Teleworking and remote schooling during the COVID-19 pandemic

The COVID-19 pandemic triggered a large increase in the amount of time that employees spend working at home. For example, analyzing data from the U.S. Bureau of Labor Statistic (BLS) American Time Use Survey and the 1979 cohort of the National Longitudinal Survey of Youth, Dey, Frazis, Loewenstein, and Sun estimate that immediately prior to the pandemic only a little more than 10 percent of workers teleworked 1 or more days per week.¹ In comparison, an analysis of a COVID-19 supplement of the 1997 cohort of the National Longitudinal Survey of Youth (NLSY97), by Aughinbaugh, Groen, Loewenstein, Rothstein, and Sun, finds that in the spring of 2021 approximately 46 percent of workers teleworked at least some of the time during the week before they were interviewed and about 25 percent teleworked the entire week.² Teleworking rates have fallen from their height at the start of the pandemic, but teleworking is still far more common than before the pandemic.³ Some well-known employers have now required that employees work in the office full time, but there still is little doubt that telework rates will remain substantially above their prepandemic rates.⁴

Another important feature of the pandemic was the increased incidence of remote learning to limit the spread of COVID-19. All U.S. public school buildings were closed by March 25, 2020.⁵ School disruptions continued well into the following school year. Estimates from the National Assessment of Educational Progress Monthly School Survey indicate that in February 2021 only 49 percent of public schools with fourth grade or eighth grade were open full time and in person for all students.⁶ By May 2021, this percentage was still only 63 percent.

In spring 2021, the NLSY97 fielded a supplement survey on the effects of the COVID-19 pandemic. The supplement interviews were conducted from February to May 2021. The supplement data include information on employment, telework, health, and children’s schooling. NLSY97 respondents were between ages 36 and 41 at the time of the supplement. Aughinbaugh, Groen, Loewenstein, Rothstein, and Sun provide a detailed description of the supplement.⁷ For present purposes, key pieces of information collected by the survey are the hours that individuals worked and teleworked in the week prior to the survey, whether children attended school in person or remotely, and the job characteristics that determine how suitable an individual’s job is for teleworking.⁸

The NLSY97 COVID-19 Supplement survey offers a unique opportunity to study the relationship between remote schooling and teleworking. While parents no longer need to contend with remote schooling, an analysis of the relationship between remote schooling and parents’ decision to telework provides insights into how the flexibility allowed by teleworking enables individuals to better manage the demands of childcare and other household responsibilities. This flexibility may be especially important for women. Furthermore, the advantages of teleworking are not available to all workers. Only a minority of jobs are well suited to working at home; the majority of jobs must be performed on site. Generally, teleworking is more prevalent among more highly skilled and higher paying jobs. The NLSY97 COVID-19 Supplement survey enables one to examine how the characteristics of individuals’ jobs determine their ability to use teleworking to cope with remote schooling and, by implication, other household demands on their time.

The next section provides a description of the data in the NLSY97 COVID-19 Supplement as well as data from additional sources. Teleworking and remote schooling equations are then presented in the data analysis section. My primary interest is in the effect that remote schooling had on teleworking. Because a substantial number of children with the option to attend school in person in the spring of 2021 may have chosen to attend remotely, endogeneity of remote schooling is a potential issue.⁹ However, after estimating a two-stage least squares equation, I am able to rule out the endogeneity of remote schooling. I complete the empirical analysis by examining how parents’ response to remote schooling by their children depended on how suitable their jobs were to teleworking and whether the response differed if only one parent was present in the household. I summarize my findings in the final section.

Data description

The analysis in this article is based on individuals who participated in both the NLSY97 COVID-19 Supplement and the previous NLSY97 round 19 data collection. The ages of individuals in this sample range from 36 to 41. In this article, my focus is on the teleworking behavior of individuals who worked during the week prior to the survey. There are 3,918 individuals in the supplement survey who worked in the week prior to the survey and for whom we have occupational information from the round 19 collection. After dropping individuals who had missing data, I ended up with a usable subsample of 2,589 observations.

Table 1 presents summary statistics for the variables used in this article. Here I highlight several variables that are key to the analysis or that come from sources other than the NLSY97. The teleworking variable used throughout this article is the number of hours the respondent worked at home in the week prior to the interview. In the analysis that follows, I use an indicator variable that equals 1 when the respondent worked at least 10 hours at home in the week prior to the survey as a teleworking variable. There is admittedly some arbitrariness to the choice of any cutoff, but I want to exclude incidental teleworking of just a few hours. Another candidate for a cutoff value would be 8 hours. I chose 10 hours because there is more bunching at 10 than 8 hours. The key results are virtually identical if one chooses an 8-hour cutoff. The results are also similar if I choose a 20-hour cutoff.

The NLSY97 has information on whether any children in the household were enrolled or educated in a public school, a private school, or a home school. In addition, the survey has information on whether children in the household attended any classes in person and whether they attended any classes remotely.¹⁰ The survey has useful information on the composition of the respondent’s household. I use variables indicating whether there are children younger than age 6 and between ages 6 and 17 in the household.

