The reduction of working time: definitions and measurement methods

Abstract

Working time reduction (WTR) is a promising policy to enhance well-being in rich countries and an important topic in discourses on a new social vision. Numerous small-scale WTR trials are either underway or planned in various contexts. Properly measuring changes in working time is necessary to evaluate these trials, but challenges abound. Traditional definitions and measurement methods may not work for fragmented, creative, and location-independent jobs. The primary aim of this article is to review relevant work-time definitions and data-collection methods, discuss their complexities, and summarize the implications for WTR research. We reviewed 243 articles and categorize indicators currently used in the literature by relying on 45 methodological studies to identify the main challenges and potential solutions. We conclude that the most dominant definitions and methods, notably usual weekly hours measured by worker surveys, are losing relevance and credibility in many contexts. With the rise of indicators focused on actual hours and measured by time diaries, work grids, interviews, automatic measurements, and time-sampling, we foresee the emergence of non-comparable, job-specific work-time indicators. We propose a new time-sampling approach to deal with some of the least measurable jobs.

Keywords:

• Working hours
• paid work
• unpaid work
• working time indicator
• four-day workweek
• post-growth

Introduction

The reduction of working time (WT) is a widely supported strategy within post-growth economics which also receives substantial attention in the general media and may serve various economic, social, and environmental goals (Kallis et al. 2013; Pullinger 2014; Soper 2020). Arguably, limitations on WT represent the most promising way to keep unemployment low without the need for economic growth (Jackson and Victor 2011; Antal 2014) and this is important because past growth has generally been environmentally unsustainable (Haberl et al. 2020; Vadén et al. 2020) and prospects for green growth in the future are highly uncertain (Antal and van den Bergh 2016; Parrique et al. 2019; Hickel and Kallis 2020). Accordingly, post-growth economic models pay a great deal of attention to working time reduction (WTR) (Spangenberg, Omann, and Hinterberger 2002; Victor 2008; D’Alessandro et al. 2020).

Nevertheless, WTR is a very broad concept including dozens of different types of reductions, with potentially divergent economic, social, and environmental impacts (de Spiegelaere and Piasna 2017). Whether reductions are compensated (by increasing hourly wages) or uncompensated; whether the scheme is individual, collective, or universal in terms of participation; whether it shortens daily, weekly, annual, or lifetime working hours; what the initial levels of working hours are; and several other factors matter. These may significantly influence how patterns of consumption and time use change, with knock-on effects on employment (Marimon and Zilibotti 2000; Bosch and Lehndorff 2001), well-being (Kamerāde et al. 2019; Lepinteur 2019), and sustainability (King and van den Bergh 2017; Antal et al. 2021; Cieplinski, D’Alessandro, and Guarnieri 2021).

Given the variety of WTR programs currently being trialed and planned worldwide, including state-run programs in Iceland, Spain, and Scotland funded by tens of millions of euros as well as large sectoral and corporate schemes in other countries (NEF 2022), a proper understanding of causes and effects is key. This requires a careful description of the systems under investigation as well as their changes. The most obvious question refers to changes in WT. If we are unable to correctly tell how WT has changed, then it is not possible to estimate the impacts of similar changes elsewhere. Therefore, it will also be impossible to consciously apply WTR as a strategy to increase well-being or to reduce environmental impacts.

Yet the measurement of WT is very complex in itself. Specialization increases differences between work activities, resulting in job-specific measurement challenges. Digitalization complicates the distinction between work and non-work by fragmenting both and allowing workers to mix these fragments. Certain activities, such as long coffee breaks supporting creative work, may not even fit into a dichotomous work/non-work classification. Therefore, very different values may be obtained for WT depending on its definition and the measurement methods. Calculating aggregate data for groups of workers adds further layers of complexity. So, while measurement methods for well-being and environmental impacts develop, the reliability of WT measurements may actually decline.

Therefore, anyone who wants to understand the impacts of specific types of WTR must pay attention to the complexities of WT measurement. The present methodological review aims to facilitate this by identifying the main problems and their potential solutions based on the literature. Such an overview is useful for both primary and secondary research. When primary data are collected, it is crucial to be aware of the various ways in which WT can be defined and measured, as well as the potential pitfalls of the indicators. In the context of ongoing WTR trials, most researchers studying well-being and environmental impacts are unlikely to be experts in WT measurement, which makes a summary of this fragmented literature very helpful. For instance, knowing when it is necessary to go beyond standard survey questions will be critical, especially with the rise of company-level case studies. In the case of secondary analyses, common mistakes of the literature include the inadequate reporting and misuse of WT indicators (e.g., incorrect references to “actual WT” or flawed country comparisons). This article shows why carefully describing indicators and choosing data sources are so important and highlights data sources suitable for specific purposes. For both primary and secondary research, awareness of the limitations stemming from WT measurement helps to avoid the overinterpretation of results, which is common in scientific studies and even more prevalent in the grey literature and media reports. Together, these contributions may help to increase consistency and avoid confusion in the emerging research field around WTR.

The structure of the article is as follows. The next section gives a short theoretical overview of WT definitions and measurement methods. The third section outlines the research methods for the survey and the fourth section presents our results and a synthesizing discussion where we present WT definitions and measurement methods used in the literature, review their complexities, and describe implications for different types of WTR studies. Finally, the last section summarizes our conclusions for WTR research.

Theoretical overview

WT definitions

The International Labour Organization (ILO) suggests seven different WT definitions for research purposes to make measurement results as comparable as possible: hours actually worked (actual WT), hours paid for (paid WT), normal hours of work (normal WT), contractual hours of work (contractual WT), hours usually worked (usual WT), overtime hours of work (overtime), and absence from work hours (absence). The ILO further provides very detailed measurement instructions for each. For instance, the definition of actual WT is more than a page long (ILO 2008, 47–48), with references to productive activity, preparations for work, waiting, travel, on-call duty, trainings, in-between time, and resting time, as well as leave, commuting, education not intended for the economic unit, and longer breaks.

The various concepts can be divided into two groups. Definitions in the first group—usual WT, actual WT, overtime, and absence—refer to activities actually taking place, disregarding formal conditions, e.g., whether work is paid or formally registered. Two of these definitions need special attention: usual WT (which in most cases refers to a week and ignores irregular events, such as irregular overtime) and actual WT (which measures actual hours worked over a specific period, usually the week before the measurement, including all unexpected events, such as absenteeism and overtime). The second group—consisting of contractual WT, paid WT, and normal WT—includes more formal administrative definitions. These are related to official salaries and contracts, as well as norms in a region or an occupation, disregarding actual activities (e.g., whether someone actually worked during their paid WT).

It should be noted that no matter how accurate these definitions are in theory, in practice they are inseparable from the instruments used to measure them. In other words, measurement methods translate hypothetical definitions into practical ones. We use the term WT indicator for the combination of a WT definition and a measurement method.

WT measurement methods

There are four main approaches and a number of methods within them to obtain primary data on WT (Figure 1).

Figure 1. How are WT data produced?

The first main approach involves asking workers about their WT. Within this approach, the most widely used method is the worker survey (WS) which uses retrospective questions to inquire about a specific time interval. WSs are widely applicable: most WT definitions can be measured and all types of workers, including self-employed and informal workers, can be involved. Participation is relatively easy for respondents and data collection through this method is generally inexpensive making it possible to ask numerous people simultaneously. WSs can be cross-sectional or longitudinal, so they are also suitable for investigating changes over time for individuals. The main limitations stem from errors in recalling, calculating, and reporting WT. The most reliable national-level data on weekly WT obtained by WSs come from the European Labour Force Survey (EU LFS) in Europe and the Current Population Survey (CPS) in the United States.

