Elsevier

Labour Economics

Volume 47, August 2017, Pages 96-106
Labour Economics

Working hours and productivity

https://doi.org/10.1016/j.labeco.2017.03.006Get rights and content

Highlights

  • We study the effect of working hours on productivity for call centre agents.

  • We use a panel of call agents and exogenous scheduling to identify this effect.

  • We find that as working time increases, output per hour decreases.

  • Fatigue seems to play an important role even for part time workers.

  • However, service quality improves slightly with longer working hours.

Abstract

This paper studies the link between working hours and productivity using daily information on working hours and performance of a sample of call centre agents. We exploit variation in the number of hours worked by the same employee across days and weeks due to central scheduling, enabling us to estimate the effect of working hours on productivity. We find that as the number of hours worked increases, the average handling time for a call increases, meaning that agents become less productive. This result suggests that fatigue can play an important role, even in jobs with mostly part-time workers.

Introduction

Hours worked vary substantially between countries, but also within countries, e.g. due to the prevalence of part-time work and working hours regulations or agreements (BickandBrüggemann, 2016, OECD, 2016). Understanding how the number of hours worked affects labour productivity is an important element of understanding labour demand, and has important implications for the regulation of working hours and firm management. Still, a lot remains unknown about the effect of working hours on labour productivity. In theory, there could be two opposite effects. On the one hand, longer hours can lead to higher productivity if a worker faces fixed set-up costs and fixed unproductive time during the day, or if longer hours lead to better utilisation of capital goods (Feldstein, 1967). On the other hand, worker fatigue could set in after a number of hours worked, so that the marginal effect on productivity of an extra hour per worker starts decreasing (Pencavel, 2015). If neither of these effects apply, or if both cancel each other out, it could also be the case that marginal productivity does not change with working time, so that output is proportional to the number of hours worked. Identifying the effect of working time on productivity is not straightforward for two main reasons. First, unobservable characteristics of industries, firms, jobs and individuals are likely to influence both working time and productivity, so that the correlation between the two variables is likely to be a biased estimate of the effect of working time on productivity. Second, external shocks could influence both working time and productivity, which again leads to a biased estimation of the effect.

In this paper, we study the influence of the daily number of hours worked on workers’ productivity using panel data from a call centre in the Netherlands from mid-2008 to the first week of 2010 (cf. DeGripandSauermann, 2012, DeGripetal, 2016). For each of the 332 workers in our sample, the data contain detailed information on the number of daily working hours, and workers’ individual performance, as measured by the average handling time of calls. The panel structure of our data set allows us to correct for time-invariant unobserved characteristics of individuals that may influence both working time and productivity. Moreover, the exact number of hours worked by a worker on a given day is determined by central planning. Expected customer demand determines the scheduling process, and schedules are hardly related to individual preferences. This enables us to obtain estimates of the effect of working time on productivity.

Estimating a model controlling for individual fixed effects and several types of time fixed effects, we find that an increase in working hours by 1 percent leads to an increase in output by only 0.9 percent, measured as the number of calls answered. This finding suggests that fatigue sets in as working time increases. The corresponding decrease in productivity is mild in this sample where most employees work part-time, but it suggests that fatigue effects would be much stronger if agents would work full-time. We find evidence of more strongly decreasing returns to hours for workers with shorter tenure, a result that is not driven by worker attrition. Using additional data on service quality, we find that longer working hours are associated with a moderate increase of call quality in working hours, a result that partly offsets the negative effect on the number of calls answered.

This paper contributes to a rich literature that studies the link between working time and productivity. Studies estimating production functions based on industry-level data find mixed evidence for the returns to working hours. Whereas some studies find increasing returns to hours (Feldstein, 1967, Craine, 1973, Leslie, 1984), which could be the result of not taking capacity utilisation rates into account (e.g. Tatom, 1980), or be due to aggregation bias (e.g. DeBeaumont and Singell, 1999), other studies conclude that output is roughly proportional to hours worked per worker (HartandMcGregor, 1988, AnxoandBigsten, 1989, Ilmakunnas, 1994). The majority of studies, however, find evidence of decreasing returns to hours (e.g. LeslieandWise, 1980, Tatom, 1980, DeBeaumontandSingell, 1999, ShepardandClifton, 2000). Typically, studies using aggregate data deal with the endogeneity of working time by using panel data and including industry fixed effects, and by instrumenting for working time using lagged values or ranks. The validity of such instruments, however, can be questioned, and the measurement of working time and output at these aggregate levels is likely to be subject to error.

