Toxic Culture Is Driving the Great Resignation

Research using employee data reveals the top five predictors of attrition and four actions managers can take in the short term to reduce attrition.

Reading Time: 12 min 


Measuring Culture

This series includes the MIT SMR/Glassdoor Culture 500, an annual index and research project that uses over 1.4 million employee reviews to analyze culture in leading companies, along with new research focused on measuring organizational culture using a scientific approach.
More in this series
Permissions and PDF

More than 40% of all employees were thinking about leaving their jobs at the beginning of 2021, and as the year went on, workers quit in unprecedented numbers.1 Between April and September 2021, more than 24 million American employees left their jobs, an all-time record.2 As the Great Resignation rolls on, business leaders are struggling to make sense of the factors driving the mass exodus. More importantly, they are looking for ways to hold on to valued employees.

To better understand the sources of the Great Resignation and help leaders respond effectively, we analyzed 34 million online employee profiles to identify U.S. workers who left their employer for any reason (including quitting, retiring, or being laid off) between April and September 2021.3 The data, from Revelio Labs, where one of us (Ben) is the CEO, enabled us to estimate company-level attrition rates for the Culture 500, a sample of large, mainly for-profit companies that together employ nearly one-quarter of the private-sector workforce in the United States.4

While resignation rates are high on average, they are not uniform across companies. Attrition rates for the six months we studied ranged from less than 2% to more than 30% across companies. Industry explains part of this variation. The graph below shows the estimated attrition rate for 38 industries from April through September, and the spread across industries is striking. (See “Industry Average Attrition Rate in the Great Resignation.”) Apparel retailers, on average, lost employees at three times the rate of airlines, medical device makers, and health insurers.

The Great Resignation is affecting blue-collar and white-collar sectors with equal force. Some of the hardest hit industries — apparel retail, fast food, and specialty retail — employ the highest percentage of blue-collar workers among all industries we studied. Management consulting, in contrast, had the second-highest attrition rate but also employs the largest percentage of white-collar professionals of any Culture 500 industry. Enterprise software, which also suffered high churn, employs the highest percentage of engineering and technical employees.

Industry explains some of the variation in attrition rates across companies but not all of it. Even within the same industry, we observed significant differences in attrition rates. The figure below compares competitors with high and low attrition rates within their industries. (See “How Culture 500 Company Attrition Rates Compare Within Industries.”) Workers are 3.8 times more likely to leave Tesla than Ford, for example, and more than twice as likely to quit JetBlue than Southwest Airlines.

Not surprisingly, companies with a reputation for a healthy culture, including Southwest Airlines, Johnson & Johnson, Enterprise Rent-A-Car, and LinkedIn, experienced lower-than-average turnover during the first six months of the Great Resignation.

Although the sample is small, these pairs hint at another, more intriguing pattern. More-innovative companies, including SpaceX, Tesla, Nvidia, and Netflix, are experiencing higher attrition rates than their more staid competitors. The pattern is not limited to technology-intensive industries, since innovative companies like Goldman Sachs and Red Bull have suffered higher turnover as well.

To dig deeper into the drivers of intra-industry turnover, we calculated how each Culture 500 company’s attrition rate compared with the average of its industry as a whole. This measure, which we call industry-adjusted attrition, translates each company’s attrition rate into standard deviations above or below the average for its industry.5

We also analyzed the free text of more than 1.4 million Glassdoor reviews, using the Natural Employee Language Understanding platform developed by CultureX, a company two of us (Donald and Charles) cofounded. For each Culture 500 company, we measured how frequently employees mentioned 172 topics and how positively they talked about each topic. We then analyzed which topics best predicted a company’s industry-adjusted attrition rate.

Top Predictors of Employee Turnover During the Great Resignation

Much of the media discussion about the Great Resignation has focused on employee dissatisfaction with wages. How frequently and positively employees mentioned compensation, however, ranks 16th among all topics in terms of predicting employee turnover. This result is consistent with a large body of evidence that pay has only a moderate impact on employee turnover.6 (Compensation can, however, be an important predictor of attrition in certain settings, such as nurses in large health care systems).