Aiming to address the limitations of WSs, another data-collection method relying on workers is the time diary (TD). Participants register what they are doing during the day, usually in 10-minute intervals. This provides data on all types of time use, including formal and informal WT, commuting, and leisure activities. The reliability of this method has been confirmed with wearable cameras and accelerometers that track physical movements (Gershuny et al. 2020). Since TD data covers all types of activities, WT according to any activity-based definition can be constructed. However, TDs are resource intensive. They require a significant commitment from respondents and rejection rates are generally high. Participants rarely register for more than one or two days. Therefore, this method tends to be used at the national level by combining the diary days of different individuals who share certain attributes (e.g., gender, age, family status, place of residence, income) into “synthetic weeks,” exploring broader patterns of time use. A less demanding time-use survey (TUS) to measure WT is a simplified TD called “work grid” (WG) in which respondents record for 15-minute time slots whether they worked or not.¹ The most reliable national-level TUS data come from the Harmonized European Time Use Study (HETUS) in Europe and the American Time Use Survey (ATUS) in the United States.

Less common retrospective and real-time methods are qualitative interviews with employees and time-sampling studies, respectively. The advantage of interviews is flexibility, allowing data collection in almost any context. However, as interviews are rarely used to measure WT, there is little information on their data quality. In the case of time-sampling, respondents register their activities—and potentially a few other things, such as their mood—when a signal arrives. A large quantity of such momentary data is then used to calculate WT.

The second main approach is to ask employers about WT. Establishment-based surveys (ESs) are mostly used to measure WT according to formal, administrative definitions, such as paid work or contracted WT, based on official payrolls, leave papers, and so forth. Before national WSs were launched, WT data came almost exclusively from ESs, so they are indispensable for historical studies. The most reliable national-level data on formal WT come from the European Company Surveys (ECS) in Europe and the Current Employment Statistics (CES) in the United States.

The third main approach is observation by researchers which is the costliest procedure and tends to be used only in very specific cases. On-site observation may be necessary to estimate the productive WT of employees within an institution or to establish industrial standards.

The fourth main approach is to rely on data traces that workers leave behind and calculate WT from these. The simplest option is to gather administrative data, such as contracts, tax papers, and so forth. A more in-depth method is to mine digital data. Due to the diversity of contexts, measurement methods must be developed case by case.

In addition, primary data collected through the previously outlined methods can be used to calculate secondary WT values, such as averages for different groups. Statistical offices produce various aggregates, such as summations that are used in national accounts.

Methods

We had three main research questions, the first two of which required data collection:

• Q1. What WT indicators are used in the literature?

• Q2. What complexities stem from their definitions and methods and how is it possible to address them?

• Q3. What are the implications for WTR research?

Answers to Q1 identify the most widely used definitions and methods. This is useful as these measures are likely to remain important—not only because of the persistence of certain advantages and disadvantages of particular indicators but also because of the drive toward standardization. Exploring Q2 then helps to use the definitions and measurement methods with awareness of their limitations. Finally, answers to Q3 summarize practical lessons for research on WTR, paying special attention to measurement strategies that address contemporary challenges and go beyond current standards.

To answer Q1, we conducted database searches with various combinations of keywords in Scopus. The strings (see Appendix A) were chosen according to the three main goals of WTR. We searched for phrases related to WT within the “Economics” subject area, finding articles on the relationship of WT with economic indicators, such as employment and productivity (WT-ECON). We combined the same WT phrases with expressions on well-being, finding articles on work-life balance and physical and mental health (WT-WB). For environmental articles, we relied on a recent systematic review (Antal et al. 2021) that combined the same WT-related search terms with expressions pertaining to the environment (WT-ENV). As our aim was to obtain a rough overview of the WT-ECON and WT-WB fields to complement the comprehensive but narrower picture of the WT-ENV area (consisting of 15 studies), we ranked the findings in terms of “relevance” in Scopus and considered the first 200 results. In addition, we used a search string focusing specifically on WTR, which generated 2,423 results.

We screened these 200 + 200 + 15 + 2,423 articles for relevance, starting with the titles and abstracts, then proceeding with the full texts. We included English language articles that looked at the relationship between WT and an economic/well-being/environmental indicator, relying on empirical WT data. This left 95 WT-ECON, 88 WT-WB, 15 WT-ENV, and 45 additional WTR studies.

For a total of 243 articles, we recorded WT definitions, data-collection methods, and the respective economic/well-being/environmental variables. As the articles often referred to external sources—for example, surveys and databases from which they took WT questions or values—we followed these routes to reach the exact definitions or sources. The result is a rough overview of the main definitions and measurement methods used in the literature. Although this is not a comprehensive or fully systematic review, the sample is appropriate to identify the main types of indicators used in these fields.

To answer Q2, we used a search query on WT measurement in the Scopus database, restricting the subject area to social sciences. This generated 1,088 articles. We screened the titles and abstracts, then the full texts, and identified 18 highly relevant articles focusing on WT measurement methods. Using these publications as a core set, we conducted multiple rounds of citation snowballing to find further relevant papers. Some of the resulting articles were highly relevant, while numerous others were informative but less central to our analysis. In the end, we worked with 45 methodological articles.

To answer Q3, we interpreted insights from the reviewed literature in light of the aims and challenges of WTR research. These discussions are relevant for the planning and evaluation of WTR trials.

Each methodological decision brought important limitations. First, the sample used to answer Q1 is not entirely systematic and not representative of any particular field. Therefore, findings are meant to be interpreted with caution. Only the broad patterns are likely to be meaningful, not the concrete numerical values that characterize our somewhat arbitrary sample. Second, the list of complexities may be incomplete. It is very difficult to properly define the area within which the methodological issues and their possible solutions are sought. The citation-snowballing method was used until we felt that the marginal value of further literature review was minimal, but multiple rounds of complete backward and forward analysis of citations were beyond the scope of this investigation. Third, our interpretations in Q3 could not cover all possible combinations of WT indicators, WTR types, and work contexts. As there are dozens of WT indicators and types of WTR and hundreds of different work contexts, the number of possible combinations is extremely high. Therefore, we tried to give relevant illustrative examples to inspire researchers who study specific cases.

Results and discussion

A diversity of indicators

Q1: A first observation is that a large number of WT definitions and measurement methods are used in the literature (Table 1). We identified at least 28 different types of indicators, not counting differences between question wordings and differences between reference periods.² These indicators were produced using more than 10 different methods.

Table 1. WT indicators are found in the literature.

WT measurement methods
	Worker surveys	Establishment surveys	Time diaries	National statistics/national accounts	National guidelines/legal definitions	Other
WT definitions	Usual WT	Paid WT	Paid/Unpaid WT	Standard average WT per employee	Normal WT	Qualitative WT indicators
	Actual WT	Productive WT	Market WT	Average hours worked per worker	Standard WT	WT based on data mining
	Paid/Unpaid WT	Contractual WT	Household WT	Average hours worked per person	Collectively agreed WT	WT based on observation
	Overtime (paid/unpaid) WT	Normal WT	Actual WT	Total average hours per employee actually worked and paid for		WT based on work grid/calendar
	Start and end of WT	Usual WT	Total WT			WT based on different calculations
	Household WT

Q2: Indicator choice is crucial for at least two reasons. First, not all indicators are equally applicable for all purposes. This is sometimes obvious—for instance, national accounts provide little insight into individual well-being. Other times the difference can only be suspected as is the case between formal and actual WT. Regarding this disparity, Worrall and Cooper (2014) report an average of 1.5 overtime hours a day for managers, which means 46 full days a year (with 8 hours per day). More recent reports from the grey literature offer even higher estimates (ADP Research Institute 2021). Similarly, actual WT rarely coincides with hours set by collective agreements or state legislation (Burgoon and Baxandall 2004).

