Say you’re looking for a new job. You go to LinkedIn to brighten up your profile and browse your social network.

But who should you go to for an introduction to a potential new employer?

A new study of more than 20 million people, published in Science, shows that your close friends (on LinkedIn) aren’t the best choice: instead, you should look to acquaintances you don’t know well enough to share a personal connection with .

The power of weak ties

In 1973 the American sociologist wrote: Mark Granovetter coined the phrase “the strength of weak ties” in the context of social networks. He argued that the stronger the bonds between two individuals, the more their friendship networks will overlap.

Simply put, you probably know all the friends of a close friend, but few friends of an acquaintance.

So if you’re looking for a job, you probably already know everything your immediate area has to offer. Intuitively, it is the weak ties – your acquaintances – that offer the most opportunities for new discoveries.

Weak ties and jobs

Granovetter’s theory feels good, but is it? A team of researchers from LinkedIn, Harvard Business School, Stanford and MIT set out to find empirical evidence about how weak ties affect labor mobility.

Their research relied on the efforts of engineers at LinkedIn to test and improve the platform’s “People You May Know” recommendation algorithm. LinkedIn regularly updates this algorithm, recommending new people to add to your network.

One of these updates tested the effects of encouraging the formation of strong bonds (recommend adding your close friends) versus weak bonds (recommending acquaintances and friends of friends). The researchers then followed the users who took part in these “A/B tests” to see if the difference affected their employment outcomes.

More than 20 million LinkedIn users worldwide were randomly assigned to well-defined treatment groups. Users in each group were presented with slightly different new contact recommendations, giving users in some groups stronger bonds and users in other groups more weak bonds.

Next, the team measured how many jobs users in each group requested and how many “job transfers” occurred. Job transfers are of particular importance as they are defined as getting a job in the same company as the new contact. A job transfer suggests that the new contact helped place the job.

Moderately weak tires are best

The study uses causal analysis to go beyond simple correlations and link linkage to employment. There are three main findings.

First, the recommendation engine has a significant impact on link formation. Users who were recommended more weak links formed significantly more weak links, and users who were recommended more strong links formed more strong links.

Second, the experiment provides causal evidence that moderately weak ties are more than twice as effective as strong ties in helping a job seeker with a new employer. What is a “moderate” weak tie? The study found that job transfers are most likely to come from acquaintances with whom you share about 10 mutual friends and rarely interact.

Third, the strength of the weak links varied by industry. While weak ties increased labor mobility in more digital industries, strong ties increased labor mobility in less digital industries.

Better recommendations

This LinkedIn study is the first to causally prove Granovetter’s theory in the labor market. The causal analysis is key here, as large-scale studies of correlations between the strength of ties and orbit transfer have shown that strong ties are more favorable, in what has hitherto been considered a paradox.

This study resolves the paradox and once again proves the limitations of correlation studies, which work poorly at disentangling confounding factors and sometimes lead to erroneous conclusions.

From a practical point of view, the study outlines the best parameters for suggesting new compounds. It turned out that the connections most helpful in finding a job are your acquaintances, people you meet in professional settings, or friends of friends, rather than your closest friends – people with whom you share about 10 mutual contacts and with whom you are less likely to interact regularly.

These can be translated into algorithmic recommendations, making the recommendation engines of professional networks like LinkedIn even more adept at helping job seekers find jobs.

The power of black boxes

The public is often wary when major social media companies run experiments on their users (see Facebook’s infamous 2014 emotion experiment).

So, could LinkedIn’s experiment have harmed its users? In theory, the users in the ‘strong link’ treatment group missed the weak links that could have led to their next job.

However, all groups had some degree of labor mobility, some more than others. Moreover, since the researchers observed a technical experiment, the study itself seems to raise few ethical concerns.

Nevertheless, it reminds us to wonder to what extent our most intimate professional decisions – such as selecting a new career or workplace – are determined by black box artificial intelligence algorithms whose workings we cannot see.

This article was republished from The conversation under a Creative Commons license. Read the original article.