In general, corporate culture is a much more reliable predictor of industry-adjusted attrition than how employees assess their compensation. The figure below displays the five predictors of relative attrition. (See “Top Predictors of Attrition During Great Resignation.”) To give a sense of their relative importance, we’ve benchmarked each element relative to the predictive power of compensation.7 A toxic corporate culture, for example, is 10.4 times more powerful than compensation in predicting a company’s attrition rate compared with its industry.

Let’s take a closer look at each of the top five predictors of employee turnover.

Toxic corporate culture. A toxic corporate culture is by far the strongest predictor of industry-adjusted attrition and is 10 times more important than compensation in predicting turnover. Our analysis found that the leading elements contributing to toxic cultures include failure to promote diversity, equity, and inclusion; workers feeling disrespected; and unethical behavior. In an upcoming article, we will dive deeper into each of these factors and examine different ways managers and employees can spot signals of toxic culture.8 For now, the important point is that a toxic culture is the biggest factor pushing employees out the door during the Great Resignation.

Job insecurity and reorganization. In a previous article, we reported that job insecurity and reorganizations are important predictors of how employees rate a company’s overall culture. So it’s not surprising that employment instability and restructurings influence employee turnover.9 Managers frequently resort to layoffs and reorganizations when their company’s prospects are bleak. Previous research has found that employees’ negative assessments of their company’s future outlook is a strong predictor of attrition.10 When a company is struggling, employees are more likely to jump ship in search of more job security and professional opportunities. Past layoffs, moreover, typically leave surviving employees with heavier workloads, which may increase their odds of leaving.

Another reason job insecurity could predict turnover is related to our measure of employee attrition, which incorporates job changes for all causes — including layoffs and involuntary terminations. We would expect frequent mentions of reorganizations and layoffs to predict involuntary turnover. According to the U.S. Bureau of Labor Statistics, however, involuntary separations have accounted for less than one-quarter of all employee exits among large companies during the Great Resignation.11 So it’s likely that poor career prospects and job insecurity contributed significantly to employees leaving on their own accord as well.

High levels of innovation. It’s not surprising that workers leave companies with toxic cultures or frequent layoffs. But it is surprising that employees are more likely to exit from innovative companies. In the Culture 500 sample, we found that the more positively employees talked about innovation at their company, the more likely they were to quit. The attrition rates of the three most innovative Culture 500 companies — Nvidia, Tesla, and SpaceX — are three standard deviations higher than those in their respective industries.

Staying at the bleeding edge of innovation typically requires employees to put in longer hours, work at a faster pace, and endure more stress than they would in a slower-moving company. The work may be exciting and satisfying but also difficult to sustain in the long term. When employees rate their company’s innovation positively, they are more likely to speak negatively about work-life balance and a manageable workload. During the Great Resignation, employees may be reconsidering the personal toll that relentless innovation takes.

Failure to recognize performance. Employees are more likely to leave companies that fail to distinguish between high performers and laggards when it comes to recognition and rewards. Companies that fail to recognize and reward strong performers have higher rates of attrition, and the same is true for employers that tolerate underperformance. The issue is not compensation below market rates, but rather recognition — both informal and financial — that is not linked to effort and results. High-performing employees are the most likely to resent a lack of recognition for their results, which means that companies may be losing some of their most productive workers during the Great Resignation.

Poor response to COVID-19. Employees who mentioned COVID-19 more frequently in their reviews or talked about their company’s response to the pandemic in negative terms were more likely to quit. The same pattern holds true when employees talk more generally about their company’s policies for protecting their health and well-being.

Short-Term Actions to Boost Retention

The powerful predictors of attrition listed above are not easy to change. A weak future outlook that spurs restructuring and layoffs may be difficult to reverse; it is too late to fix a poor response to the pandemic; and a toxic corporate culture cannot be improved overnight. Relentless innovation provides companies like Tesla or Nvidia with a competitive advantage, so they must find ways to retain employees without sacrificing their innovation edge.