Information on how suitable an individual’s job is for working at home comes from three sources. First, in the previous round 19 data collection, individuals were asked about the tasks they performed on the job. For example, there is information on whether half their workday or more is spent doing physical tasks and whether there is a great deal of face-to-face contact with people other than coworkers or supervisors.¹¹ Second, the National Longitudinal Survey has information on an individual’s occupation. Using O*NET, Dey, Frazis, Loewenstein, and Sun have applied the Dingel-Nieman framework to create a 0–1 variable indicating an occupation’s suitability for telework.¹² I refer to this indicator as Teleworkable1. Third, Dalton, Dey, and Loewenstein use the Business Response Survey and the Occupational Employment and Wage Survey, both collected by BLS, to estimate the proportion of workers in an occupation that teleworked in the summer of 2021.¹³ I call this variable Teleworkable2.

In this article, I use two additional pieces of information from other sources: Google cell phone location data and Safe Graph cell phone data. Google cell phone location data, which provides information on visits to workplaces in the respondent’s round 19 county of residence, allows one to measure the change in county-level activity at workplaces between a baseline period before the COVID-19 pandemic (January 3, 2020, to February 6, 2020) and the period of the supplement interview (February 2021 to May 2021).¹⁴ Greater reductions in activity at workplaces in the spring of 2021 were associated with increased COVID-19 related restrictions and greater reluctance, on the part of employees, to head into the worksite. Consequently, the change in county-level activity should be negatively correlated with teleworking.

Safe Graph cell phone data provides a measure of reduced school activity. In my analysis, I use a countywide measure of the percentage of “schools closed” that has been made available by Parolin and Lee.¹⁵ This measure counts a school as closed if cell phone calls from it in the spring of 2021 had fallen by 50 percent or more from the period before the COVID-19 pandemic. Note that while there was reduced school activity at a school that was counted as “closed,” there may have been some in-person learning taking place. I will therefore refer to this variable as the proportion of schools that were at least partially closed. Naturally, there was a greater likelihood that a respondent’s children attended school remotely when an increased proportion of schools were at least partially closed.

Data analysis: basic specification

The following equation expresses the incidence of telework T as depending on remote schooling incidence R, job characteristics JC, and all other variables X:

(1)          .

The results of estimating equation (1) are presented in table 2.¹⁶ The coefficient on the variable remote school is positive and highly statistically significant, indicating that parents of children attending school remotely were 8 percent more likely to telework than parents of children who were attending schooling in person.¹⁷ As expected, the county level change in workplace activity is also negative and statistically significant.

The coefficients on Teleworkable1 and Teleworkable2 are positive, as expected, and statistically significant. The task variables also generally have the expected sign. For example, teleworking is less likely in jobs that involve a lot of physical tasks or face-to-face contact. In the analysis that follows, I construct a single variable that captures all of the available information on how suitable a job is for teleworking. I do this by using the estimated coefficients on the JC variables. Specifically, I let , where  are the coefficient estimates for b₂.¹⁸ By construction, if one estimates equation (1) as

(1′)   ,

then the estimate of the Teleworkable coefficient b will equal 1 and the estimated coefficients b₀ and b₁ on R and X will be exactly the same as the estimated coefficients on R and X in equation (1).

Table 2 shows the estimated teleworking equation for the female and male subsamples. The results are broadly similar. However, the coefficient on remote schooling is substantially larger in the female equation. In fact, the estimated coefficient is not statistically significant in the male equation. I examine the female and male responses in more detail in the next section.

Parents who teleworked may have found remote schooling more manageable and, therefore, may have been more likely to choose remote schooling for their children when given the option. If so, then the estimated coefficient on remote schooling in the teleworking equation is biased upward.¹⁹ Another obvious variable affecting the likelihood that a child attends school remotely is the countywide partial school closure rate. Letting the variable Clos denote the school closure rate, the incidence of remote schooling can be written as follows:

(2)           .

Substituting equation (1′) into equation (2) yields the following reduced form equation:

(2′) ,

where  .

If teleworking makes remote schooling more likely, the coefficient  on Teleworkable should be positive. The results of estimating equation (2′) are presented in table 3.²⁰ The coefficient on the school closure rate is large and statistically significant, indicating that the school closure rate has the potential to serve as a powerful instrument. Other coefficients of note are the positive coefficients on the presence of children ages 6 to 17 in the home and on the indicator that the respondent is Black, as well as the negative coefficients on the central, southern, and western region indicators. However, the estimate of  is wrong signed and statistically insignificant, providing evidence that remote schooling can be taken as exogenous in the teleworking equation. Further evidence is provided by the two-stage least squares estimates presented in table 2.²¹ Although not quite statistically significant at the 95-percent confidence level given the high standard error, the coefficient on remote schooling is positive and large. Moreover, the Durban chi-squared and Wu-Hausman F tests fail to reject exogeneity of remote schooling at any conventional confidence level. 

Estimates of the reduced form remote schooling equation for the female and male subsamples are presented in table 3, and estimates of the corresponding two-stage least squares equations can be found in table 2. The results for the subsamples are in accord with those for the overall sample. In the remainder of the article, I focus on these female and male subsamples.