Figure 2. (A) Shares of WT definitions. (B) Shares of WT indicators by the reference period. Note. In 5–10% of the articles, it was not possible to clearly determine what definition was used.

Second, variations between seemingly similar indicators may be substantial. For example, actual WT according to WSs and ESs may be very different, especially for multiple job holders (Bluestone and Rose 1998). On one hand, as ESs are specific to establishments, a worker who has two part-time jobs of 20 hours per week each is registered at 20 hours, perhaps even twice. On the other hand, the WS registers 40 hours. Moreover, actual WT in ESs refers only to officially registered WT, so unpaid overtime, informal WT, and self-employed work are usually not measured. Therefore, actual WT tends to be significantly lower in ESs than WSs (Fleck 2009).³ Studying the same workers with the two methods, differences were above 10% in jobs where WT is less measurable (Abraham et al. 2013).⁴

Q3: Choosing the right indicator is often not straightforward, but no general rules can be given because there are dozens of ways in which WT can change, hundreds of different work contexts, and various potential research goals. In fact, the diversity of indicators is a response to this complexity. Since formal WT data rarely give a good approximation of the duration of work, activity-based indicators are almost always necessary to track progress toward WTR and to study its impacts. However, the direct influence of WTR policies will be on formal WT, so the analysis usually requires information on both.

Dominant definitions

Q1: The most dominant WT definitions are activity based: usual WT is followed by actual WT. In terms of the reference period of retrospective measures, weekly working time is by far the most common (Figure 2).

Figure 3. Shares of WT measurement methods. Note. The numbers in brackets next to the surveys on the right side of the figure are not percentages, but the absolute number of articles in our sample in which they were used.

Q2: The most commonly used WT definitions differ from each other. Table 2 shows weekly usual and actual WT, collected from the same respondents in an LFS.

Table 2. Comparison of hours usually worked and hours actually worked, full-time employees, 2014 (Source: Eurostat 2018).

	WT indicators		Difference
	Hours usually worked	Hours actually worked	In hours	In %
EU-28	40.4	35.6	4.8	11.8
Belgium	39.2	34.3	4.9	12.6
Bulgaria	40.9	39	1.9	4.7
Czech Republic	40.8	36.2	4.6	11.3
Denmark	37.8	33.6	4.2	11
Germany	40.5	35.6	4.9	12.2
Estonia	40.6	36.8	3.8	9.3
Ireland	39	35.3	3.7	9.6
Greece	40.8	38.4	2.4	5.8
Spain	40.2	35.9	4.3	10.7
France	39	31.9	7.1	18.2
Croatia	40.9	37.1	3.8	9.4
Italy	38.7	34.4	4.3	11.1
Cyprus	41.5	37.2	4.3	10.4
Latvia	40.4	37.6	2.8	7
Lithuania	39.4	36.3	3.1	7.9
Luxembourg	40.1	36.4	3.7	9.2
Hungary	40.7	37.9	2.8	6.9
Malta	40.4	37.2	3.2	7.8
Netherlands	38.9	35.9	3	7.6
Austria	41.5	35.1	6.4	15.5
Poland	41.1	37.5	3.6	8.7
Portugal	41.7	36.5	5.2	12.4
Romania	40.8	40.1	0.7	1.8
Slovenia	41	35.1	5.9	14.5
Slovakia	41	36.7	4.3	10.6
Finland	39	32	7	10.8
Sweden	39.9	32.4	7.5	18.8
United Kingdom	42.4	37.1	5.3	12.5
Iceland	44.5	38.8	5.7	12.7
Norway	38.5	32.1	6.4	16.7

Note. The EU average and countries with the largest differences are highlighted.

On average, usual WT was 11.8% higher (range: 1.2–18.8%) than actual WT in the European Union (EU) in 2014. Broader findings are similar: in all methodological papers, usual WT was higher than actual WT. This is mainly because of the difference between the definitions. Asked how many hours one usually works, people likely report the most common value, not the average. If the distribution is skewed, which is likely because downward divergence from the most common value is more characteristic than upward divergence, then the usual WT will be above average WT. This is particularly true for workers with long work weeks, as their WT distribution is likely to be more negatively skewed (Lin 2012). Similarly, if the social norm is that “more work is better,” then usual WT may be overreported, especially by workers with varied work schedules who do not have standard weeks and by workers who find such calculations difficult.

Another challenge is to use an appropriate reference period. If the period is short (e.g., a day), then the recall bias will generally be smaller, but the indicator will be less informative because longer periods show variation (workdays differ). If the period is long (e.g., a year), then retrospective measures will have a large recall bias. Therefore, longer-term WT data is more reliable if based on repeated data collection using shorter reference periods. This procedure is costly, explaining why weekly values are dominant and why longer-term data, such as annual values are generally produced by large organizations like national statistical offices. A noteworthy consequence is that we know much less about daily, annual, or lifetime WT than about weekly WT.

Q3: Despite its systematic distortions, usual WT will likely remain dominant for some time. It is meaningful for smaller groups of workers or a smaller number of data points than actual WT, which is crucial for smaller-scale case studies. It can provide information on past WT because it does not refer to a definite time period, which facilitates “before/after” WTR comparisons. And it is methodologically much less challenging to measure usual hours than actual hours for a long enough period to assess the impacts of WTR policies.

Nevertheless, current changes in the world of work, notably the unpredictability and fragmentation of WT in many jobs, make usual WT increasingly inadequate. To some extent, such problems can be mitigated by directing attention to non-work periods and overtime. In particular, asking about absences and overtime before total WT in WSs improves data quality for both usual and actual WT because respondents reflect more on irregularities (Eurostat 2018). In addition, a shift toward actual hours is a solution, even if this complicates measurements and works better for larger samples.

The characteristics of WTR schemes matter for the limitations of WT data using different reference periods. It is impossible to understand the causes or the long-term impacts of life strategies with a very long time horizon, for example, the reduction of WT after years of long hours with commonly used reference periods. Similarly, the effects of organization-level WTR schemes that increase the length of holidays are often missed by activity-based definitions focusing on weekly hours, which underlines the complementary role of formal WT data or questions focusing on holidays. Variable work schedules also call for the use of annual WT. When weekly hours are adequate, then keeping their limitations in mind is key. For instance, reductions from long weekly hours may significantly reduce the bias of usual WT, so reductions may appear larger than they are.

Dominant methods

Q1: By far the most dominant WT measurement method is the WS. The most commonly used data sources are national household-based surveys (e.g., LFS, CPS). WSs are followed by national statistics, establishment surveys, time diaries, and other methods (Figure 3).

Figure 4. Causes of large WS errors.