Our analysis identified four actions that managers can take in the short term to reduce attrition. (See “Short-Term Steps for Companies to Increase Retention.”) As in the graph above, each bar represents the topic’s predictive power relative to compensation. This time, the topics predict a company’s ability to retain employees compared with industry peers. Providing employees with lateral career opportunities, for example, is 2.5 times more powerful as a predictor of a company’s relative retention rate compared with compensation.

Provide opportunities for lateral job moves. Not all employees want to climb the corporate ladder or take on additional work or responsibilities. Many workers simply want a change of pace or the opportunity to try something new. When employees talk positively about lateral opportunities — new jobs offering fresh challenges without a promotion — they are less likely to quit. Lateral career opportunities are 12 times more predictive of employee retention than promotions. We observed the same pattern in multinationals: The more frequently employees discussed the possibility of international postings, the more likely they were to stick with their current employer.

Sponsor corporate social events. Company-organized social events, including happy hours, team-building excursions, potluck dinners, and other activities outside the workplace are a key element of a healthy corporate culture, so it’s no surprise that they are also associated with higher rates of retention.12 Organizing fun social events is a low-cost way to reinforce corporate culture as employees return to the office, and it strengthens employees’ personal connections to their team members.

Offer remote work options. Much of the media coverage of the Great Resignation has focused on the importance of remote work in retaining employees. Unsurprisingly, when employees discussed remote work options in more positive terms, they were less likely to quit. What you might not have expected is the relatively modest impact of remote work on retention — just a bit more powerful than compensation in predicting lower attrition. Remote work options may have a modest effect on employee turnover because most companies in an industry converge on similar policies. If companies cannot differentiate themselves based on remote work options, they may need to look elsewhere — providing lateral job opportunities, for instance, or making schedules more predictable — to retain employees.

Make schedules more predictable for front-line employees. When blue-collar employees describe their schedules as predictable, they are less likely to quit. Having a predictable schedule is six times more powerful in predicting front-line employee retention than having a flexible schedule. (A predictable schedule has no predictive power for white-collar employees.)

This finding is consistent with a study of 28 Gap stores, in which employees at randomly-assigned locations received their work schedules two weeks in advance, and their managers were barred from canceling their shifts at the last minute. Employees in the control stores were subject to the usual scheduling practices.13 The stores with predictable schedules increased retention among their most experienced associates. Compared with the workers at the control stores, the employees with fixed schedules had a 7% improvement in their quality of sleep. The benefits were especially pronounced for workers with children, who reported a 15% reduction in stress.

Much of the media coverage of the Great Resignation focuses on high turnover among burned-out knowledge workers who are dissatisfied with their stagnant wages. Our findings are broadly consistent with this narrative. Industries that employ large numbers of professional and technical employees, like management consulting and enterprise software, have experienced high turnover. We found indirect evidence that burnout may contribute to higher levels of attrition among companies that excel at innovation. It’s worth noting, however, that our direct measures of burnout, workload, and work-life balance do not emerge as key predictors of industry-adjusted turnover.

The simplistic narrative of white-collar burnout misses other critical realities of the Great Resignation. Our findings reinforce recent government statistics showing that blue-collar intensive industries like retail and fast food are experiencing unprecedented levels of attrition.14

More fundamentally, we found that corporate culture is more important than burnout or compensation in predicting which companies lost employees at a higher rate than their industries as a whole. A toxic corporate culture is the single best predictor of which companies suffered from high attrition in the first six months of the Great Resignation. The failure to appreciate high performers, through formal and informal recognition, is another element of culture that predicts attrition. A failure to recognize performance is likely to drive out a company’s most productive employees. This is not to argue that compensation and burnout don’t influence attrition — of course they do. The important point is that other aspects of culture appear to matter even more.

