Hourwise

Overtime, unpaid hours and the long-hours tail

Hours beyond a standard schedule are the part of the distribution where measurement is weakest and where the statistical categories inherit their shape from employment law rather than from research.

A legal threshold becomes a statistical one

In the United States, the forty-hour week is a legal construct before it is a statistical one. The Fair Labor Standards Act of 1938 established the threshold above which covered employees must be paid a premium rate, and defined classes of employees exempt from that requirement. Payroll systems record overtime because the law obliges employers to account for it, and statistical programmes that draw on payroll data inherit the distinction.

The consequence is a definitional asymmetry. Overtime worked by a covered employee generates a record. Additional hours worked by an exempt employee generate none, because no payment attaches to them. Both are hours of work; only one leaves an administrative trace.

The unmeasured half

Household surveys are the main instrument for reaching hours that payrolls cannot see, and they reach them by asking. That returns the problem to recall. Reported hours heap on round numbers, and the heaping is most pronounced in exactly the range where the interesting variation sits — the difference between a fifty-hour week and a fifty-five-hour week is unlikely to survive a respondent's summary.

There is also a prior question that no instrument resolves cleanly: what counts. A message answered at ten in the evening, a document read before a meeting, time spent thinking about an unfinished problem. Daniel Hamermesh, whose work on the timing and scheduling of work has drawn heavily on time-use data, has written extensively on how sensitive these measures are to definitional choices made before any data is collected. The choice is not technical. It determines the size of the quantity being measured.

What research on long hours has found

Dora Gicheva studied the relationship between very long working weeks and early-career outcomes in high-end professional labour markets, and reported that the association between hours and wage growth is not linear across the distribution — the relationship at the top behaves differently from the relationship in the middle. Lonnie Golden has worked on the related question of schedule variability, where the issue is not the total but its predictability from week to week.

Running underneath both is the overreporting result described by John P. Robinson and his co-authors, who found that the discrepancy between diary reconstructions and survey reports grows as reported hours rise. If that result holds, the long-hours tail is the least reliable part of the distribution, and it is also the part most often cited.

The objection

The overreporting finding has never gone unchallenged. The counter-argument is that diary methods were designed when work was spatially and temporally bounded, and that they systematically miss work that is dispersed into short intervals across a day. A reconstruction that records an evening as leisure will not capture twenty minutes of work inside it unless secondary activities are collected and treated as work — and whether they should be is itself disputed.

Both critiques can hold simultaneously. Survey recall may overstate long weeks while diary reconstruction understates fragmented work, in which case the true figure sits between two estimates that are each biased in a known direction by an unknown amount. That is an unsatisfying conclusion, and it is the one the literature currently supports.

References

Back to the library