Q2: Worker surveys: Answers to questions like “How many hours per week do you usually work?” or “How many hours did you work last week?” may not always reflect actual behavior. In the United States, respondents spend about 3–10 seconds to answer such questions (Robinson, Chenu, and Alvarez 2002). For many people, this is probably not enough to accurately interpret the question and provide a reliable answer (Walthery and Gershuny 2019). After all, interpretations of “work” may not be consistent, people may not remember all work episodes, and may not be able to properly add them up, and calculating averages over a time period is even more difficult. The measurement of WT in surveys may be affected by random errors arising from unclear questions or the loss of respondents’ motivation, and systematic errors stemming from incorrect or biased evaluations of the length of work and non-work activities (Niemi 1983, 1993; Robinson and Bostrom 1994; Carrasco and Domínguez 2015). Figure 4 shows when errors tend to be larger.⁵

Figure 5. Differences between compiled WT indicators using diverse denominators: the example of annual working hours per worker in the United Kingdom. Source: Giattino, Ortiz-Ospina, and Roser (2020) and adapted from Our World in Data (https://ourworldindata.org/working-hours). Note. OECD data (OECD 2020) refers to average annual hours actually worked per worker; Penn World Table (PWT) data (Feenstra, Inklaar, and Timmer 2015) refers to average annual hours worked by persons engaged; Bick et al. data (Bick, Brüggemann, and Fuchs‐Schündeln 2019b) refers to average annual hours worked per employed person; Huberman and Minns data (Huberman and Minns 2007) refers to annual hours of full-time production workers (male and female) in nonagricultural activities.

A consistent finding, and likely the largest source of bias, is that people reporting WT well above the average tend to overreport (especially females) while people reporting WT well below the average tend to underreport (especially males) (Frazis and Stewart 2004; Bonke 2005; Walthery and Gershuny 2019). The larger the difference from the average, the larger the bias tends to be. Regarding other factors, the biggest difference between WS and TD data is usually for atypical workers including the self-employed, agricultural workers, people working from home, people who have many unsocial working hours, and time-squeezed women (Eldrige and Pabilonia 2007). Respondents who feel tensions between work and home life and people who more frequently do several things simultaneously tend to overestimate paid hours. Absences and breaks are not always appropriately deducted, so individual WT is more often higher than lower in WSs compared to TDs (e.g., Frazis and Stewart 2014; Kan and Pudney 2008). Social norms may also partly explain why women tend to overstate market work while men overstate non-market work.

The calculation of average actual WT from WSs brings new sources of errors. For instance, the most widely used labor-force surveys significantly underestimate the number of days off and sick leave in most countries, which leads to a general overestimation of actual WT by WSs. This is because respondents who did little work the previous week often deny responses or talk about normal weeks instead, workers with zero hours are often excluded from the sample (Wingender 2018), and in many countries, there is seasonality in data collection (e.g., the Christmas week is not sampled (Bick, Brüggemann, and Fuchs‐Schündeln 2019b) or the sample sizes for certain weeks are small (Frazis and Stewart 2014).⁶ In the EU, using the average of LFS microdata over all 52 weeks has been shown to overestimate the number of workweeks by 7.5% (by missing 3.4 vacation weeks), with significant differences between countries (Bick, Brüggemann, and Fuchs‐Schündeln 2019b).

The magnitude of the difference between WS data and TDs (or WGs or calculated WT) was studied in a number of the articles that we reviewed, usually yielding differences of up to 10–30%, with substantial variations between groups of workers (Appendix B).

Establishment surveys

Data ased on ESs are usually limited to formal definitions of WT, notably paid or contractual hours. ESs also lack the data necessary for micro-analysis, such as the socio-economic status of the workers, and getting such data is often difficult or impossible for reasons of confidentiality (Bonke 2005). Since ESs give preference to larger firms and formal employment, employees of smaller firms, the self-employed, and informal workers are typically excluded from the sample (Messenger, Lee, and McCann 2007; Bick, Fuchs-Schündeln, and Lagakos 2018). As a result of these limitations, ESs are usually applied only as complementary sources of information in the social sciences (Abraham et al. 2013; Bick, Fuchs-Schündeln, and Lagakos 2018; Ohanian and Raffo 2012).

National statistics and national accounts

National accounts (NA) are a large subset of national statistics that use a variety of sources to consistently summarize the economic activities of a country. These accounts include data on WT, which are calculated in country-specific ways. Relying on national accounts and making further adjustments, international databases contain time series for WT covering many countries. The most frequently applied sources are the database assembled by the Organization for Economic Co-operation and Development (OECD) and the Total Economy Database (TED), the latter of which also serves as a basis for the Penn World Tables (PWT). For the EU, even more detailed data (e.g., sectoral WT) are available from the EU Klems database.

Such “compiled” indicators have their own issues. To begin with, for the calculation of an average per capita WT at the national level, a reference group must be chosen, which will serve as the denominator variable. Average WT may then refer to the whole population, a reasonable age group, the working population, or a subset of the working population (e.g., the formally employed or nonagricultural workers).

Statistical offices must also decide what data they take as a starting point: some use WSs while others use ESs. Subsequent methodological decisions are about adjustments, such as correcting inaccuracies stemming from absences and overtime hours (Wanger, Weigand, and Zapf 2016). While certain data are available almost in real-time (e.g., sick leave), others have a considerable time lag and sampling errors, so they often appear as estimates (e.g., data on secondary jobs from household surveys).

Despite various attempts at standardization, the latest empirical research suggests that NA data are not suitable for international calculations in which the exact level of WT is required (e.g., for productivity comparisons). Thus, the mentioned databases only allow the identification of WT trends within a specific country (Fuchs-Schündeln 2019; Ohanian and Raffo 2012; Wingender 2018).⁷ For international comparisons, it is more promising to use adjusted WS data (e.g., starting from standardized LFSs) than national statistics (Bick, Fuchs-Schündeln, and Lagakos 2018).

Wingender (2018) found that NA data based on ESs are generally lower than data based on WSs (actual WT from LFSs).⁸ As the example in Figure 5 shows, values in different data sources can easily differ by 10%. Furthermore, OECD and TED data have been substantially revised over time: WT values for the same countries in the same years could change by more than 10% (Bick, Brüggemann, and Fuchs‐Schündeln 2019a).

Figure 6. The share of articles considering non-market WT.

Time diaries

TD is considered to be almost statistically unbiased for measuring activity-based WT. Despite the fact that recall bias and the impact of social desirability are not ruled out entirely, they are assumed to be very modest—in fact, they are not even quantified (Kan and Pudney 2008). TD data is more reliable for regularly occurring activities than for irregular ones (e.g., market work as opposed to irregular overtime) because the reference period of TUSs is typically one or two days (Juster, Ono, and Stafford 2003). The accuracy of aggregated TD data can be improved in two ways: by increasing the sample size and by increasing the length of data collection for each participant. The former is appropriate if aggregate WT levels and distributions between socio-economic groups are investigated, while the latter helps if longer-term behaviors of the same persons or households are studied.

TUS-data collection and analysis require large investments and this feature has strongly limited the availability of TD data until recently.⁹ National statistical offices only carry out large-scale TUSs every few years and the TD method is also very burdensome for participants (Chatzitheochari et al. 2018), especially in the case of multi-day diaries. For weekly diaries, response rates of around 30% are considered very high (Robinson et al. 2011). Nevertheless, several experts stress the need for more weekly diaries to make TD data more comparable with data from other measurement methods using longer reference periods (Kan and Pudney 2008; Robinson et al. 2011; Sullivan and Gershuny 2018). This procedure is easier if market WT is in the focus and the WG method is used (Robinson, Chenu, and Alvarez 2002), which, however, is rare in the literature.

Interviews

Interviews allow workers to clarify interpretations, so more consistent data can be obtained than with any other retrospective method. Any WT definition and reference period can be studied. The main drawback is that conducting and analyzing interviews for large samples is very expensive, and findings cannot be generalized.

Time-sampling

Time-sampling methods gather data in real-time, but not continuously. Participants report their activities at specific points in time during the day, usually triggered by a beeper or a smartphone application. As postponing responses is not allowed, the recall bias is negligible but many answers—around 25% according to some analysts (Jeong 2005)—tend to be missing. Because the length of activities is not measured, a large number of data points is required to accurately measure WT in smaller groups. While the method is hardly applicable at the individual level, it is an option for groups above 30–50 workers, depending on the frequency of measurements per participant (Sonnenberg et al. 2012; van Hassel et al. 2017).