Our research identified four steps — offering lateral career opportunities, remote work, social events, and more predictable schedules — that may boost retention in the short term. Leaders who are serious about winning the war for talent during the Great Resignation and beyond, however, must do more. They should understand and address the elements of their culture that are causing employees to disengage and leave. And above all else, they must root out issues that contribute to a toxic culture. Our next article will explore, empirically, what constitutes a toxic culture and how organizations can address this challenge.


Measuring Culture

This series includes the MIT SMR/Glassdoor Culture 500, an annual index and research project that uses over 1.4 million employee reviews to analyze culture in leading companies, along with new research focused on measuring organizational culture using a scientific approach.
More in this series


1. Microsoft sponsored a survey of over 30,000 employees across 31 markets in January 2021 for its Work Trend Index. See “The Next Great Disruption Is Hybrid Work — Are We Ready?” Microsoft, March 22, 2021,

2.Job Openings and Labor Turnover Survey,” U.S. Bureau of Labor Statistics, accessed Dec. 6, 2021, The data represents seasonally adjusted quits for total nonfarm employers in the U.S. from April through September 2021.

3. To test the accuracy of our estimates of employee attrition, we compared them with the November U.S. Bureau of Labor Statistics (BLS) Job Openings and Labor Turnover Survey (JOLTS) for total separations (including employee resignations, layoffs, and other sources of job separations) for private companies with more than 5,000 employees. The BLS total separation rate was 10.8% for April through September 2021, and our estimates were 10.1% for the same period.

4. To estimate turnover at the company level, we identified all job transitions where a user left their current employer for any reason, including quitting, retiring, or being laid off, and divided these by corporate head count. Job transitions were measured for April through September 2021 for 538 Culture 500 companies. The attrition rates are adjusted for sampling bias related to who has an online profile and for lags in when users reported transitions on their profiles. To test the robustness of our estimates of attrition, we separately estimated the hiring rate for each company for the same period. Hiring rate is defined as employees who joined the company divided by corporate head count. We would expect hiring and attrition rates to be correlated, as companies replace employees who leave. The Spearman correlation coefficient was 0.81 between the hiring and attrition rates.

5. We assigned each Culture 500 company to a primary industry and calculated how many standard deviations the focal company’s attrition rate was above or below the industry mean. We used the resulting industry-normalized attrition rate as the dependent variable in our subsequent models. To identify which factors were most important in predicting each company’s normalized attrition rate, we used an XGBoost model and calculated the SHAP (Shapley additive explanations) value for 172 topics measured by the CultureX Natural Employee Language Understanding platform. The SHAP value approach analyzes all possible combinations of features in a predictive model to estimate the marginal impact that each feature has on the outcome — in our case, which cultural elements have the biggest impact in predicting a company’s industry-normalized attrition rate. For an overview of SHAP models, see S.M. Lundberg, G. Erion, H. Chen, et al., “From Local Explanations to Global Understanding With Explainable AI for Trees,” Nature Machine Intelligence 2, no. 1 (January 2020): 56-67.

6. A.L. Rubenstein, M.B. Eberly, T.W. Lee, et al., “Surveying the Forest: A Meta-Analysis, Moderator Investigation, and Future-Oriented Discussion of the Antecedents of Voluntary Employee Turnover,” Personnel Psychology 71, no. 1 (spring 2018): 23-65; and D.G. Allen, P.C. Bryant, and J.M. Vardaman, “Retaining Talent: Replacing Misconceptions With Evidence-Based Strategies,” Academy of Management Perspectives 24, no. 2 (May 2010): 48-64.

7. Relative importance is calculated by dividing each topic’s SHAP value by the SHAP value for the compensation topic. When highly predictive features are closely related (such as job insecurity and restructuring), we report their combined predictive impact.

