A comparison of hours of work at home estimates between the Current Population Survey and the American Time Use Survey

In this article, I assess the accuracy of estimates of telework hours from the Current Population Survey (CPS), the U.S. Bureau of Labor Statistics (BLS) main labor force survey, by comparing these estimates with estimates of work from home from the American Time Use Survey (ATUS), a time-diary survey also sponsored by BLS. The discussion proceeds as follows, First, I discuss the data on telework and work from home collected in, respectively, the CPS and the ATUS. Second, I compare hours of telework reported in the CPS with hours of work from home reported in the ATUS, both for the population as a whole and for selected characteristics. Third, I compare the total number of hours worked, irrespective of location, between the two surveys.

Measuring telework in the CPS and the ATUS

In October 2022, CPS interviewers began measuring telework by asking respondents: “At any time last week, did you telework or work at home for pay?”¹ If respondents answered “Yes” to the preceding question, then interviewers asked respondents: “Last week, you worked [individual’s total hours worked last week] hours. How many of these hours did you telework or work at home for pay?”² There are two factors that may result in respondents providing inaccurate responses to questions about teleworking. First, respondents may not accurately remember all episodes of work from home over the prior week. Second, because only a single household member answers all CPS questions for a household, proxy reporting for other members of the household may result in errors. 

ATUS data on working from home are generated as follows. An ATUS interviewer asks respondents to describe and report the duration of their activities for a 24-hour period starting at 4 a.m. the previous day. Respondents answer only for themselves. For most activities, respondents also report the location of the activity, so the occurrence and duration of episodes of work from home are recorded.³ This reporting allows estimation of the quantity of work from home based on a respondent recalling a single day.  

Previous researchers have used the ATUS to assess the accuracy of hours-of-work reports in the CPS, so it is natural to compare CPS estimates of telework hours with ATUS estimates of hours of work from home.⁴ However, two important differences between the CPS and the ATUS limit the comparability of the surveys' telework measures. First, and most important, while CPS respondents are asked about “telework” or “work at home for pay,” the ATUS time diary does not distinguish between telework and other categories of work from home. For example, a worker may perform uncompensated work from home. Because telework and other categories of work from home are not distinguished in the ATUS, researchers should expect estimates of hours of work from home in the ATUS to exceed estimates of telework hours in the CPS. (In the next section, I discuss what types of workdays involving work from home are more or less likely to be counted as telework.) Second, the single-day reference period for the ATUS implies that researchers cannot estimate the number of workers who telework over a longer period. For example, workers who work from home 3 days a week may be interviewed about one of the days they work in the office or about a nonworking day. These workers will not be counted as working from home in the ATUS. However, the ATUS will accurately estimate the mean number of hours per week worked at home.⁵

Comparing telework in the CPS and work from home in the ATUS

For the period from October 2022 to December 2023, I compare average hours of telework reported in the CPS with average hours of work from home reported in the ATUS by respondents who report working in the prior week. CPS households (more strictly, addresses, because respondents who move between surveys are not followed) are in the CPS sample for 8 months over a 16-month period (4 months in the survey, 8 months out, and 4 months in). The ATUS sample is drawn from households that have completed their participation in the CPS. The ATUS interview occurs within a few months (typically, 3 months) after the final CPS interview in month-in-sample 8 (MIS8). Only a single member of the household is interviewed for the ATUS. Because the ATUS response rate is lower than the CPS response rate, comparisons of the CPS and the ATUS might be biased if ATUS respondents work a different number of hours from home than the average CPS respondent.

I calculate the difference between ATUS hours of work from home and CPS hours of telework as follows:

where D denotes the difference between hours of work from home between the CPS and the ATUS adjusted for the difference between samples, H denotes hours of work from home, i denotes an individual observation, t denotes the reference month for the estimate (the superscript denotes the survey), and the MIS8 subscript indicates that the observation is in MIS8 (the absence of a third subscript indicates that all months in sample are included). The second term in brackets adjusts for the difference between the ATUS and CPS samples. The adjustment between the ATUS and the CPS samples is calculated by subtracting the total hours of work from home of respondents in both the CPS and the ATUS from the total hours of work from home of respondents only in the CPS. For both samples, hours of work are reported in the last month in sample in the CPS.⁶ I calculate the second term with only ATUS observations from 2023, because ATUS respondents in 2022 were not asked the current CPS work-from-home question when they were interviewed for the CPS 3 months earlier, prior to October 2022.

