Since 2016, we’ve conducted an annual equity review of Glassdoor’s own employee compensation as part of our ongoing efforts to promote workplace transparency and pay equity. This pay equity report builds on Glassdoor’s long history of promoting transparency in the workplace and our recent reinvigorated commitment to radical transparency.
We are pleased to report that we find no adjusted pay gap by gender or race/ethnicity at Glassdoor in 2021. This is the sixth consecutive year we’ve found no statistically significant adjusted gaps by gender (2020, 2019, 2018, 2017, 2016). This is also the third year we’ve analyzed pay by race/ethnicity and, similarly, we find no statistically significant adjusted pay gaps. Each year, we use an approach similar to the methods described in our guide, which outlines how employers can analyze pay equity within their own organizations.
Our Data and Methodology
For our analysis, we collected demographic and payroll data for our U.S. full-time employees as of June 1, 2021. We define an employee’s total pay as the sum of an employee’s annual salary, reflecting any changes from the spring review cycle, and any bonuses/commissions they received in the 6-month period covered by the review cycle.
To create a more apples-to-apples comparison of pay between similar employees, we estimate adjusted pay gaps. To calculate the adjusted pay gap, we estimate a linear regression regressing the logarithm of pay on a set of controls, including demographic characteristics like race and gender, job details like title, department, level and tenure, geography and performance ratings.
If you’d like to learn more about our approach, we’ve included three explainers at the bottom of this post to discuss:
- Our Methodology: A more detailed description of our methodology, including who and what is included in our analysis.
- Unadjusted vs. Adjusted Gaps: What is the difference between adjusted and unadjusted pay gaps? What are the benefits and challenges associated with each metric?
- Statistical Significance and Confidence Intervals: What do “statistical significance” and “confidence intervals” mean? How do we use them to interpret results?
Our Findings: No Adjusted Pay Gaps by Gender or Race/Ethnicity among Glassdoor Employees
We find no statistically significant adjusted pay gaps at Glassdoor in 2021, after controlling for differences between employees including factors like job title, location, tenure, level and performance ratings. The adjusted pay gap provides a more apples-to-apples comparison that assesses whether we are offering “equal pay for equal work”.
Chart 1 shows the unadjusted (more on this further below) and adjusted pay gaps by gender and race/ethnicity. After including controls, the unadjusted pay gaps shrink. For women, the unadjusted pay gap is 19.6 percent, but after controls, the gap shrinks to a statistically insignificant 1 percent. Similarly, the adjusted pay gaps for Asian, Hispanic/Latinx and Black workers shrink to statistically insignificant after including controls. The thin lines on each column are error bars. If the error bar crosses zero, then the difference is not statistically significant, and in the chart below, you can see none of the adjusted pay gaps are statistically significant.
The unadjusted pay gaps shown in Chart 1 are wider than the adjusted pay gaps because the unadjusted pay gaps don’t take into account differences in observable characteristics of employees like job title, level or tenure. In particular, unadjusted pay gaps are driven in large part by differences in representation in higher-paying roles, levels or departments, a trend economists call “occupational sorting.” For example, if women are underrepresented in higher-paying fields like engineering, then the unadjusted pay gap will widen. Occupational sorting accounts for over half of the unadjusted gender pay gap economy-wide, according to our research from 2019.
Glassdoor has set goals to hire more employees from underrepresented groups, especially into higher-paying technology roles and leadership, which should naturally help mitigate the effect of occupational sorting and, thus, the unadjusted pay gap. Maintaining no adjusted pay gaps while improving representation of underrepresented groups should help us reduce unadjusted pay gaps over time.
Conclusion
We’re pleased to report that after accounting for differences in jobs, location, tenure and other observable factors, we find no adjusted pay gaps by gender or race/ethnicity among U.S.-based Glassdoor employees in 2021. This remains unchanged from our previous analyses in the previous five years.
At Glassdoor, we believe in the power of transparency to empower employers, employees and job seekers alike to identify and address problems. In particular, we believe pay transparency is a powerful tool in the fight for pay equity. In addition to offering pay data to our users on Glassdoor, sharing our analysis publicly represents Glassdoor’s commitment to hold itself accountable as it continues to advocate for pay equity.