8. Jason Sockin finds that “respect/abuse,” a topic that overlaps with our definition of toxic culture, is the single best predictor of employee satisfaction. See J. Sockin, “Show Me the Amenity: Are Higher-Paying Firms Better All Around?” SSRN, Nov. 18, 2021,

9. D. Sull and C. Sull, “10 Things Your Corporate Culture Needs to Get Right,” MIT Sloan Management Review, Sept. 16, 2021,

10. B. Zweig and D. Zhao, “Looking for Greener Pastures: What Workplace Factors Drive Attrition?” PDF file (Mill Valley, California: Glassdoor, 2021), A recent meta-analysis also found that job insecurity was a strong predictor of voluntary employee turnover. See A.L. Rubenstein et al., “Surveying the Forest,” 23-65.

11. In the November BLS JOLTS, seasonally adjusted layoffs and discharges for all private companies with more than 5,000 employees represented 24% of total separations for April through September 2021.

12. D. Sull and C. Sull, “10 Things.”

13. J.C. Williams, S.J. Lambert, S. Kesavan, et al., “Stable Scheduling Increases Productivity and Sales: The Stable Scheduling Study,” PDF file (San Francisco: Center for WorkLife Law, 2018),

14.Job Openings and Labor Turnover Summary,” U.S. Bureau of Labor Statistics, Dec. 8, 2021,

Reprint #:


More Like This

Add a comment

You must to post a comment.

First time here? Sign up for a free account: Comment on articles and get access to many more articles.

Comments (24)
Courtney Breen
@DonaldSull - Have you considered examining attrition rates in government or nonprofit organizations? I am very curious about toxic culture in those settings. I also think it would be interesting to hone in on healthcare as an industry, including for profit, nonprofit, and government agencies, and examining it from that angle. Naturally I ask because I work in this field and have observed and experienced many phenomena you discuss in this article.
Wayne Anderson
With great respect, while I think there is a tremendous amount of good information in this article, I would observe that compensation is also a huge driver in the "Great Resignation".  During a tight labor market, the competitive opportunity is unlocking people who have spent the last few years getting experience in whatever is in demand to monetize that experience. 

The most efficient way to monetize that experience for the individual worker is not the current employer - which typically has systems in place that contain the cost of labor for the existing workforce (e.g. limits on raises, decelerators based on level band, etc).  This leads to an imbalanced economic incentive that overcomes the risk of moving to a new company.  Especially in an era where employer loyalty is less of a motivating factor based on the erosion of job protections and "cold" data-driven or segment-driven layoffs.
Donald Sull
@Bonnie Bailey. In analyzing the free text, we classify a review as discussing burnout when the respondent explicitly mentions burnout or synonyms (e.g., "run down," "ground down," etc). The construct of burnout is closely related to worker stress, work overload and mental wellbeing, so we also aggregate these related topics.
Bonnie Bailey
Great article!  How did you measure burnout?
Donald Sull
@Caitlin Sattler on Mar 17, we will publish the article outlining the "Toxic Five" components of a toxic culture and reporting the organizational costs of a toxic culture. We're currently working on article describing how to improve a toxic culture. Best guess we'll publish that one in May.  So stay tuned!

Editor's Note: The new article on toxic culture has published:
Caitlin Sattler
Hi, I was wondering when the next article signaled at the end of this piece will be posted/available?
Donald Sull
@Todd Inskeep Compensation levels are highly correlated with industry.  Average compensation in grocery or fast food industry, for example, looks very different than it does in management consulting or enterprise software. In this study we predict a company's attrition rate compared to its industry, so a good chunk of the differential in compensation will be accounted for by industry control.  

In a separate study, we analyzed predictors of employee satisfaction (a decent proxy for exit intent) among front-line employees whose jobs do not typically require a college degree. These are typically low paying jobs. Again we find that toxic culture is the best predictor of low employee satisfaction.

In our client work at CultureX we can control directly for each employee's compensation (and in some cases employees compensation relative to external benchmarks), and we find that variance in toxicity of subcultures is a much stronger predictor of attrition than variance in compensation. 