Studies using firm-level data, or data from workers in individual firms or in specific sectors are typically better at dealing with the endogeneity of working hours. A few studies use panels of firms to estimate the link between working time and firm or establishment productivity (Créponetal, 2004, Schank, 2005, Kramarzetal, 2008, GianellaandLagarde, 2011). They tend to find that output is roughly proportional to the number of hours worked.1 Due to the data structure, these studies are able to control for the endogeneity of working time caused by time-invariant firm characteristics. However, shocks that would affect both working time and productivity could still form a potential source of bias.

Studies using data about individual workers in a firm, or about workers in comparable firms date back to the early 20th century, when studies descriptively analysed the relationship between working hours and output, or compared output before and after a change in working hours (Goldmark, 1912, Vernon, 1921, Kossoris, 1947).2 More recent studies, however, exploit exogenous sources of variation in working hours to address the relationship between hours and worker level productivity. An early example is the study of citrus harvesters by Crocker and Horst (1981), who use the size of the grove worked on as a source of variation in working time. Brachet et al. (2012) conduct a difference-in-differences analysis to compare performance of paramedics working on short and long shifts. Using data from munition plants in Britain during the First World War, Pencavel (2015) uses variation in working time coming from the demand for shells to estimate the effect of working time on productivity. Dolton et al. (2016) use data from the Hawthorne experiments (conducted between 1924 and 1932) to exploit the fact that workers were subjected to different working times in different periods. While Crocker and Horst (1981) find that output is proportional to hours worked, Brachetetal, 2012, Pencavel, 2015 and Dolton et al. (2016) find evidence of decreasing returns to hours. A contrasting result is found by Lu and Lu (2016), who exploit changes in mandatory overtime laws for nurses. They find that the introduction of overtime laws actually reduced the quality provided by nurses, an effect that can be explained by changes in staffing policies of permanent and contractual (temporary) nurses.3 We contribute to this literature by exploiting exogenous variation in working hours, which is due to the call centre's central scheduling.

Most of the studies that are able to exploit exogenous variation in working time to identify the effect of working hours on productivity have concentrated on either manual workers from the first half of the 20th century (Pencavel, 2015, Doltonetal, 2016),4 or on the health sector using more recent data (Brachetetal, 2012, LuandLu, 2016). In this paper, we provide evidence about call agents in a call centre. Our results can have informative value for a broader range of medium-skilled level jobs in the service sector, and are relevant for policies such as working time regulation.

The remainder of the paper is structured as follows. In the next section, we outline our conceptual framework. Section 3 presents the empirical model we estimate and our identification strategy. Section 4 describes the data we use. Section 5 presents our main estimation results. In Section 6, we conduct a number of robustness checks, and we formulate conclusions in Section 7.

Section snippets

Model

Typically, studies of the relation between working time and productivity estimate a model of the type:Y=f(H,X)+ϵwhere Y is a measure of output, H a measure of hours worked, X is a set of variables which are also relevant for output (the capital stock being a typical candidate), and ϵ is the error term. Very often, the relationship between log output and log hours is estimated assuming a Cobb-Douglas production function. Because we focus on productivity at the level of the individual worker, for

Empirical strategy

Using panel data, our empirical model can be written asln(1AHTit)=ϕ+β·ln(Hit)+γ·Xit+δ·Tt+μi+ϵitwhere the subscript i stands for an individual agent, and t represents the time period (day or week) examined. In this section, we first discuss how we control for potential confounders, and second how we exploit scheduling as a source of exogenous variation in working hours.

Description of the context

We use rich company data on call agents employed in a call centre located in the Netherlands. This customer service centre handles calls from current and prospective customers of a mobile telecommunication company. The call centre comprises five departments, which are segmented by customer groups. To focus on workers which have comparable performance, we limit the sample to the largest department, for customers with fixed contracts.