Personal observation

Direct personal observation eliminates the subjectivity of respondents from the data. However, it is an intrusive procedure, which may alter the behavior of the subjects—this is known as the Hawthorne effect (Guarisco, Oddone, and Simel 1994). It is also very expensive to train observers, and it is not guaranteed that even a well-trained observer will be able to accurately identify all work activities (Oddone et al. 1995). Finally, it is difficult to apply this method for people with irregular or unsocial working hours, and for mobile or remote work.

Automatic measurements

Automatic measurements—for example, by smart cards and digital timers—are usually designed to measure productive WT excluding any slack time and breaks (Collewet and Sauermann 2017). Most types of automatic measurements are limited to highly standardized workplaces like industrial facilities, call centers, or supermarkets (Hodder 2020), and to digital work like software development. These measurements are not perfect either: time spent on non-measurable but possibly work-related activities (e.g., thinking) are often missed, while some non-work activities may appear as work.

Portable and wearable automatic devices, such as mobile phones providing global positioning system (GPS) data, accelerometers, and wearable cameras—can be more precise and more applicable for various types of work, but some technological barriers persist (Elevelt et al. 2021) and using them for larger samples is not always feasible (Gershuny et al. 2020). Furthermore, while these tools may seem to provide objective data, the Hawthorne effect is likely to be very strong because participants are aware of being constantly monitored.

Q3: Depending on the aims of a WTR study, the limitations of different methods will matter. We highlight some of these complications and their potential solutions, which we expect to be particularly important in future research.

White-collar workers with long WT could be a prime target of WTR efforts: they can afford uncompensated reductions and for them the marginal benefit of additional free time is high. However, their WS data are often highly distorted. To mitigate these distortions, repeated surveys of actual hours can be used instead of the usual WT. In addition, separate questions can be asked about WT variability, then people with highly variable or very long WT can be treated separately. If the number of such participants is small, interviews work well. If the number is larger, then WGs may be appropriate. If the fragmentation of WT makes even the 15-minute periods of the WG too long for accurate reporting (Jett and George 2003; Wajcman and Rose 2011; Chen and Karahanna 2018; Puranik, Koopman, and Vough 2020) or the WG is too burdensome, then time-sampling is an option. Regardless of the method, pre-measurement interviews are useful to explore context-specific problems of WT measurement.

As WT values according to different measurement methods can differ by 10–30%, and most reduction schemes affect WT by a similar amount, using two different methods before and after WTR is a mistake. In the case of smaller changes in WT, smaller samples, and more uncertain methods, cautious interpretation is advised. Whether pre-WTR and post-WTR measurement errors are likely to differ should be qualitatively assessed by going through the main biases. In particular, the distribution of working hours and the level of time squeeze may change perceptions of working time (Carrasco and Domínguez 2015), so changes in these deserve attention.

For methods with a strong Hawthorne effect, it is important to emphasize that various work vs. leisure preferences can be sensible. In addition, exploring individual norms may help to assess whether data are more likely to be distorted upwards or downwards.

For studies relying on aggregate data, such as many environmental papers that investigate the time series of WT and annual resource use or emissions, it is better to use internationally consistent data (e.g., Bick, Brüggemann, and Fuchs‐Schündeln 2019b). When cross-country comparisons are only used to contextualize or interpret research findings, then the risk of misusing aggregate data sources is even higher because authors rarely dig into the details for a few introductory or interpretative comments. Therefore, the OECD should replace its cross-country bar chart with one showing trends over time on its main page (OECD 2022).

Precise indicators

Q1: Most articles use data from WSs that simplify complex WT definitions into one or two short sentences. More detailed instructions may be given to interviewers: for instance, the EU LFS question on actual hours is one line long while the explanatory notes fill a page (Eurostat 2016), but it is unlikely that many respondents inquire about these instructions.

A substantial number of papers do not report WT indicators properly and definitions are confused or missing. The WT-WB literature is the least precise, followed by WT-ENV, while the WT-ECON literature is the most precise, usually giving indicators in full.

Q2: Classifying activities as work or non-work and defining the breaks to be excluded for all types of jobs would require very detailed descriptions in self-reported measurements. This may not be feasible for larger, more diverse samples.

A traditional issue is how lunch breaks are treated. Even 30-minute lunch breaks make up more than 6% of average daily hours in most countries. This adds up to 120 hours a year, or 15 full days (with 8 hours per day), so differences of this magnitude deserve attention. A related question is about longer coffee breaks, which may or may not serve work purposes (Dupre and Lallement 2007). Today, questions about the boundaries of WT are especially relevant as creative tasks gain importance, people increasingly work from home, and breaks become increasingly irregular (Messenger and Gschwind 2016; Feng and Savani 2020; Hodder 2020). This leads to the question of whether traditional WT or productive WT is more relevant (Collewet and Sauermann 2017). The answer depends on the focus of the analysis (e.g., well-being, environmental, or economic impacts).

Q3: There are at least three lessons for WTR. First, the precision of WT definitions and the feasibility of primary data collection must be balanced against each other. Most importantly, survey questions can be too short and not precise enough, or too long and exhausting for respondents. The optimum depends on the research context.

Second, in the case of smaller (or slower) changes in WT, as well as for statistical analyses, precision is particularly important. Self-reported survey data or aggregate WT values may be too crude for such purposes. If other types of data are not available, then a very serious limitation must be acknowledged.

Finally, the length of breaks and overtime may change as a result of WTR. This means that activity-based WT definitions must be precise enough to measure these periods. One type of WTR receiving a great deal of media attention is when workers complete the same tasks in less time (e.g., four days) and continue to get the same salary.¹⁰ While traditional WT decreases substantially, productive WT may be unchanged. If neither production nor total consumption changes significantly, then economic, environmental, and well-being effects may be smaller than with reductions of productive WT. Exploring how productive WT or the amount of output change is thus useful beyond measuring traditional WT. Obviously, this is more important if productive WT or output are weakly linked to traditional WT.

Non-market work

Q1: Non-market or household (including care) work has a huge impact on the well-being of individuals (Steptoe et al. 1998; Rotenberg et al. 2008; Boye 2011; Kamerāde et al. 2019) and potentially also on environmental impacts and the living standards of households (Boeri, Burda, and Kramarz 2008). If services provided by non-market work in one household are paid for in another, then the same household income results in different discretionary spending (Folbre 2020). Furthermore, despite recent convergence, household WT is still unequally shared by men and women, so it is crucial from a gender perspective (Gershuny et al. 2020). In other words, ignoring household work, as most statistics do, is misleading from the perspectives of both welfare and well-being. Yet non-market WT receives little attention in the reviewed literature (Figure 6). Studies that do consider household and other forms of non-monetized work use TDs more often than for market work.

Q2: As there are no administrative records and neither automatic data collection nor direct observation are practically feasible, the length and distribution of non-market WT can only be asked directly from the workers (or other household members). However, there are no international standards to define household WT. As a result, various interpretations co-exist, for example, childcare may be included (Sayer and Gornick 2012) or excluded (Berik and Kongar 2013). This disparity greatly reduces the reliability of survey data and the validity of comparisons. Moreover, household WT is usually constituted by several short episodes, often carried out simultaneously with other activities, thus retrospective data on household WT is highly unreliable and controversial. One study found household WT to be 36% lower in a WS than in a TD (Bonke 2005), while many others found the opposite result (e.g., WS values have been reported in some cases to exceed TD values by 100% (Robinson et al. 2011), by 70% for men and 30% for women (Juster, Ono, and Stafford 2003), and by 56% for men and 50% for women (Marini and Shelton 1993) in different contexts). Note that even TD data are sometimes controversial—for instance, coding childcare at night is a challenge (Kitterød and Pettersen 2006).

