Google, Facebook or Apple? Wisdom of the crowds.

It’s an often repeated question — I heard a lot of people left X and joined Y, is that true? Is Y the hottest company right now? The answers are often anecdotal - a friend of a friend moved from X to Y last month, or a long list of subjective rants on the internet about X or Y. The data is usually outdated, some of the popular search results refer to pre-IPO Facebook. How can one find an answer that is grounded in data and up to date?

Migration Volumes

The most straightforward thing to do is to find out the raw number of employees who migrated from X to Y. Finding the exact number is obviously very hard, but a good approximation can be found using LinkedIn advanced search which allows you to set current company and previous company as filters. For example, using LinkedIn advanced search reveals the number of employees at Uber who worked at Facebook before

Migration-Volume(Facebook ⇒ Uber) = 233

And the migration in the opposite direction

Migration-Volume(Uber ⇒ Facebook) = 17

But comparing the raw numbers moving from X to Y can be misleading as it depends on the relative sizes of X and Y. Large companies will have a large number of employees leaving, as well as joining them.

Migration Probability

We fix the problem in comparing migration volumes by normalizing them by the number of employees of the previous company. This gives us the probability of migration from X to Y. It is the ratio of the number of employees at Y who worked at X before, compared to the current employees at X. Below are sample queries:

Current-employees(Apple) = 121487

Migration-Volume(Apple ⇒ Google) = 1192

Migration-Probability(Apple ⇒ Google) = 1192 / 121487 = 0.0098

Similarly, Migration-Probability(Google ⇒ Apple) = 705 / 69521 = 0.0101

So the probability of migrating from Google to Apple is higher than the probability of migrating from Apple to Google! Using this as a metric to measure migration seems counter intuitive to the employee satisfaction surveys. The problem is Apple simply has lot more people, so the denominator is large and that drives down the migration probability to any particular company. So this metric doesn’t appear that useful as it favors companies with large number of employees.

Migrant Densities

We can try to fix the problem with migration probability by trying a different normalization instead — comparing the fraction of employees at X who came from Y. This is the migrant density of X at Y, computed as the fraction of employees at Y who worked at X before. For example,

Current-employees(Uber) = 20488

Migration-Volume(Microsoft ⇒ Uber)= 486

Migrant-Density(ex-Microsoft @ Uber) = 486 / 20488 = 0.0237

Similarly, Migrant-Density(ex-Uber @ Microsoft) = 19 / 134431 = 0.0001

So may be Uber has an edge over Microsoft, but this metric seems rather unfair to Microsoft. It’s easy to make up a large fraction at a small company. This metric now favors small companies too much.

Migration Lift

One way to balance this opposing tension between big and small companies is to compute an adaptation of lift scores. Lift scores measure the strength of a signal over a naturally occuring value. From the above example, since Microsoft is a large company, it’ll naturally make up a significant population of any smaller tech company. The lift score measures whether at any company, the fraction of ex-Microsoft employees is larger than the expected value.

For example, one can compute the lift of migration from Microsoft to Google as

Migration-Volume(Microsoft ⇒ Google) = 3939

Ex-Employees(Microsoft) = 250642

Current-Employees(Google) = 69521

Migration-Lift (Microsoft ⇒ Google) = 3939 / (250642 * 69521) = 2.2 x 10-7

This computes the extra “lift” of migration from Microsoft to Google, compared to all the migration to Google that would have taken place anyway.

The table below shows lift scores for migrations within Google, Microsoft, Apple, Facebook, Amazon and Uber.

The graph below is a sorted view of the migration lift.

Several interesting facts:

• One can see the strong lifts from Microsoft ⇒ Google, Google ⇒ Facebook and Facebook ⇒ Uber and their relative strengths. It’s surprising that such a simple metric can track the ongoing shift in technology.
• Although the strongest lift is from Google ⇒ Facebook, the reverse lift from Facebook ⇒ Google comes at third spot and is quite strong as well, underlining the neck to neck competition. In comparison, there is a strong lift from Facebook ⇒ Uber, but not in the opposite direction.
• The lift from Amazon ⇒ Microsoft is higher than the lift from Microsoft ⇒ Amazon. So the theory held in some circles that Amazon is bleeding Microsoft doesn’t appear true and the two are at equal footing. In fact there are strong lifts away from Amazon towards all other companies, perhaps a reflection of the employee sentiments there.
• The lifts towards Uber seem very strong with negligible lifts away from it. Uber is also the only company in the group that hasn’t gone IPO. It’ll be interesting to revisit the numbers post IPO.

Caveat for the analysis is that data gets stale, please refresh the inputs when looking at the results beyond 2016.