Heat stress at the workplace is an occupational health hazard that reduces labour productivity1. Assessment of productivity loss resulting from climate change has so far been based on physiological models of heat exposure1. These models suggest productivity may decrease by 11–27% by 2080 in hot regions such as Asia and the Caribbean2, and globally by up to 20% in hot months by 20503. Using an approach derived from health economics, we describe self-reported estimates of work absenteeism and reductions in work performance caused by heat in Australia during 2013/2014. We found that the annual costs were US$655 per person across a representative sample of 1,726 employed Australians. This represents an annual economic burden of around US$6.2 billion (95% CI: 5.2–7.3 billion) for the Australian workforce. This amounts to 0.33 to 0.47% of Australia’s GDP. Although this was a period when many Australians experienced what is at present considered exceptional heat4, our results suggest that adaptation measures to reduce heat effects should be adopted widely if severe economic impacts from labour productivity loss are to be avoided if heat waves become as frequent as predicted.
Climate change may have profound effects on labour productivity, although few studies have estimated its economic costs5. Negative impacts of hot weather include, for instance, higher work accident frequency because of concentration lapses, higher levels of fatigue and poor decision making because time perceptions change6, 7, 8, and increased stress hormone levels which also affect cognitive performance and decision quality9, 10. Workplace heat stress was discussed in the context of increasing heat exposure from climate change2, 11. By the time of the latest IPCC report12 there was strong agreement that labour productivity will decrease as a result of increases in wet bulb globe temperature (WBGT; ref. 12), which is correlated with heat stress and a need to take breaks from labour13, 14. The only costed study of productivity impacts of heat, in Germany during 2004, estimated losses of between US$771 million and 3.4 billion for the year15.
Correlations between WBGT and physical stress, on which these studies relied, are influenced by clothing, acclimatization and micro-environments that affect evaporative cooling16. Furthermore, extrapolations from physiological models to productivity changes ignore sub-clinical impacts of heat-related disorders among those who can afford to avoid working in hot weather for their livelihoods. WBGT-based models may therefore underestimate productivity loss among people who have, or can withdraw to, cool work environments even if exposed to heat away from work.
Sub-clinical effects of heat are analogous to those of other health issues, particularly chronic diseases. Just as with disease, people affected by heat can respond to hot weather either by staying home from work (absenteeism) or attending work but performing less efficiently (presenteeism; ref. 17). Taking this analogy further, we explore here the effects of heat at work on absenteeism and presenteeism using a tailored version of the work productivity and activity impairment (WPAI) questionnaire18, which has been widely applied in health economics19. To the best of our knowledge this is the first time questionnaires developed to study the economic burden of diseases have been applied to heat stress. The only comparable study investigated the negative impacts of heat stress on Thai workers’ propensity for injury at work7, an indirect measure of productivity loss. Other studies of heat impact combined meteorological data with workers’ compensation claims20 or examined hospital discharge data for work-related accidents6.
Our study is also the first to examine the costs of heat stress in Australia, a country much affected by heat21. Within Australia, extreme heat is the most dangerous form of natural hazard, accounting for more deaths (55%) than all other natural hazards combined, even though deaths linked to heat stress are consistently under-reported21. Since 1950 there has been a significant increase in the number of heat waves, both in Australia22 and globally23, with 2013 and 2014 breaking many records4. The IPCC (ref. 24) concluded that continued warming is ‘virtually certain’ and that there is ‘high confidence’ of more frequent heat extremes in Australia25. The IPCC also considers that heat is likely to have substantial impacts on human health in Australia and calls for more research on the socio-economic impacts of climate change, including the effects on workforce participation24. As Australia has a mild to hot climate, and so will receive few benefits from amelioration of cold seasons, this study concentrated on costs because increased heat is likely to have a far greater negative than positive effect on worker productivity25.
The aims of our study were to quantify the cost of productivity loss resulting from heat stress at work, identify factors affecting productivity loss, assess which occupations were most affected, and interpret our results through the lens of health economics. Our estimates of workplace heat stress costs can be used not only to quantify the costs of unmitigated climate change and the benefits of mitigation but also to highlight the benefits of heat-stress-prevention programmes and relief strategies at work. We obtained data from an online survey in 2014 of 1,726 adults in a paid job (18 to 65) across Australia. We investigated self-reported reductions in productivity due to presenteeism and absenteeism. Presenteeism in our study is the productivity loss from reduced intensity and/or quality of labour input due to heat; absenteeism refers to the number of days of work missed because of heat17. Following the approach suggested by ref. 7, we took 12 months as the recall period to avoid any seasonal heat stress bias. Although self-reported estimates need caution in their interpretation26, the bias to which they are subject was reduced by keeping the questionnaire simple, sampling a large number of people across two sample periods (autumn and spring) and by controlling for factors known to affect productivity loss in other circumstances (see Methods).
The sample was representative in terms of gender, age, income and geographical area (Supplementary Table 1). Of the final 1,726 respondents, the majority (78%) had good or excellent health, suggesting that they had worked usual hours in the previous year.
Three-quarters (n = 1,289) of all respondents had been affected by heat at their workplace (17% of them often, 58% sometimes, 25% never; Table 1). By comparison, 50% of Thai industrial workers had at least sometimes been affected by heat stress at work7.
Table 1: Heat and its effects on productivity at work, self-assessed by respondents for the 12 months previous to either May or October 2014.
