Abay Asfaw
Purpose/Objectives
The objective of this study is to quantify the impact of race and ethnicity on the incidence of teleworking and examine to what extent this impact is mediated by education and occupation. Quantifying the relative role of mediators in explaining racial and ethnic teleworking disparities would help inform intervention efforts and best practices.
Background
Studies have shown that the incidence of teleworking is lower among non-Hispanic Blacks and Hispanics than among non-Hispanic Whites and non-Hispanic Asians. However, little is known about the underlying causes of racial and ethnic disparities in teleworking or about the mechanisms through which race and ethnicity affects teleworking. In this study, we hypothesized that racial and ethnic disparities in teleworking are mediated by educational attainment and occupations. Education is expected to have a substantive influence on teleworking by affecting employment trajectories including pay level, titles, occupations, and positions. Racial and ethnic disparities in the proportion of workers employed in different occupations is also expected to have an impact on the ability of workers to telework.
Methods
Methods
Data: The data source for this study was the Current Population Survey (CPS) sponsored by the Bureau of Labor Statistics (BLS) and the U.S. Census Bureau. CPS is a nationally representative monthly survey that provides comprehensive data on the labor force characteristics and other demographic information of the U.S. population. Starting in May 2020, BLS added five questions to the CPS to measure the effects of the COVID-19 pandemic on the labor market. One of the questions is “At any time in the last 4 weeks, did you telework or work at home for pay because of the coronavirus pandemic?” We used this question to measure the incidence of teleworking during the pandemic – 1 if the respondent teleworked and 0 otherwise. We used data from May 2020 through March 2021. Overall, we considered 508,495 respondents who were working during the survey weeks and for whom there was no missing information on all covariates and mediator variables. These respondents represent 143.9 million workers. “Race and ethnicity” was our independent variable. The CPS collected information on the race of respondents as White, alone; Black, alone; Asian, alone; and any other race alone or race in combination. The survey also collects information on the ethnicity of respondents as Hispanic and non-Hispanic. We used this information and created the race and ethnicity variable as “non-Hispanic White (White)”, “non-Hispanic Black (Black)”, “non-Hispanic Asian (Asian)”, and “Hispanic”. Workers whose ethnicity was identified as Hispanic could be of any race. We did not consider “any other race and mixed races” because their share in the total workforce was small (2.5%). We used “college education” and “occupation” as mediator variables (see Figure below). In the survey, respondents were asked about the highest grade they completed. We used this information to identify respondents with college education. Respondents were also asked about their occupation. We used the BLS Standard Occupational Classification (SOC) system to aggregate specific occupations to 22 two-digit SOC. We included sex, age groups, marital status, number of children in the household, and full-time or part-time status as control variables. We also included survey month (May 2020-April 2021) as a covariate in order to account for trends in teleworking since May 2020.
Analysis: We were interested in explaining racial and ethnic disparities in teleworking during the COVID-19 pandemic. We were also interested to understand the pathways, whereby race and ethnicity might affect teleworking during the pandemic. One possible explanation is differences in college education and occupation among different racial and ethnic groups. In other words, these factors are assumed to be a step in the pathways between race and ethnicity and the incidence of teleworking. To empirically examine this relationship, we included these variables as mediator variables. Education might affect occupation. However, occupation was included as a separate mediator variable because several other factors might also affect occupation.
As shown in the Figure, race and ethnicity can directly affect teleworking during the pandemic or indirectly by affecting the mediator variables. The mediator variables, may partially, or entirely, explain the association between race and ethnicity and the teleworking.
In a linear model, the problem can be specified as:
T=α_a+β_a R+γ_a X+ε Reduced model (1)
T=α_b+β_b R+δM+γ_b X+µ Full model (2)
Where T is the outcome variable, R is “race and ethnicity”, X is a vector of covariates, M is a vector of mediator variables, and ε and µ are error terms. βa measures the total and βb the direct impact of race and ethnicity on teleworking during the pandemic. The difference between βa and βb measures the indirect effect of race and ethnicity on the ability of workers to telework because of the pandemic. This approach would help to examine the extent to which the impact of race and ethnicity on teleworking can be explained by variations in education and occupation. In other words, δ measures how much of the influence of race and ethnicity on teleworking was mediated by education and occupation.