In this department, all call agents have the same task,

Estimation results

In this section, we present the results of estimating the relation between working hours and productivity. We first address our main measure of productivity, ln(1/AHT), and then the two measures of the quality of calls. As mentioned above, we conduct the estimations both at the daily and at the weekly level.

Robustness checks

In Section 3, we discussed potential sources of endogeneity in our estimations. In this section, we check whether these potential problems indeed play a role in our data.

Conclusions

In this paper, we have estimated the impact of working time on productivity, both at the daily and at the weekly level. We have used panel data on individual workers of a call centre in the Netherlands, which contain information about the number of hours worked, the number of calls answered, the average handling time of calls and indicators of call quality for every day worked by an agent from mid-2008 until the first week of 2010. The panel character of the data enables us to control for

References (45)

  • R. Craine

    On the service flow from labour

    Rev. Econ. Stud.

    (1973)
  • B. Crépon et al.

    RTT, productivité et emploi: nouvelles estimations sur données d'entreprises

    Econ. Et. Stat.

    (2004)
  • T.D. Crocker et al.

    Hours of work, labor productivity, and environmental conditions: a case study

    Rev. Econ. Stat.

    (1981)
  • A. De Grip et al.

    The effects of training on own and co-worker productivity: evidence from a field experiment

    Econ. J.

    (2012)
  • A. De Grip et al.

    Tenure-performance profiles and the role of peers: evidence from personnel data

    J. Econ. Behav. Organ.

    (2016)
  • R. DeBeaumont et al.

    The return to hours and workers in U.S. manufacturing: evidence on aggregation bias

    South. Econ. J.

    (1999)
  • Devicienti, F., Grinza, E., Vannoni, D. 2015. The Impact of Part-Time Work on Firm Total Factor Productivity: Evidence...
  • Dolton, P., Howorth, C., Abouaziza, M., 2016. The Optimal Length of the Working Day: Evidence from Hawthorne...
  • M.S. Feldstein

    Specification of the labour input in the aggregate production function

    Rev. Econ. Stud.

    (1967)
  • A. Garnero et al.

    Part-time work, wages, and productivity evidence from belgian matched panel data

    Ind. Labor Relat. Rev.

    (2014)
  • Gianella, C., P. Lagarde, 2011. Productivity of hours in the aggregate production function: An evaluation on a panel of...
  • Goldmark, J., 1912. Fatigue and Efficiency: A Study in Industry, New York: Russel...
  • Cited by (68)

    • Working hours and job satisfaction in China: A threshold analysis

      2023, China Economic Review
      Citation Excerpt :

      Second, we utilize the instrument-variable-based two-stage residual inclusion approach to address the endogeneity of working hours. Previous studies have revealed that how long one works is endogenous (Collewet & Sauermann, 2017; Cygan-Rehm & Wunder, 2018). Ignoring this endogeneity would generate biased results and misleading policy implications.

    • Use it too much and lose everything? The effects of hours of work on health

      2022, SSM - Population Health
      Citation Excerpt :

      Combining these two observations suggests there is a non-linear relationship between work hours and health outcomes which we seek to capture using a quadratic form in work hours. Although not directly focusing on health outcomes, one explanation for the results generated in Pencavel's (2014, 2016) analysis of the nonlinear relationship between working hours and productivity in munitions factories in Britain during the First World War is stress and fatigue generated by long working hours that then affects productivity (see also Collewet & Sauermann, 2017). We contribute to the existing literature in two ways.

    View all citing articles on Scopus

    The authors would like to thank Jordi Blanes-I-Vidal, Alexandra de Gendre, Andries De Grip, Annemarie Künn-Nelen, John Pencavel, two anonymous referees, and seminar participants at the Paris School of Economics, and audience at EALE 2016 for valuable comments and suggestions. Jan Sauermann gratefully acknowledges financial support from the Jan Wallanders och Tom Hedelius Stiftelse for financial support (Grant no. I2011-0345:1).

    View full text