We remain committed to furthering workplace transparency and promoting pay equity. In addition to publishing our own pay checkup, we urge employers to utilize our free employer guide to examine their own pay practices.
Explainer: Our Methodology
To calculate the adjusted pay gap, we estimate a linear regression regressing the logarithm of pay on a set of controls. Log-transforming pay helps to prevent a few high salaries from skewing the average, a common issue in pay data. Our controls include demographic characteristics like race and gender, job details like title, department, level and tenure, geography and performance ratings. The adjusted pay gaps we report are the coefficients estimated for the gender or race dummy variables when including all controls.
Our measure of total pay includes annual salary looking forward and bonuses/commissions paid over the prior 6-month period. Because our analysis excludes some compensation like RSUs and bonuses/commissions from the prior semiannual review cycle, our final measure of total pay may underreport employees’ actual annual total compensation.
Our analysis is limited to employees based in the United States and paid in U.S. dollars. Employees who did not or declined to provide demographic information were included in the analysis as “Missing” or “Declined to Identify”. To protect the privacy of smaller groups, we only report gender pay gaps for women compared to men and race pay gaps for Black, Hispanic/Latinx and Asian employees compared to white employees.
Explainer: Unadjusted vs. Adjusted Gap
Our pay equity analysis calculates two key metrics: (1) the unadjusted pay gap and (2) the adjusted pay gap. The unadjusted pay gap is the percentage difference in a group’s average pay relative to the reference group. For example, if the average pay for men is $100,000 and the average pay for women is $80,000, the unadjusted pay gap will be 20 percent. That 20 percent gap could be converted into the statement that women make 80 cents for each dollar men make, a popular way to describe the gender pay gap.
One challenge the unadjusted pay gap metric faces is that it doesn’t explain why a gap exists. Our past research indicates that an important driver of the gender pay gap is “occupational sorting”, the underrepresentation of women in higher-paying roles.
To form a more apples-to-apples comparison, we can use the adjusted pay gap. The adjusted pay gap uses statistical methods to compare pay for employees in similar roles with comparable experiences, levels and backgrounds. This helps give a better understanding whether there is “equal pay for equal work”.
The challenge with the adjusted pay gap is that it can statistically remove differences that are driven by gender or race/ethnicity. For example, if women are underrepresented in higher levels of management because they are passed over for promotion on the basis of their gender, controlling for management level may result in the finding that pay within a management level is equitable even if promotion into that level is not. Similarly, while performance ratings are often pointed to as a potential source of bias, including them as a control is still important to make sure we’re comparing apples-to-apples. In sum, the adjusted pay gap does not guarantee bias does not exist, but helps give a holistic picture of the state of pay equity.
The focus of our pay equity analysis therefore is the adjusted pay gap to ensure we are paying employees equally for equal work. However, we also report the unadjusted pay gap to be transparent about the state of pay at Glassdoor. Because the unadjusted pay gap is in large part driven by underrepresentation of women and minorities in higher-paying roles, our goals to hire more underrepresented groups into those higher-paying roles should help us reduce our unadjusted pay gap over time.
Explainer: Statistical Significance and Confidence Intervals
The adjusted pay gap we report in our analysis is a statistical estimate. As such, unexplained differences in pay can make us less confident in our estimate. Similarly, we may become less confident in our estimate for smaller groups where there is less data to work with.
We consider our estimate to be statistically significant if it falls within a 95 percent confidence interval, meaning that we are 95 percent confident that the value we’re estimating actually falls within the interval.
When we say we find no statistically significant pay gap, we mean that our 95 percent confidence interval includes zero percent (or no gap), which means we cannot statistically distinguish the gap we estimate from no gap at all. It does not guarantee no gap exists, but simply means we cannot observe one with the data we have.
As we collect more data and as Glassdoor hopefully grows, we should be able to more precisely estimate the adjusted pay gap, thereby shrinking the confidence interval. And the smaller the confidence interval, the more confident we can be that there are no adjusted pay gaps.