Q3: When market work is reduced, non-market work often changes substantially. Whether this happens is a key question for WTR studies: it matters for well-being and determines changes in time use, with a strong impact on the composition of consumption. Some authors envisage downshifters engaging in eco-friendly non-market work (e.g., Nørgård 2013; Schor 2010), while others contend that polluting activities like travel will dominate (Lindsay, Lane, and Humphery 2020). Therefore, information on non-market WT is necessary to assess the individual and social desirability of WTR.

As WSs are so unreliable, researchers have proposed TDs and interviews as approaches for measuring non-market WT (Niemi 1993; Juster, Ono, and Stafford 2003; Robinson et al. 2011). If this is not possible, then question wording is key for clarity and congruent answers (Schulz and Grunow 2012). In particular, asking about separate activities like cooking, shopping, and cleaning is preferable to aggregated questions on total household WT over a longer reference period. If gender issues are investigated, then asking about the percentage shares of various activities is also a good idea.

Levels of analysis and a web of working times

Q1: Most articles (∼80% in our sample) use WT data of one or more countries from large representative surveys, focusing exclusively on the national level. These surveys usually register the WTs of all members of a household, but they are generally not longitudinal and do not measure non-market WT. Substantially fewer papers deal with lower levels, such as sectors, companies, or households. Among lower-level investigations, studies on the medical sector are dominant (>50%). The overwhelming majority of articles focus on the WTs of particular individuals or groups, treating them as separate, non-interacting entities.

Q2: If WT patterns or dynamics are to be broken down by groups, then national level studies are constrained by the availability of data on respondents. Information on job and family contexts will often be too crude to compare groups of interest (e.g., managers and professionals) or to control for relevant variables (e.g., personal values). Furthermore, statistical models can easily become too complex to have real explanatory value, mainly because so many variables influence WT.

Furthermore, the WTs of individuals, groups, and even nations strongly influence each other, like interconnected nodes in a network (Gerstel and Clawson 2018). Within households, the WTs of partners interact, for example, through the sharing of household work. This has been studied in a few contexts (Kitterød and Pettersen 2006; Presser 2006; Xu et al. 2019), but the results are too general to be informative for specific changes in WT. How WTR will affect partners’ (or other household members’) WT is thus usually unpredictable.

Other interactions, for instance between colleagues due to substitution or complementarity and between companies due to competition or supply-chain effects, are almost entirely unknown. Such interactions depend on the state of the economy (e.g., whether newly hired workers were previously employed). Changes in opening hours can also influence time use in broader communities (Anttila, Nätti, and Väisänen 2005).

Finally, even countries may have WT interactions. Reductions in one country that affect production, exports, or imports may affect WT abroad. These interactions depend on international power relations, which partly explains why people in poor countries do so much more paid work (Bick, Fuchs-Schündeln, and Lagakos 2018). Paying attention to WT leakage is thus important from a justice perspective.

Q3: Due to numerous smaller, non-collective WTR programs running today, individual, household, and firm-level studies are increasingly relevant. This requires primary data collection by researchers in various contexts.

Direct effects of WTRs are easier to describe than indirect ones because “[c]hanges in any given individual’s time create a cascade across other people, organizations, and institutions” (Gerstel and Clawson 2018, 80). There is no methodological tradition to reliably quantify such effects. This limits the types of conclusions that can be reached regarding the overall effects of WTR.

To address this limitation, relevant interactions should be identified. This often requires 1) focusing on the combined work schedules of family members and 2) looking at total WT at organizations by investigating the number of people employed as well as the number of hours worked by employees, and 3) considering the state of the macroeconomy.

In most cases, only theoretical speculation and interviews are feasible to explore situations. A more detailed checklist of issues and corresponding methods of analysis, as well as reasons for which expectations about understanding the web should be tempered, are given in Appendix C.

Development of WT measurement over time and steps forward

Q1: Unlike many fields of social science that have been revolutionized by digitalization and large datasets, changes in WT measurement and analysis have been incremental. WSs, first used in 1909, are still dominant (Pencavel 1986). So, despite developments in the measurement of well-being (e.g., smart tools supplying physiological data), environmental impacts (e.g., advances in input-output modeling), and economic effects (e.g., the radically increased availability of transaction data), the WT-WB, WT-ENV, and WT-ECON relationships are still difficult to study. While WT definitions and sampling strategies were improved and standardized and electronic data collection spread, the world of work has changed so much that the overall reliability of WT data has likely decreased (Robinson and Bostrom 1994; Robinson et al. 2011; Walthery and Gershuny 2019).

Q2: WT measurement developed insufficiently mainly because of the exploding diversity of work activities and job types compared to the industrial age. While several well-being, environmental, and economic indicators can be measured in a uniform way (e.g., heart rates, emissions, or credit-card transactions), this is impossible with WT. In fact, aggregating all types of activities in a single WT makes little sense for certain types of jobs. For physicians and other medical personnel, sleeping during on-call time is clearly different from treating patients, with large impacts on well-being and non-work time use, so the shares of these activities are crucial. Neither of these activities are comparable to the fragmented work of a researcher who may have new ideas while engaging in sports. So, WT data are not necessarily informative without the measurement of intensity and productivity. This is more challenging for individualized, flexible, and creative work than for industrial jobs where intensity is externally determined. As noncreative work is increasingly mechanized, ongoing specialization brings new work activities, and working from home spreads, the trend is toward poor measurability.

Q3: From a WTR perspective, the situation is challenging. Subnational case studies, including most current trials and schemes worthy of investigation (NEF 2022), require case-specific measurements, preferably with high precision. Yet, changes in WT will appear to be different depending on the indicator. As no definition or method is universally applicable, context-specific indicators have to be developed. This will make pre-WTR/post-WTR comparisons more meaningful, while also making comparisons between jobs, and aggregate data, less relevant.

As for measurement tools, interviews, WGs, relying on digital traces, and time-sampling can be expected to gain importance as opposed to WSs. Recognizing the complexities of measurements, researchers should develop interview guides. For WGs, smart and user-friendly registration tools hold the potential to reduce respondent burden and rejection rates (Chatzitheochari et al. 2018). For digital traces, the key is careful interpretation, which will be context-specific and not always straightforward, causing divergence between indicators for different jobs. For time-sampling, one could imagine participants registering percentage values expressing the extent to which their current activities can be classified as work: often 0 or 100%, but sometimes values in between, potentially with an option to switch on a default answer (e.g., 0% while sleeping). For larger samples or longer sampling periods, this could be one of the most widely applicable methods, and a rare option to measure very fragmented workdays.

Table 3 gives a summary of the main practical lessons that have been covered above. Awareness of these complexities and the potential ways in which they can be addressed helps to move forward in the field of WT measurement. This is essential to learn from WTR trials and to facilitate their successful implementation.

Table 3. Main dilemmas of WT measurement and approaches to address them.