To be clear, not saying that compensation does not matter to attrition. Of course it does. Also, low compensation is correlated with other attributes of a job that influence employees' likelihood of quitting. McKinsey or Google will not only pay better, but also offer a variety of amenities to attract and retain employees that workers won't receive at Walmart or Stop & Shop. But when we control for industry or company, we still find toxic culture has a huge impact on employee attrition.
Todd Inskeep
There's a lot of commentary around how compensation is a limited predictor of retention and leaving jobs.  I think trying to isolate hard issues like compensation which is easily measured from softer issues like "toxic culture" understates the role of compensation in employee decisions.  

Said more simply, I think poor pay increases the toxicity of the culture.  Poor pay makes work-life balance harder, poor pay makes schedule control a bigger issue... Poor pay makes everything worse. Its only when you get to a certain level that pay becomes a lower level issue.  I've seen other studies that indicate once a family gets to something like $120-150k and can feel "comfortable" then pay becomes less of an issue.  

I also would expect that pay gets heavily biased by expectations.  And the rising pay differential between executives, the C-suite, and the lower rate roles at most companies factors into many considerations like job stability, performance recognition and even the response to Covid.    Executives are getting 50-500% bonuses (with a variety of tax reduction help...). 

BLUF: Compensation affects more than just the compensation concern.
Donald Sull
@Michael Farrell Great question on age breakdown. When employees fill out a Glassdoor review, they have the option to list their age. Just over one-third of reviewers list their age, so we are reluctant to report out data by age.  Strongly suspect your observation about higher turnover among younger employees is correct on average, but we didn't analyze that.  

We limited our analysis to companies in the Culture 500, which does not include government agencies. At one point we did look at Glassdoor data on government agencies, We found that a much smaller percentage of government employees wrote reviews compared to employees in for-profit companies. The exceptions were the armed forces and the US Postal Service, which do have a large number of reviews.
Donald Sull
@David Mullin. Great point on variance in toxic culture across industries. In a forthcoming article, we'll report out incidence of toxic culture across 40 industries for Culture 500 companies. Spoiler alert--industries with high percentage of front-line employees (retail, fast food), have highest levels of toxicity. Just as you noted.
Donald Sull
@John Doe. The attrition rates for six months April-September 2021, corresponding to the first six months of what the "Great Resignation."
John Doe
Are these attrition rates annualized, or only over 6 months?
David Mullin
It would be interesting to see toxic culture by industry.  I've never worked in a retail setting or food and beverage where someone wasn't causing problems for everybody.
Michael Farrell
Is there an age breakdown?  In my discussions with younger folks, it seems that they expect to change jobs somewhat frequently compared to their older teammates. 

Was there any data on government agencies, be they federal, state or local?
Donald Sull
@Muhammad Hannan. May well be the case that government assistance to employees influenced their propensity to leave jobs. In this analysis, we focus on the intra-firm cultural attributes that predict firm-level attrition relative to its industry. My best guess is that the impact of government assistance to employees would be roughly comparable across firms in the same industry, but that is not my area of expertise.
Donald Sull
@Simon Drake. We are not arguing that toxic culture is the only factor predicting attrition, just that it is the most predictive of those that we used as features in our models. It may well be the case that more consultants became entrepreneurs or migrated to freelancer status. Unfortunately our data does not allow us to test that hypothesis. There is solid evidence that business formation increased during COVID (, although these are companies founded by all entrepreneurs, not just former consultants.
Donald Sull
@James Logan. For an overview of how we developed the topics we used, please see  our earlier article, particularly the section "codifying culture with machine learning and human expertise." We have subsequently validated and refined our taxonomy by linking topics to established concepts in research on culture and also using unsupervised and semi-supervised topic modeling approaches.
James Logan
Interesting approach and insights! Wonder if you could share more about how you derived the 172 topics from the Glassdoor reviews?
Simon Drake
Fascinating article and well structured, thank you. I suspect it's a bit more complicated than just "toxic cultures", though I think this has been a driver for increased attrition and disengagement. Did you research more deeply into the sectors? I'm particularly interested in the high attrition in Management Consulting? Could it be a higher percentage of prospective entrepreneurs starting new businesses? or taking the gig economy freelancer approach, which continues to grow and become more widely accepted/understood? Thanks for sharing.
Timo Marquez
I doesn't surprise me that "toxic culture" is the driver for this massive job/career shift. From the article there are some interesting insights which demand a lot of thought:

☠️ Wasn't there a "toxic culture" (failure to promote diversity, equity, and inclusion; workers feeling disrespected; and unethical behavior) before Covid? Did the Covid context augment the perception of "toxicity" to the point a drastic/fast change was needed?

🌐 Is would be interesting to know the details among factors for the Industries: Apparel (GAP, Guess, Nike, UnderArmor,...), Management Consulting and Internet (GAFAs et al) ? pay, work/life balance, management, other?

👨🏻‍💻 For the "Innovation predictor", It would be interesting to understand this in context of why join in the first place? i.e. Do people join "innovative companies" anticipating they'll leave in few years? Maybe cross the results with a HR onboarding/exit study?

🎬 The short-term actions to boost retention don't address at all the 10x toxic culture. How would the Toxic Culture predictor compare relative to each short-term action ?
Muhammad Hannan
It would be interesting to learn if this study accounted for the unprecedented government assistance for the employees!  From personal experience in food and beverage sector, I do believe that played a major role for blue collar workers to take the step on switching jobs and career with the financial help from extended unemployment benefits nationwide.
Donald Sull
@Dan Holdsworth we're not making any assumptions about composition of employees. In this paper, we are simply predicting each company's attrition rate relative to its primary industry.  And we find that companies that are more innovative are more likely to have higher attrition. 

The question then becomes how do we interpret this finding. Great point that the composition of employees who self select to work in very innovative companies (SpaceX, Tesla, Nvidia) very likely differs from those who choose to work in more staid competitors (Boeing, Ford, Intel). 

Your hypothesis is very plausible that employees who choose to work in bleeding-edge innovators are more likely to jump ship. We know, for instance, that the median tenure of employees in start-ups is about half the average of employees in more established companies. 

But it's possible that the converse is true. Somebody who works at SpaceX could not stand the bureaucracy and boredom of a more traditional competitor. This would make them less likely to quit. 

The matching of types of employees to extremely innovative firms is a super interesting question, but we'd need a different study to shed light on the mechanisms that underlie higher attrition in bleeding-edge firms. 

Thanks for raising the issue!
Dan Holdsworth we're not making any assumptions about composition of employees. In this paper, we are simply predicting each company's attrition rate relative to its primary industry.  And we find that companies that are more innovative are more likely to have higher attrition. 

The question then becomes how do we interpret this finding. Great point that the composition of employees who self select to work in very innovative companies (SpaceX, Tesla, Nvidia) very likely differs from those who choose to work in more staid competitors (Boeing, Ford, Intel). 

Your hypothesis is very plausible that employees who choose to work in bleeding-edge innovators are more likely to jump ship. We know, for instance, that the median tenure of employees in start-ups is about half the average of employees in more established companies. 

But it's possible that the converse is true. Somebody who works at SpaceX could not stand the bureaucracy and boredom of a more traditional competitor. This would make them less likely to quit. 

The matching of types of employees to extremely innovative firms is a super interesting question, but we'd need a different study to shed light on the mechanisms that underlie higher attrition in bleeding-edge firms. 

Thanks for raising the issue!
Dan Holdsworth
      I think that in writing this article, you have made an error in your initial assumptions about employees. You are assuming that the employees of all companies are all the same to start with; I don't think that this assumption is true.

If you instead assume that people who join riskier-seeming, more innovative companies are themselves more inclined to be risk-takers and more inclined to be flighty then much of the observed variation between companies is removed. The employees who jumped ship were already more likely to do this before they joined these companies.