Hourwise

How official statistics count a workweek

Three separate systems produce a number called hours of work. They were built at different times, for different purposes, and they do not measure the same thing.

Three instruments, one label

The first instrument is household recall. In the United States, the Current Population Survey — run by the Bureau of Labor Statistics with the Census Bureau — asks respondents how many hours they usually work each week, and separately how many hours they actually worked in the reference week. Those two questions produce different distributions, and both are reported.

The second is the employer payroll record. The Current Employment Statistics programme collects hours from establishment payrolls. What a payroll knows is hours an employer paid for, which is not the same quantity as hours a person worked: paid leave can enter the count, and unpaid work outside scheduled time cannot.

The third is the time-use diary. The American Time Use Survey, fielded since 2003, asks a respondent to reconstruct a single designated day in sequence, from one activity to the next, without asking about work in the abstract at all. Hours of work are then computed from the reconstruction rather than reported by the respondent.

Where the instruments part company

John P. Robinson and Geoffrey Godbey, working with diary collections assembled over several decades, argued in Time for Life that diary-based estimates of paid work were lower than the figures the same population reported when simply asked. Robinson and Ann Bostrom had set out the argument for a general audience in the Monthly Labor Review a few years earlier. The gap they described was not uniform: it was small for people reporting moderate schedules and widened considerably among those reporting the longest weeks.

Jerry A. Jacobs examined the accuracy of self-reports in the same journal and reached a more qualified conclusion, noting that the discrepancy depends heavily on which comparison is made and on how the diary day is weighted up to a week. Harley Frazis and Jay Stewart, both economists at the Bureau of Labor Statistics, later worked directly on why the agency's own hours series move differently from one another, treating the divergence as a reconciliation problem between definitions rather than as one series being wrong.

What each method gives up

A diary avoids asking for a summary judgement, which is its main advantage. The cost is that it observes one day. Individual weekly totals reconstructed from a single day are noisy, and the method works best across large populations rather than for any one respondent. Diaries also record a primary activity, so work carried out alongside something else can be undercounted unless secondary activities are collected and used.

Recall questions cover the whole week, which is what most analysis needs, but they ask for an estimate. Reported hours cluster on round numbers — forty, fifty, sixty — a heaping pattern visible in the raw distributions and one reason a reported figure should not be read as a precise count.

Payroll records are administrative and do not depend on memory at all. What they give up is scope: they describe the employment relationship as the employer accounts for it, and the Bureau of Labor Statistics has long distinguished hours paid from hours worked when constructing measures for productivity.

The dispute is not settled

Whether the diary figure or the survey figure is closer to the underlying quantity remains contested. The case for diaries rests on the argument that people summarising a busy week round upward. The case against rests on the observation that work has become more fragmented, and that short intervals scattered through an evening are precisely what a diary reconstruction is least likely to capture. Both arguments can be right at once, in which case the two methods err in opposite directions and neither figure is a ceiling or a floor.

References

Back to the library