Embracing diverse values in company culture pays off—literally

Berkeley Haas Assoc. Prof. Sameer Srivastava found organizations that embrace diversity are more innovative

Berkeley Haas Assoc. Prof. Sameer Srivastava found organizations that embrace diversity are more innovative

Zappos says its 10 core values are a “way of life,” while Netflix details seven aspects of its culture, nine “highly valued behaviors and skills,” plus deal-breakers like “no brilliant jerks.” Nordstrom has just one rule of thumb: “Use good judgment in all situations.”

More and more companies are asking employees to adopt a set of core values, seeking to build a culture that will give them an edge. But while getting everyone on the same page makes it easier for people to work together, too much of the same thinking can stifle creativity. What’s the right balance of cultural values to drive profitability, growth, and innovation?

An analysis of 500,000 Glassdoor.com reviews of S&P 500 firms found that companies whose employees disagree on core values are, indeed, less profitable than similar companies where workers are culturally aligned. Meanwhile, the firms that are the most highly valued and innovative have something in common: They embrace a diverse range of cultural values throughout the organization.

Sameer Srivastava

Assoc. Prof. Sameer Srivastava co-directs the Computational Culture Lab and the Berkeley Culture Initiative.

“Past research has suggested there’s a tradeoff between diversity and productivity,” said Berkeley Haas Assoc. Prof. Sameer Srivastava, co-author of the study, which is forthcoming in Administrative Science Quarterly. “We suggest it’s a false tradeoff. You can have a multiplicity of ideas and values and also have cultural alignment on those ideas and values.”

A new way of thinking about cultural diversity

The paper defines a new way of thinking about diversity in organizations and reconciles a fundamental contradiction in current thinking. On the one hand, deep differences in how people think can create problems when they have to coordinate on tasks; research has found that a strong and unified culture increases productivity and efficiency. On the other hand, diverse viewpoints and perspectives can help people respond to change and uncertainty, and ultimately recombine ideas into something novel.

To get a more nuanced view of the cultures of different organizations and their relationship to business performance, Srivastava and collaborators Amir Goldberg of Stanford and Matthew Corritore of McGill drew on the power of the Computational Culture Lab, which Srivastava and Goldberg co-direct. The joint Berkeley-Stanford lab uses data science to develop new ways of measuring organizational culture. (Srivastava, Goldberg and co-researchers previously analyzed 10 million internal emails from a technology company to learn about culture fit within an organization.)

This time they looked at differences between organizations, turning to Glassdoor, a job search platform with 17 million monthly users who post anonymous reviews of their employers. The company has a data science team that agreed to share data with the research team to gain new insights.

The power of machine learning

The researchers used natural language processing and machine learning to identify hundreds of topics in comments about company culture—ultimately choosing 500 topics to cover as wide a range as possible. After “training” their statistical model to identify patterns in culture-related sentences, they scanned over 500,000 reviews of 492 S&P 500 firms posted between 2008 and 2015. They matched companies with similar characteristics for purposes of comparison, limiting the sample to companies with at least 25 reviews.

By looking at how many topics were mentioned in reviews of each company, how many times they were mentioned, and how much commonality there was between reviews, they were able to determine whether a company’s culture was diverse or uniform, and divided or aligned.

For example, they classified a company as divided when they found little overlap in topics mentioned by reviewers—as might happen at a company where customer service reps prioritize delivering “wow,” but engineers care only about technical progress, and the finance team is laser-focused on profits. They found other companies where everyone talked about the same few topics: They were culturally unified, but had little diversity.

It was the companies in which employees, on average, talked about many different culture-related topics that seemed to hit the sweet spot for innovation. That kind of diversity was strongly associated with a higher market valuation—as measured by Tobin’s Q. Those firms also produced more patents on average, as well as higher-quality patents that were built on by other companies, than similar firms where the typical employee mentioned relatively few topics related to culture.

Conversely, the researchers found that firms with highly divided, rather than unified, cultures were less profitable: those types of cultures were associated with lower returns on assets (ROA).

Diversity of values, distribution of values

Interestingly, these statistical relationships were true no matter which specific values were mentioned (e.g. collaboration, adaptability, playfulness). The important factors were the variety of culture-related topics discussed and how consistently people mentioned those topics throughout the organization.

The paper suggests that in assessing a company’s culture, it’s important to look beyond which values are emphasized to how they are distributed in a group. While diversity may arise from differences between people—which the researchers call “interpersonal diversity”—it’s also true that individual people often hold multiple values, which may even be contradictory. They define this as “intrapersonal diversity.” This view builds on research that finds when people have a broad “toolkit” of cultural resources, they have greater capacity for creativity and adaptability.

New technique for measuring culture

In addition, the power of data science and natural language processing offers an exciting new way for organizations to understand what makes for a successful culture. Traditional approaches such as collecting demographic information like age, gender, or ethnicity may or may not relate to underlying beliefs, and surveys are not only expensive but also relatively static. This approach allows researchers to examine topics that people are actually talking about and how these topics vary over time.

“This gives us a much more granular measure of culture over time,” said Srivastava, who also co-leads the Berkeley Culture Initiative, which he founded with Prof. Jennifer Chatman to develop new approaches to organizational culture research.

Srivastava cautions that a limitation of Glassdoor data is that people are writing for an external audience, and they choose to write reviews for a complicated set of reasons—including in response to campaigns by their employers. To the extent possible, the researchers did account for these dynamics in their analyses, he said.

You Might Also Like