and others received more PYMK suggestions to people with whom they had weak ties.

The scholars also defined tie strength in two ways: by interaction intensity, based on the number of messaging interactions people had, and in structural terms, based on the number of mutual friends two users had in common.

All told, the experiment involved around 20 million LinkedIn users, who over the five years ended up creating about 2 billion new connections on the site, recorded over 70 million job applications, and wound up accepting 600,000 new jobs identified through the site.

“This is by far the largest longitudinal, randomized controlled experiment on the strength of weak ties ever conducted,” Aral says. “I don’t think there’s any real debate about that.”

And from that mass of data, a clear pattern emerged. As the researchers write in the paper, “the stronger the newly added ties were, the less likely they were to lead to a job transmission.”

The upside-down U shape

The fact that weak ties led to more job opportunities, overall, is just one of multiple related findings from the study. The scholars also found that the connection between structural tie strength and job transmission does not exist in a simple inverse form.

“The strength of weak ties is not linear,” Aral says. Charting the relationship between structural tie strength — the number of connections you have in common with someone — and usefulness, the scholars found those two properties have an upside-down U-shape form. The connecting bar between the two sides of the “U” is where moderately weak ties are, representing the highest-yield connections that people have on LinkedIn.

“Moderately weak ties are the best,” Aral says. “Not the weakest, but slightly stronger than the weakest.” The inflection point is around 10 mutual connections between people; if you share more than that with someone on LinkedIn, the usefulness of your connection to the other person, in job-hunting terms, diminishes.

However, when it comes to interaction intensity — how often you are in communication with someone — the results look a bit more linear. In this case, what emerges is closer to the idea that the weakest ties produce the most results, and the strongest ties produce the least job transmission.

“These two measures behave differently,” Aral says. “It’s important to think about weak ties in this multifaceted way, with interaction intensity and structural bridging.”

Finally, the usefulness of weak ties varies by industry on LinkedIn. The power of weak ties on the site is especially strong in high-tech industries.

“Weak ties are better in more digital industries,” Aral says, defining those as fields that are “more suitable for machine learning, artificial intelligence, more software intensive, more suitable for remote work, an so on. In those industries, weak ties are even more important. In analog industries, stronger ties can be more important.”

This could be due to what Aral has in previous research called the “refresh rate” of digital industries, in which they keep evolving quickly, making it more important to have a wide range of connections — especially weak ties — in those fields. Still, Aral notes, “We encourage more research, because we need to know more about why this variation seems to exist across industries.”

Developed at MIT

The genesis of the study goes back several years, when Saint-Jacques was pursuing his PhD research at MIT Sloan, advised by Brynjolfsson (then at MIT) and Aral. The group developed the idea of the research project, and after Saint-Jacques joined LinkedIn, had the opportunity to engineer the large-scale experiment.

The sheer size of the study, Aral notes, made it possible for the researchers to draw their multiple conclusions with confidence.

“The scale of the experiment is necessary for that, because you need a lot of statistical power to examine the question with such granularity,” Aral observes.

Other scholars say the study is a significant addition the literature on social and professional networks and weak ties.

For his part, Aral says he regards the current study as part of a larger effort, involving both himself and other members of MIT’s Initiative on the Digital Economy, to grasp the real-world impacts of digital social platforms.

“The main thrust of conversation around those platforms in the world today has been about how they affect society, like teen mental health, democracies and our elections and the spread of fake news, and whether misinformation affects the global pandemic and vaccine hesitancy,” Aral says.

In this case, Aral adds, “What this study really highlights is, we have to add to that: How are these platforms affecting the global economy? … This shows that one of the algorithms on LinkedIn can affect employment patterns, and it’s the largest professional network on the planet. We need to add that to the discussion about the impact of digitial social networking on the world.”