Filed under scheduling

Advance notice: how researchers try to measure an unstable schedule

Two research programmes approach short-notice scheduling from different directions, and their disagreements are mostly about method rather than about findings.

Among labour researchers, the length of time between a roster being published and the shift being worked has become a standard way of describing schedule instability. It is easy to state and hard to measure. The two bodies of work that have shaped the field over the past decade took different routes to it, and the gap between them is instructive.

The cohort approach

Susan Lambert, Peter Fugiel and Julia Henly, working at the University of Chicago, analysed the National Longitudinal Survey of Youth 1997 to describe scheduling conditions among early-career workers. Because the NLSY97 follows the same cohort over many years, it allows scheduling to be placed alongside earnings history, education and job tenure. Their analysis reported that a large share of hourly workers in the cohort knew their schedule only a short time in advance, and that variability in weekly hours was common rather than exceptional.

The design carries an obvious boundary. A birth cohort is an age group, so the results describe early-career workers and cannot be read as a statement about the workforce as a whole. Older workers with longer tenure may face different conditions, and the survey cannot say so either way.

The platform-recruited survey

The Shift Project, directed by Daniel Schneider and Kristen Harknett, took a different route. Rather than drawing from an existing probability sample, it recruits hourly workers at large service-sector firms through targeted advertisements on social media platforms, then weights responses to population benchmarks. The advantage is scale and specificity: the project can report on named sectors and describe conditions at a level of detail that general-purpose surveys rarely reach.

The cost is sampling. A sample assembled through advertising is not a probability sample, and weighting reduces but does not remove the risk that respondents differ systematically from non-respondents. Researchers working with the data have been explicit about this, and comparisons against benchmark surveys form part of the project's published methodology. Readers evaluating any single figure from this line of work should know which it is.

What both designs share

Both rely on what workers report. Neither has routine access to the scheduling software that actually produces the rosters, and the divergence between an employer's system record and a worker's recollection is not documented at scale. Question wording also varies: asking how far in advance a schedule is known is not the same as asking how far in advance it is posted, or how often it changes after posting. Surveys that ask different questions produce different numbers, and those numbers are sometimes compared as though they measured the same thing.

A further limit is temporal. Most instruments ask about a reference period of a week or a month. Scheduling practices vary by season, and a survey fielded outside a peak trading period may describe a quieter version of the same job.

Why the disagreement is useful

The two approaches fail in different directions, which is the most useful property a pair of methods can have. The cohort study has defensible sampling and limited coverage. The platform-recruited survey has broad coverage and contested sampling. Where their conclusions converge — that short notice is a routine feature of hourly scheduling in several large service sectors rather than a rare event — the convergence carries more weight than either study alone.

Where they diverge, the divergence is usually traceable to design rather than to the underlying reality. That is worth stating plainly, because summaries of this research in the general press often present a single percentage without the instrument that produced it.