Dilemma		Approaches to move forward
Activity-based definitions	Usual WT more widely applicable but more distorted than actual WT (in most cases: usual WT > actual WT), especially for long & flexible WT, powerful work norms	-Usual: reference period not necessarily very recent (can be used for before/after comparisons if measurement starts after WTR); smaller samples and individual WT; more regular and less fragmented WT -Actual: reference period must be recent (e.g., yesterday, last week); larger samples and average WT of the sample; more variable and fragmented WT -Statistical methods to correct biases (e.g., Lin 2012; Eurostat 2018)
	Comparability vs. Context-specificity	-Context-specific definitions often better than traditional, standardized ones, especially if WT is difficult to define using a single indicator (e.g., medical workers’ WT that includes different types of on-call time).
	Precision vs. Feasibility (more precise definitions require longer descriptions, which increases respondent burden, potentially lowering response rates or increasing costs)	-Precision more important: statistical studies, small changes in WT → define length of breaks that must be deducted from WT, measure frequency of breaks (may change as a result of WTR) -Precision less important: if trends are studied and activities/intensity of work do not change significantly as a result of WTR (e.g., intensity is externally fixed); if work schedules are not variable
Reference period	High information content (long period) vs. Small recall bias (short period)	-Long reference period more important: variable weekly WT -Alternative: repeated shorter term data collection over longer periods -Separate questions on holidays & longer absences (number of annual working weeks)
	Seasonability of WT; Irregular periods (e.g. Covid-19) Life strategies impacting WT over multiple years (e.g. Career / accumulation goals)	-Separate questions on seasonability, irregularities, impacts of life strategies
WS	High WT overreported, low WT underreported	-Awareness that WT reductions are likely to be overestimated if WT before WTR is substantially higher than after WTR -Statistical methods to correct biases (e.g., Kan and Pudney 2008)
	Large reporting error by atypical workers: self-employed, agricultural worker, work from home, work during unsocial hours, time-squeezed women	-Separate questions on WT variability, treating respondents with very irregular WT separately (e.g. Interviews) -Separate questions on the strength of work-life boundaries -Questions about absences and overtime before asking WT -Using other methods: interviews, WG, TS – potentially together with pre-measurement interviews to explore the context -Statistical methods to correct biases (e.g., Gershuny et al. 2007; Walthery and Gershuny 2019)
	Sick days, holidays underestimated in annual data (e.g., in aggregate WT averages), especially if work norms are strong	-Repeated actual WS or complementing WS data with other data sources (e.g., administrative data on sick leave) -Questions on work norms, telling that various work/leisure preferences can be sensible -Using databases relying on balanced sampling strategies (similar sample sizes for all weeks) (e.g., Bick, Fuchs-Schündeln, and Lagakos 2018)
ES	Long-term datasets vs. Information content low (formal definitions; self-employed, informal employment excluded; inappropriate for individual or disaggregated data)	-Only as complementary method or in historical studies
NA	Accessibility of up-to-date data vs. international comparability (OECD, TED data based on country-specific calculations: reference group, adjustments, corrections over time)	-OECD, TED datasets appropriate for time trends of national WT values -Purposefully created databases required for international comparisons (e.g., Ohanian and Raffo 2012; Wingender 2018; Bick, Brüggemann, and Fuchs‐Schündeln 2019b)
TD	Costly aggregated data (individual very difficult) less good for irregular activities	-WG method if focus is only on WT, not other time uses -Technology to decrease respondent burden (e.g., Chatzitheochari et al. 2018) -Advanced statistical methods to improve the reliability of aggregate data (e.g., Cornwell, Gershuny, and Sullivan 2019)
Other methods	Interviews: rich data vs. Costs, not generalizable	-For individual WTR or small samples -Purposive sampling -Develop interview guide for standardization
	Time sampling: many data points required, better for groups than individuals; work/non-work separation difficult	-Technology to decrease respondent burden (e.g., user-friendly application: one-touch response using a phone) -Application that allows for finer responses than work / non-work dichotomy
	Personal observation: very high costs, Hawthorne effect, work from home,	-Measurement cost is lower if WT is just one of many workplace factors studied (e.g., anthropological studies of workplace relations for individual WTR or systemic analyses of workplace processes)
	Automatic: cannot be applied in many contexts e.g., non-standard settings	-Passive data collection methods (e.g., location data) if relevant and privacy and legal considerations allow (e.g., Elevelt et al. 2021) -Often better to use as a complementary method
Indirect effects	Non-market work: hard to define & measure	-Separate questions about time spent with non-market activities (cooking, cleaning, etc.), not total non-market work -Question on how non-market activities are shared within the household, especially if gender issues are studied
	Indirect effects on others’ market work: hard to quantify	-List of interactions and relevance check – interviews, theoretical consideration

Conclusion

The world of work has been radically transformed over the last few decades, and the speed of change has only increased with the COVID-19 pandemic. New WT patterns are emerging with flexible, location-independent, and more fragmented working hours, to which the methodology of WT measurement has not yet been fully adapted. This article reviewed how these challenges are addressed in the literature, discussed the complexities, and drew conclusions for WTR research. To properly monitor and evaluate the impacts of the various WTR schemes and trials emerging around the world, WT measurement would have to satisfy several conditions. Ideally, indicators would refer to the same workers before and after the change (like repeated interviews), refer to a longer time period (like usual WT in WSs), be reliable (like personal observation), accurately measure even short periods of work (like time-sampling), and provide data in real time (like automatic measurements). In addition, the method should be able to measure non-market work (like TDs), allowing the calculation of various types of WT over the chosen groups of workers, large or small, independent of work context, also considering indirect effects. And all this should come at a reasonable cost without overburdening subjects.

As no such method exists, the approach and indicator must be selected by experts who analyze WT from a wide range of options reviewed in this article. The optimal solution will depend on the work context, the type of WTR, and the research goals, which collectively yield thousands of combinations (Table 3). In more traditional cases, the currently dominant indicator—usual WT measured by WSs—may still be practicable. However, current trends limit the scope of its application, requiring other definitions, notably actual WT, productive WT, and non-market WT, as well as other measurement methods, such as TDs, WGs, interviews, automatic measurements, and time-sampling.

If new indicators gain ground, the comparability of WT data between contexts will be diminished. This is a cost to be paid for optimizing between accuracy and feasibility on a case-by-case basis. The benefit is that context-specific indicators will make the analysis of trends over time at lower levels more meaningful, which is crucial for the future of WTR research. The divergence of WT indicators calls for indicator-specific standards regarding WT definitions and data-collection methodology. In addition, careful reporting will be inevitable to avoid confusion stemming from seemingly small, but often quite important differences discussed in this article.

Finally, we propose a method that combines time-sampling with a scale to indicate the extent to which activities at the moment of receiving the signal could be considered work. This approach could be used for a variety of contexts including the fragmented work of atypical and creative workers. With two scales, market and non-market WT could be measured simultaneously, using context-specific definitions discussed before data collection. This could be a large step toward reducing uncertainties around unpaid overtime and toward a better understanding of WTR impacts at the household level.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Work grids are sometimes called work-schedule calendars.

2 For example, “How many hours do you work in a typical week” and “How many hours do you actually work per week?” both referring to usual hours, as well as “usual weekly hours” and “usual daily hours.”

3 When administrative sources report actual WT, those values tend to be the lowest (Fleck 2009).

4 Not all definitions are equally sensitive to methodological differences. The definitions of formal WT indicators are more permanent and objectively verifiable than activity-based definitions, which still differ from each other. It may not make a large difference whether the question “How many hours do you usually work in a week?” is asked in a survey or an interview. More questions arise in the case of actual WT, which refers to a shorter and more definite period.

5 Compared to TD values. We note that no perfect method exists, so every comparison is, to some extent, questionable (Frazis and Stewart 2014).

6 This is especially relevant for historical data. Current sample sizes are less uneven. A smaller limitation for the international comparability of LFS data is that even if sampling strategies were uniform over the countries, their seasonality would cause cross-country differences because of the uneven distribution of vacation days over the year (Bick, Brüggemann, and Fuchs‐Schündeln 2019a).