As indicated above, our outcome variable was binary. Therefore, we estimated a logistic regression and used a user-written Stata command ‘KHB’ which decomposes direct and indirect effects for nonlinear models (Kohler et al., 2011). In the CPS data most of the respondents were interviewed more than once and this might violate the assumption of independence of observations. To address this problem, we used the Huber/White/sandwich estimator. We also used the survey weights to achieve national representativeness and to correctly incorporate oversampling of specific subgroups groups.
Findings
Table 1 presents the total, indirect, and direct effects of race and ethnicity on teleworking. The second column of Table 1 presents the total effect of race and ethnicity on teleworking not adjusting for the mediators. During May 2020 through March 2021, the odds of Black and Hispanic workers to telework were 38% and 54% lower than that of White workers, respectively, controlling for covariates. On the other hand, during the study period, the odds of Asian workers to telework was 2.1 (95% CI: 2.00-2.24) times higher than that of White workers, controlling for covariates. Using the KHB method, we decomposed these total effects to estimate the indirect and direct effects of race and ethnicity on teleworking. For Black and Hispanic workers versus White workers, education and occupation explained 90% and 91% of the total effects of “race and ethnicity” on teleworking, respectively. However, the directs effects were also statistically significant. After controlling for mediators and covariates, the odds of Black and Hispanic workers to telework were 5% and 8% less than that of White workers. For Asian workers, however, the mediator variables explained only half of the effect of race and ethnicity on teleworking. The direct effect of race and ethnicity for Asian workers showed that the odds of Asian workers to telework was 46% higher than that of White workers, after controlling for mediators and covariates.
The next important issue was to examine which of the mediators contributed most (negatively or positively) to teleworking. Compared to White workers, college education decreased the likelihood of teleworking for Black and Hispanic workers. High percentage of Black workers relative to White workers employed in transportation and material moving, healthcare support, building and grounds cleaning and maintenance, protective service, production, and food preparation and serving occupations decreased their likelihood of teleworking. Similarly, high percentage of Hispanic workers relative to White workers employed in construction and extraction, building and grounds cleaning and maintenance, transportation and material moving, food preparation and serving, and production occupations reduced their likelihood of teleworking. For Asian workers, college education increased their likelihood of teleworking followed by their high share in computer and mathematical occupations compared to White workers.
Discussion
Teleworking is one strategy used by employers to reduce the spread of COVID-19. In addition to facilitating flexibility and improving work-family balance, teleworking reduces the risk of COVID-19 exposure and other influenza like illnesses (ILI). However, despite an increase in the percentage of workers teleworking since the pandemic started, marked racial and ethnic disparities persist. During May 2020 through April 202, compared to White workers, the odds of Black and Hispanic workers teleworking was 38% and 54% lower, respectively, controlling for covariates. On the other hand, the odds of Asian workers of teleworking was more than double that of White workers, controlling for covariates. Prior studies on racial and ethnic disparities in teleworking have not examined how these differences were mediated. We used a mediation analysis to measure the role of education and occupation in racial and ethnic teleworking disparities. The results showed that college education increased the likelihood of teleworking for Asian workers compared to White workers but not for Black and Hispanic workers compared to White workers. This might indicate that compared to White workers, college education did not improve the likelihood of Black and Hispanic workers to telework. Concentration of Black workers in occupations such as transportation and material moving, healthcare support, and protective services and Hispanic workers in construction and extraction, building and grounds cleaning, transportation and material moving, and food preparation occupations significantly reduced their likelihood of teleworking. There could also be some multicollinearity between college education and occupation, and this might underestimate the impact of these variables on teleworking.
Conclusions
Racial and ethnic disparities in the proportion of workers employed in different occupations could be one of the reasons for racial disparities in teleworking. Addressing these structural issues could help reduce racial disparities in teleworking and consequently disparities in COVID-19 and other ILI infections at work. Further analysis might explore why the impact of college education on teleworking was negative for Black and Hispanic workers.