Remote schooling interactions

Not all parents were equally able to work at home when their children attended school remotely. Individuals in jobs more suitable for teleworking would have been better able to work at home in response to the demands on their time brought about by remote schooling. One would also expect the effect of the change in workplace activity to depend on how suitable an individual’s job is for teleworking. For the sake of generality, I also add interactions of Teleworkable with the other explanatory variables so that the telework incidence equation becomes the following:

(3)      

.

Table 4 shows the results of estimating equation (3) for the female and male subsamples. The only statistically significant coefficients in both the male and female equations are Teleworkable, remote schooling for sufficiently high values of Teleworkable, and the change in the county-level activity at workplaces. The estimation results clearly show that parents’ response to remote schooling depended crucially on how suitable their jobs were for teleworking. The estimated effect of remote schooling on the telework incidence of females is small and not statistically different from zero at the 10th percentile of the Teleworkable variable, but the effect is very large at higher values of the Teleworkable variable. At the median value of Teleworkable, remote schooling led to a roughly 11.5-percentage-point increase in the likelihood of teleworking. At the 75th percentile of Teleworkable, the increase in teleworking is about 17 percentage points. 

Remote schooling by a parent’s children had a smaller effect on the telework incidence of male parents than female parents. At the median value of Teleworkable, the estimated effect of remote schooling on the likelihood that men teleworked is small and statistically insignificant. However, the estimate effect is larger for higher values of Teleworkable. At the 75th percentile of Teleworkable, remote schooling led to an approximately 9-percentage-point increase in the likelihood that men teleworked.

The flexibility provided by teleworking in responding to remote schooling may have been especially useful to single parents having the sole responsibility for childcare. To test this possibility, I interact remote schooling with an indicator for whether the spouse or partner was present in the household. Denoting this indicator with the variable named Absence, I estimate the following equation:²²

(4)

Estimation results are presented in table 4. The estimation results indicate that the absence of a spouse or partner magnified the effect of remote schooling on mothers’ teleworking incidence, with the impact being larger the more suitable the job for teleworking. At the median value of Teleworkable, remote schooling led to about a 22-percentage-point increase in the likelihood of teleworking when the spouse was absent, compared with a roughly 9.5-percentage-point increase when the spouse was present.²³ At the 75th percentile of Teleworkable, the corresponding amounts are about 38 and 13.5 percentage points, respectively.

The estimated interaction of spousal or partner absence with remote schooling is quite imprecise for men. The estimated interaction is signed wrong and not statistically different from zero.

Summary

The COVID-19 pandemic resulted in a very large increase in teleworking. In addition, school closings led to a large number of students attending school remotely. The results from the NLSY97 COVID-19 Supplement survey conducted in the spring of 2021 make it possible to examine the relationship between these two occurrences. Thirty-two percent of parents in the sample, whose children were enrolled in school, worked at home 10 hours or more in the week prior to the time they were surveyed. My estimates indicate that remote schooling attendance by children increased the likelihood that parents worked at home.²⁴ 

Not surprisingly, the responsiveness of teleworking to remote schooling depended crucially on how suitable an individual’s job was to teleworking. Using information on individuals’ occupations and other characteristics of their jobs, I have constructed the variable Teleworkable as an indicator of how suitable an individual’s job is to teleworking. Remote schooling by a child had no effect on the likelihood that a parent teleworked if their job was not well suited for teleworking. However, the estimated effect was substantial when the parent's job was well suited for teleworking. Remote schooling by a child had an especially large effect on teleworking by women in telework suitable jobs.²⁵ For example, my estimates indicate that at the median value of Teleworkable remote schooling increased the likelihood that women teleworked by 11.5 percentage points. At the 75th percentile of Teleworkable, this figure is about 17 percentage points. In contrast, at the median value of Teleworkable, remote schooling had a small and statistically insignificant effect on the likelihood that men teleworked. But at the 75th percentile of Teleworkable, remote schooling caused the likelihood of teleworking by men to increase by roughly 9 percentage points.

Women’s response to remote schooling by their children depended on whether their spouse was present in the household. For instance, at the median value of Teleworkable, my estimates indicate that remote schooling by their children increased the likelihood that women teleworked by about 22 percentage points when the spouse or partner was absent, compared with 9.5 percentage points when the spouse was present. The degree to which the absence of a spouse or partner magnified the effect of remote schooling was greater for jobs that were better suited for teleworking. For instance, at the 75th percentile of Teleworkable, remote schooling by their children increased the likelihood that women teleworked by about 38 percentage points when the spouse or partner was absent, compared with 13.5 percentage points when the spouse was present

While parents no longer need to contend with remote schooling, the flexibility allowed by jobs that are well suited for teleworking enables individuals in such jobs to better meet the demands of childcare and other household responsibilities. For example, parents with jobs that allow them to work at home may be able to telework on the days that their children cannot attend school because children are sick or when bad weather or a holiday results in a school closure. This consideration may be more important for women, who still seem to bear the majority of household responsibilities, than for men. The increased flexibility provided by teleworking may be especially important for one parent households.