7 Yet they are (mis)used in several reviewed papers for international comparisons.

8 This explains, for instance, why Germany appears to have very low annual WT values in certain comparisons (e.g., the OECD database). The country’s calculations start from ES values.

9 For a recent description of harmonized TD data and an overview of methodological changes that enabled international comparisons, see Cornwell, Gershuny, and Sullivan (2019).

10 There are many recent examples from some the best-known news sources (Simpson 2022; Nova 2021; Joly 2022).

Appendix A. Search strings

 

Q1

WT-WB search string

• (“working time” OR “working hours” OR “work hours” OR “worktime” OR “work time”) AND (“wellbeing” OR “well-being”)

Results ordered by relevance, first 200 results exported (date of search: 18 June 2020)

WT-ECON search string

• (“working time” OR “working hours” OR “work hours” OR “worktime” OR “work time”) AND LIMIT-TO (SUBJAREA, “ECON”)

Results ordered by relevance, first 200 results exported (date of search: 22 July 2020)

WT-ENV search string

• Articles found by Antal et al. (2021) with the string (“working time” OR “working hours” OR “work hours” OR “worktime” OR “work time”) AND (“energy use” OR “energy consumption” OR “energy demand” OR degrowth OR footprint OR emission OR resource*)

WTR search string

• (“working time” OR “working hours” OR “worktime” OR “work time” OR “hours worked” AND reduction*)

This search generated 2423 results (date of search: 11 November 2020).

Q2

• (“working time” OR “working hours” OR “worktime” OR “work time” OR “hours worked” OR time-use AND measure*) AND LIMIT-TO (SUBJAREA, “SOCI”)

The search found 1088 articles (date of search: 07 January 2020).

Appendix B. Differences between WS and TD/WG/calculated WT values

 

 

Appendix Table B1. Survey data distortions: summary of the main articles analyzing the WS method.

Data sources	Authors	Tested indicators	Extent and direction of average deviation	Groups with the highest and lowest deviation	Specific conclusions*
Finnish Time Use Survey	Niemi (1993)	WS actual WT/TD (1 or 2 days)	WS actual ∼3% higher than TD	Lowest: white-collar males (lower level), reporting ∼40 hours/week: ∼2%; Highest: self-employed females, reporting ∼50 hours/week: ∼30%	Factors of memory and social desirability have a significant impact on WS data.
Current Population Survey; Time-use surveys from Research Center of University of Michigan and Maryland	Robinson and Bostrom (1994)	WS actual WT/TD (1 or 2 days)	1965: WS 1 hour/week higher; 1975: WS 4 hours/week higher; 1985: WS 6 hours/week higher	Lowest: males reporting ∼40 hours/week: ∼7%; Highest: females reporting 50+ hours/week: ∼25–40%	The higher the actual WT reported in WS, the higher the deviation from TD.
National Survey of the Changing Workforce (USA)	Jacobs (1998)	WS actual WT, WS usual WT (last year)/Calculated WT: Departure to and return from work	Errors largely random in nature	No significant difference between groups (data on “hours usually worked last year” tend to have less dispersion than those that involve “hours worked last week.”)	All differences are statistical artifacts (using a definition based on departure to and return from work, deducting travel time).
Current Population Survey; American Time Use Survey	Frazis and Stewart (2004, 2014)	WS actual WT, WS usual WT/TD (1 or 2 days) with different definitions of work	WS actual ∼1,8% higher than TD; WS usual ∼7% higher than TD	Lowest: WS actual WT for full-time males: ∼2%; Highest: WS usual WT for females: ∼10%	Frequency of holidays is lower in the CPS reference period than in ATUS reference period, leading to an overestimation by WS compared to TD.
Current Population Survey; American Time Use Survey	Lin (2012)	WS usual WT/TD (1 or 2 days) with different definitions of work	WS actual ∼7% higher than TD	Lowest: hourly-paid, male, service and sales workers: ∼7%; Highest: salaried, female, professional, and office support workers: ∼12%	Usual WT is inappropriate to be directly compared to TD as it is not the mean but the most common value of weekly WT.
Danish Time Use Survey	Bonke (2005)	WS usual WT/TD (1 or 2 days)	TD ∼2% higher than WS usual WT (because of special sample)	Lowest: females reporting ∼35 hours: ∼10% (WS higher); Highest: males reporting ∼10 hours: ∼125% (TD higher)	Those reporting fewer hours underestimate WT, while those reporting many hours overestimate WT.
American Time Use Survey; Multinational Time Use Study; UK Harmonized European Time Use Study	Gershuny et al. (2007)	WS data (unspecified)/TD (1 or 2 days), weekly work grid (WG)	WS ∼14% higher than TD and ∼13% higher than WG	Lowest: females: WS-TD ∼1%; Highest: males: WS-WG ∼14%	More full-week TD data needed.
Belgian National Survey	Robinson et al. (2011)	WS usual WT/weekly TD	WS higher than TD (% cannot be determined)	Lowest: males reporting ∼30–35 hours/week; Highest: females reporting 50+ hours/week	More intensive and verifiable methods (e.g., direct observation) needed.
UK Labour Force Survey; UK Time-use Survey	Walthery and Gershuny (2019)	WS actual WT, WS usual WT/weekly WG	WS higher than WG (% cannot be determined); WS actual closer than WS usual to WG	Lowest: male regular workers; Highest: female irregular workers	Larger difference for people with more varied daily WT.

The focus is on systematic deviations from a reference method and the groups with the largest deviations.

Appendix C. A web of working times

 

 

Appendix Table C1. Potential secondary effects, production side, with examples.

Who is affected?	WT reduction	WT increase	WT indicators/methods
Household members	Partner reduces WT to enjoy leisure together with WTR participant	Partner increases market WT because WTR participant does a larger share of non-market work	TD (market + non-market WT), time-sampling, interviews
Colleagues	Productive WT of a colleague reduced because work process is dependent on WTR participant	Substitution by colleagues (overtime) or newly hired workers (whether they were previously employed matters)	Automatic measurements (in highly rationalized workplaces), WS (actual WT/usual WT if repeated measurements are not allowed), interviews (with workers or employers)
Workers of other organizations	Workers at companies supplied by an organization where WTR happens reduce WT because sectoral production is cut	Competitors supplying the same product or service as the organization where WTR happens increase WT to fill the market gap	Interviews
Members of the same community	Shorter opening hours of organization where WTR happens (e.g., kindergarten) force others to reduce, shorter WT becomes a norm	Organization where WTR happens introduces more shifts (e.g., 2 × 6 instead of 1 × 8), increasing opening hours and reducing pressure to finish early	Interviews
Others	Household-service providers needed less because more non-market work is done by WTR participant	Labor input from other countries needed more because production in country where WTR happens is cut	Interviews

As Appendix Table C1 shows, various types of links may exist. The task is to assess the potential strength of these links. Depending on the WTR scheme, the strength maybe 0% (e.g., partners not changing WT at all) or 100% (e.g., full substitution at the workplace). Obviously, thinking through potential tertiary effects or changes further down the cascade is more important for stronger effects. Whether it is realistic to assess (or even measure) all relevant impacts will depend on the scheme. An additional difficulty is that, depending on changes in salaries, consumption may also change, affecting others’ WT. Unfortunately, economic models would require too many assumptions to be useful here, and the effects of most WTR schemes are way too small to be perceptible in macro-level investigations. Therefore, WTR schemes will differ substantially in terms of the level of uncertainty caused by indirect effects, but these will often be non-negligible.