The CPS telework question asks respondents about their work status during the calendar week before the interview. The reference week is usually the week containing the 12th of the month.⁷ For comparability, I restrict the ATUS sample to observations in which the time-diary day is in a CPS reference week. Harley Frazis and Jay Stewart have shown that hours of work recorded in the ATUS are higher for CPS reference weeks than for other weeks, which is to be expected given the relatively low frequency of holidays during weeks containing the 12th of the month.⁸

I show the results of comparing telework hours from the CPS and the ATUS in table 1.⁹ For all workers, hours of telework reported in the CPS are substantially lower than work-from-home hours reported in the ATUS. CPS telework hours averaged 5.6 hours per week, while ATUS work-from-home hours averaged 9.1 hours per week. The difference in hours between the CPS and the ATUS is reduced from 3.5 to 3.1 hours after adjusting for differences between survey samples. (All differences in hours between the ATUS and CPS mentioned in the remainder of the article are also adjusted for differences between the samples.)

One might suspect that respondents are more likely to regard some types of working days as including paid telework than they are other types of days. Weekdays are arguably more likely than weekend days to be regarded as paid workdays. Periods of working at home that are on the same days as work at other locations may be regarded as extra unpaid work rather than paid telework.

Researchers cannot determine if an entry in the ATUS time diary designates what a respondent would regard as paid work from home or unpaid work from home. However, from 2017 to 2018, the ATUS conducted the Leave and Job Flexibilities Module, which contains questions about telework along with the ATUS regular time-diary questions. In the responses to this module, researchers can distinguish between the following responses: respondents who state that they never work at home, they work at home but are not paid for that work, they work at home and are paid for that work, and they work at home and are partially paid and partially unpaid. Table 2 shows estimates from this module of hours of work from home for workers in various self-described categories of paid and unpaid telework, by categories of workday. In table 2, I refer to workdays that are not on weekends and for which there are no episodes of work outside the home as “standard” work-from-home days, and all other days with work from home as “nonstandard.” Table 2 provides suggestive evidence in favor of the hypothesis that standard workdays are more likely to be regarded as telework days by respondents than nonstandard workdays are. The first row of table 2 shows hours of work from home for respondents who answered “no” to either of the following questions: “As part of your (main) job, can you work at home?” or “Do you ever work at home?” Despite some respondents denying working from home in the module, these respondents have episodes of work from home in their time diary; however, almost all of this work is on nonstandard work-from-home days.

The second through fourth data rows of table 2 show average hours of work from home for different responses to the question: “Are you paid for the hours that you work at home, or do you just take work home from the job?” The possible responses are “paid,” “take work home,” or “both.” For respondents who state that they are not paid for working from home, only a small proportion of their hours of working from home are on standard work-from-home days. For respondents who answer “both” or “paid,” most of their hours working from home are on standard days, with respondents who report “paid” having the highest proportion of standard days.

While there is evidence that at least some hours worked at home on nonstandard workdays are not considered paid telework by respondents, some work from home on these workdays is, undoubtedly, considered paid telework. This is demonstrated in table 2 by the telework hours on nonstandard days shown in the row labeled “Works at home, paid.” Restricting telework hours to standard work-from-home days may still provide a useful rough lower bound for paid telework. Accordingly, the sixth column of table 1 shows hours of telework restricted to standard days, with ATUS respondents working 7.5 hours per week at home on such days. The sample-adjusted difference in telework hours per week between the CPS and the ATUS is about 1.6 hours, which is approximately half the difference observed when work-from-home hours are not restricted to standard days in the ATUS.

