How we built the all-new LinkedIn Top Companies methodology

By Michael Lombard and Laura Lorenzetti

Two thousand hours a year. Given the average 40-hour work week, that’s how much time we dedicate to our jobs every year — and that’s not even including commute time (even if it’s now only from your bedroom to your kitchen table). A job isn’t just a paycheck, it’s a step in a career arc, whether it’s a jumping off point, a skill builder, a stepping stone or a culmination of years of hard work.

Last year, after 4 years of publishing the LinkedIn Top Companies list, we knew we wanted to do a better job of capturing that reality. The Top Companies list since its inception showed professionals where people like them were most eager to work. But it didn’t tell you how well those companies set you up for 100,000+ hours of a career. So prior to the 5th anniversary, we asked ourselves two questions: What does it mean to build a great career? And how do we uncover the companies that best support people for a rewarding career??

Those questions kicked off a dialogue between our editors and data scientists, academic experts and career coaches. We took the research and married it up with our own proprietary data that would allow us to measure the corporate world over a long period of time.?

As we debut this brand new methodology, our lead data scientist Michael Lombard and I want to pull back the curtain and share the dialogue we had to build and test our new Top Companies methodology.

Want to skip to the big reveal? You can see this year’s 50 Top Companies in the U.S. here.?

LAURA [Editor]: We started our methodology revamp with that one big question: What makes a good career today? “Good” is a relative term — and that can mean a lot of things to a lot of people. At the end of the day, a good career — and a great job — is deeply personal, but I believe there are key factors that could apply to the majority of workers in the U.S. and beyond. So, I set out to figure out what those universal factors might be.?

My first stop: published research. I scoured academic journals like Organization Science, topical publications like Harvard Business Review and individual, detailed research with thrilling titles like “Time and job satisfaction: a longitudinal study of the differential roles of age and tenure.” I also turned to the experts directly, speaking with people like Wharton professor Matthew Bidwell, Harvard Business School professor Francesca Gino and organizational psychologist and Wharton professor Adam Grant. These experts took the time to answer my myriad questions, from “Is longer tenure always better?” (not necessarily) to “Do perks matter?” (not really) and “What about layoffs?” (It’s complicated: Too many is bad, but some may be necessary to evolve a successful business.).?

Grant uncovered an interesting tidbit when he worked on a research project with Facebook — No. 19 on our list this year — that gets to the heart of what makes a good career. There’s a saying that workers quit a boss, not a job. But that’s not what Grant and Facebook found. “The decision to exit was because of the work,” Grant wrote, in partnership with Facebook leaders. “They left when their job wasn’t enjoyable, their strengths weren’t being used, and they weren’t growing in their careers.”

This expert input helped me hone in on key areas that lead to career fulfillment: the opportunity to progress both at a single company and beyond, a supportive and connected workplace culture, the ability to gain new skills and diversity in the types of people you get to work alongside.?

At that point, it was time to turn to the data. Michael’s work began.

MICHAEL [Data Scientist]: Once Laura had identified the key pillars, we needed to translate this to LinkedIn data. We are fortunate to have a tremendous amount of career-focused data. But, just as scarce data can be a problem, having too much data isn’t great either. My first step was to map each pillar to an internal data source: Promotion rates became our position tables, recruiting interest became our communication tables and so on. This is data speak to say, each idea we had about career development had to map to data that LinkedIn possessed.?

I drafted my first attempt at this back in 2019 on an obtrusive whiteboard that I’m sure irked my desk neighbors. That turned into basic SQL queries to get counts, summary statistics, and such. And that quickly grew to include all sorts of filters and quality control checks. I spent half my time pulling data and the other half?in conversations with domain-area experts at LinkedIn to learn the ins and outs of their datasets, such as how our standardization teams map subsidiary companies or how we differentiate recruiter activity from other member activity on our platform.

Some of these queries grew more complex, while others imploded after we found better data sources or our strategy changed. My notes extended beyond my whiteboard to the walls and windows around my desk. Russell Crowe’s John Nash may have taken to writing on windows to unlock the governing dynamics of the universe. I did the same as I tried to quantitatively map out the driving forces of a good career.?

After weeks of whiteboarding, writing code, deleting code, munching on Unreal peanut butter cups while staring into space, deleting more code and writing up documentation, I had enough data to share with Laura. This is always a wonderful moment...until people start looking at it. Things you might have thought were good signals turn out to be useless noise. Clever approaches turn out to be too clever for their own good. What might have worked in Japan has no relevance in Spain. But at that first moment, when you have a complete list, you at least have the fleeting feeling of accomplishment.

LAURA [Editor]: The first version of a data dump can feel chaotic, but that’s where the partnership between a data scientist and an editor hits its stride.?

Michael handed off a spreadsheet with hundreds of thousands of cells filled with data and thousands of companies under consideration. (And, yes, it was a very slow-loading document.) Our fearless Top Companies editor, Ashley Peterson, and I started organizing it. We first looked at growth at the companies, understanding how a particular employer gained or lost employees over the past year. We then made sure that what we saw matched what the company had reported publicly. Accuracy is king for any ranking. Then we walked through each data set under consideration to understand how and why it shifted companies rankings. We wanted to make sure the data accurately reflected what’s happening in the market, and we wanted to only include data that had strong, positive signals to what we know helps build a good career.