I examine the association of cross-survey differences in reported telework with other characteristics in the lower part of table 1. The weighted percentage of the sample accounted for by workers with given characteristics is shown in table 3. One characteristic of interest is the suitability of an occupation for telework. I use a classification devised by Jonathan I. Dingell and Brent Neiman to classify detailed occupations (in the worker’s main job) as suitable or unsuitable for telework.¹⁰ As expected, occupations suitable for telework have substantially greater hours of working from home reported in both surveys than occupations not suitable for telework. However, the difference in telework hours between CPS and ATUS is similar between groups.

Educational attainment has been shown to be associated with telework, even after taking into account the suitability for telework of the detailed occupation.¹¹ Rows 6 and 7 of table 1 show that workers with a bachelor’s degree or a higher level of educational attainment report more hours of telework than workers with lower levels of educational attainment. The difference between hours of work from home in the CPS and the ATUS are of comparable magnitude across education groups. Similarly, as shown in rows 4 and 5 of table 1, the difference between the CPS and the ATUS estimates of hours worked at home is not statistically significantly different between men and women.

Rows 8, 9, and 10 of table 1 show results by class of worker (wage and salary or self-employed) for single jobholders and for multiple jobholders. Although the estimates in the table are imprecise, both self-employed workers and multiple jobholders reported substantially more work from home in the ATUS than in the CPS: 7.5 more hours of work from home for self-employed workers and 6.7 more hours of work from home for multiple jobholders. Wage and salary workers reported 2.5 more hours of work from home in the ATUS than in the CPS. The difference between the ATUS and the CPS is statistically significant at the 10-percent level for work-from-home hours of self-employed workers and wage and salary workers.

When I restrict the analysis to standard days, I no longer find a statistically significant difference between ATUS and CPS reports. But the case for restricting the analysis to standard workdays is weaker for multiple jobholders because they work, on average, more than twice as many hours on weekends than do single jobholders (3.1 hours and 1.4 hours, respectively).¹²

To investigate whether one source of the difference between the surveys is the presence of proxy responses in the CPS, I divide the sample by whether ATUS and CPS responses were by the same individual or whether CPS responses were by a proxy. As would be expected, the point estimates shown in rows 11 and 12 of table 1 indicate that the discrepancy in reported work-from-home hours between the CPS and the ATUS is greater for proxy respondents than for self-respondents. However, the estimated CPS–ATUS differences in work-from-home hours between self-respondents and proxy respondents are within an hour of each other and are not statistically different at standard levels of significance.

Results for total hours of work

The results presented in the previous section raise an interesting question: do the growth of telework since the COVID-19 pandemic and the substantial difference between work-from-home hours reported in the CPS and the ATUS imply that total hours of work, inclusive of work outside the home, are also greater in the ATUS? Prepandemic research by Harley Frazis and Jay Stewart found that the total hours of work reported in the CPS and the ATUS are similar. But if reported hours of work outside the home remains equal between the surveys, then the greater number of hours of work from home reported in the ATUS would also imply a greater number of total hours of work reported in that survey.¹³

Table 4 shows estimates of total hours of work in the CPS and the ATUS. Unlike reported hours of work from home, the total reported number of hours is little different between the CPS and the ATUS. Total reported hours of work outside the home (derived as total hours of work minus hours of work from home) are in fact greater in the CPS than in the ATUS. The difference between the CPS and the ATUS in total reported hours of work outside the home are roughly equal in magnitude (and of opposite sign) to the difference in hours of work from home between the surveys. The standard errors of these differences (not shown) are approximately equal to those shown in table 1.

Summary

From October 2022 to December 2023, ATUS respondents reported 9.1 hours per week of work from home hours, substantially higher than the telework hours reported by CPS respondents (5.6 hours per week). Part of this difference of over 3 hours may be accounted for by the CPS question referring to “work at home for pay.” Respondents may view some work at home as catching-up work and not as paid work. Evidence from the 2017–18 ATUS Leave and Job Flexibilities Module is consistent with this view. However, excluding weekend work and work that occurs on the same day as work away from home from estimates of telework still leaves a difference of about 1.5 hours, which suggests a substantial response bias in the CPS. The discrepancy in hours of work from home does not affect comparisons of estimates of total hours worked between the ATUS and the CPS. This is because reported hours of work, aside from telework, are greater in the CPS than hours of work outside the home are in the ATUS.