Michael guided us on statistical rigor and answered questions about specific data points. He must have grown tired of hearing “But why?” so many times. But it was that back and forth that honed the data, leading us to throw out data categories that weren’t reliable signals. (One example of a data set we discarded was job views. That data accurately reflected in-demand companies but didn’t speak directly to career progression, which was our ultimate goal with this new methodology.) This helped us zoom in on what matters for career building and led us to our current data combination, which is based on 7 pillars that you can find at the bottom of this article in the detailed methodology. Over time, we expect the pillars to stay the same, but the data sources to expand.

MICHAEL [Data Scientist]: How can we be sure our data accurately reflects the truth? A cop out answer is that this is a Sisyphian task, never fully completed. There is some truth to that — our methodology is a living thing, with the data reflecting the changing world of work as we experience it. We analyzed market penetration rates, regressed our findings against outside sources such as the World Bank and the Bureau of Labor Statistics, and evaluated various scoring mechanisms for every pillar we selected. The work is tedious but important: One pillar’s background work completed by my teammate Mar Carpanelli fills 30 pages of summary statistics, outlier definitions, and comparative scoring techniques. These documents made legal and peer review sessions much simpler. And review we did: code reviews upon code reviews across multiple data scientists and economists, legal reviews of all kinds to ensure we were protecting member privacy, committee reviews of methodological approaches, communications reviews of presentations, and so on. When we received questions from partners about whether we had considered X or Y, we almost always could not only say we had considered it but point to the exact quantitative reason that particular approach did not work. To ensure accuracy, even the simplest of counts ended up as pages of notes and back and forth.

Once our economists had signed off, lawyers gave the go-ahead, and communication teams had approved our approach, we then met with LinkedIn editors in every country that generated a list to get their feedback. I can confidently state that I never expected “discussing corporate holding structures of Japanese companies at 11 pm” to be part of my job description. The on-the-ground expertise of our LinkedIn News partners is paramount to the success of this project. From idiosyncrasies of corporate structures (I, for one, never knew that Mitsubishi Fuso in Japan was owned by Daimler AG) to naming conventions for the unemployed or freelancers (I beg of you to not name your company Freelancer) to corporate scandals or illegal activities that may be common knowledge to an entire country but breaking news to a single data scientist in New York. Each review led to adjustments in our methodology and additional filters to remove noise. More importantly, each review led to improvements in our data set that we never could have discovered by just looking at these numbers in a vacuum.

LAURA [Editor]: As Michael so eloquently lays it out, a methodology like Top Companies is never built in isolation. It takes a small data-crunching village, hundreds of combined hours, the tech savvy to be able to scrape the BLS website and the reporting chops to dig up even the most obscure of layoff reports. Once our methodology was locked, we checked the final data over and over before we locked the ranking. As I mentioned before, accuracy is king.??

We want this list to serve members well and be a year-round resource to identify the right opportunity for you. While there are some key, overarching themes to a fulfilling career, identifying the right job is personal. And this list is designed to be a starting point for you — these companies are investing in their employees’ careers and one of them (or several of them) might have the right job for you as the next step in your career journey.

We’d love to hear from you: What matters most to you as you build your career? What do you look for in a company? And you can see all 50 U.S. Top Companies here.

Methodology

Our methodology uses LinkedIn data to rank companies based on seven pillars that have been shown to lead to career progression: ability to advance; skills growth; company stability; external opportunity; company affinity; gender diversity and educational background. Ability to advance tracks employee promotions within a company and when they move to a new company, based on standardized job titles. Skills growth looks at how employees across the company are gaining skills while employed at the company, using standardized LinkedIn skills. Company stability tracks attrition over the past year, as well as the percentage of employees that stay at the company at least three years. External opportunity looks at Recruiter outreach across employees at the company, signaling demand for workers coming from these companies. Company affinity, which seeks to measure how supportive a company’s culture is, looks at connection volume on LinkedIn among employees, controlled for company size. Gender diversity measures gender parity within a company and its subsidiaries. Finally, educational background examines the variety of educational attainment among employees, from no degree up to Ph.D. levels, reflecting a commitment to recruiting a wide range of professionals.?

To be eligible, companies must have at least 500 employees as of Dec. 31 in the country and reductions in staff (including attrition and layoffs) can be no higher than 10% (based on LinkedIn data or public announcements). Only parent companies rank on the list; majority-owned subsidiaries and data about those subsidiaries are incorporated into the parent company score. Certain data counts — like gender diversity and educational background — are normalized based on company size across the pool of companies eligible for the list. The methodology time frame is Jan. 1, 2020 through Dec. 31, 2020. All of the data used is aggregated and/or de-identified.

We exclude all staffing and recruiting firms, educational institutions and government agencies. We also exclude LinkedIn, its parent company Microsoft and Microsoft subsidiaries.

Correction: Certain data counts were normalized for company size. The previous version stated that all data counts were normalized for company size.