The survey was delivered through a commissioned online survey in the first two weeks of May 2014. The study was approved by the Charles Darwin University Human Research Ethics Committee (H13119). The sample was drawn from an online panel recruited by MyOpinions PermissionCorp. MyOpinions has an active panel of 300,000 verified respondents. MyOpinions has developed, and continues to maintain, an actively managed panel which adheres to a strict ‘research only’ policy governed by industry bodies such as ESOMAR, AMSRS and AMSRO. MyOpinions is also accredited to ISO 20252 and ISO 26362 professional standards and guidelines. Approximately half of the panel has been recruited from offline sources. Depending of the length of the survey, MyOpinions offers small incentives in the range of AUD 2–6 for completion of a survey.
Data.
Our data were obtained from the MyOpinions panel, from which a random sample of adults (between 18 and 65 years) in paid employment was drawn so that it was representative of the Australian population in terms of gender, age and geographical distribution. To avoid very hot periods and to minimize bias that could arise from recent experience of heat waves, we conducted the survey in two periods, one in early May (Austral autumn), the other in early October 2014 (Austral spring). A total of 2,193 people were sampled (994 in the first wave and 1,199 in the second wave) with 1,736 people completing the survey (79%). The average time to complete was 13.2 min. Those who needed less than 7 min were deleted (n = 2). Entries of respondents who were absent from work owing to illness for more than 150 days a year were also deleted (n = 4) as were those claiming an annual income >AUD 1,000,000 (n = 4). The final data set included 1,726 people.
The questionnaire.
The questionnaire consisted of two parts. The first part was about peoples’ work (occupation, sector of employment, employer, income, weekly working hours, proportion of time working outside (low = less than 10%, medium = up to 50%, high = more than 50%) and physical exertion while working). Physical exertion while working was asked as a scale question from 1 to 10, 1 meaning that the physical burden of work is very low, and 10 meaning that it is very high. The variable was transformed into three categories: low (scores 1–3), medium (scores 4–7) and high (scores 8–10).
At the end of this first part we asked two trigger questions on the impact of heat on respondent’s work. First, if respondents answered positively to having been stressed by heat at work in the previous 12 months, they were asked to identify the number of work days missed (absenteeism) and the number of days they went to work but were less productive because of heat stress (presenteeism). Second, if respondents said that they had been at work but were less productive because of heat stress on at least one day in the past 12 months, they were asked about the extent to which their work had been impaired, expressed as the percentage impairment due to heat stress. These questions were taken from the work productivity and activity impairment (WPAI) questionnaire18 used in health economics and modified to our purpose. One modification was that, following the approach of ref. 7, we used a recall period of 12 months instead of a recall period of seven days. Those respondents who had been less productive because of heat stress were then asked if they compensated for the productivity loss by working longer hours and, if so, on how many of the less productive days they compensated for lost time and for how many more hours they then worked (less than 30 min, between 30 and 60 min, more than 60 min).
The second part of the questionnaire was presented to all respondents and included questions about their demographic background (gender, age, education, postcode, nationality) and lifestyle (health status, smoking behaviour, alcohol consumption, level of exercise).
Calculation of production loss.
Total production loss (TPL) was calculated as TPL = PLA + PLP, where PLA is the annual production loss from absenteeism and PLP the annual production loss from presenteeism.
PLA was calculated for each individual as NA∗DI where NA = number of days absent per year due to heat stress and DI = daily income. Daily income was derived from respondents’ stated annual gross income. We assumed 250 working days per year and a 5-day working week. For those working full time we assumed 38 h/week, for part-time workers 19 h/week. Those who did not fall into the full- and part-time employment categories stated their actual weekly working hours.
PLP was calculated for each individual as HL∗NP∗HI, where HL = hr lost per less productive day, NP = number of d yr−1 of lower productivity and HI = hourly income. HL was calculated as p∗H where p = the percentage by which productivity was reduced on less productive days and H = number of hours per day spent working for payment.
Statistical analysis.
The R statistical package (version 2.15.3) was used to carry out statistical tests on the effect of various socio-economic, lifestyle and employment characteristics of respondents (see Table 1) on the number of days absent and present but less productive, on the level of reduced productivity, on the economic loss as a result of absenteeism and presenteeism, and on the annual total economic loss. Because the relevant parameters were not normally distributed, we applied non-parametric Kruskal–Wallis (KW) tests, followed by multiple comparison tests (equivalent to Tukey HSD) using the command kruskalmc in the R library pgirmess.
Survey limitations.
Self-reported estimates are subject to both random and systematic bias26. Although unable to remove all bias, we attempted to reduce and/or manage it in four ways. First, bias is increased by complexity46, but our questionnaire, like the medical models on which it was based18, was relatively simple. Second, we had a substantial sample size of 1,726 respondents spread evenly across two sample periods. Third, although neither of the two periods (May and October 2014, Austral autumn and spring respectively) was likely to be subject to absolute temperature extremes that could have influenced an immediate response, for one group summer had occurred within the previous six months, whereas for the other group summer was in the first half of the 12-month recall period. However, the responses of the two groups were statistically indistinguishable (apart from an extra day, on average, absent owing to illness), as were the characteristics of those sampled (Supplementary Table 3), suggesting that the survey period (May or October) did not cause undue bias. Finally, presenteeism and absenteeism in response to heat stress was broadly comparable to that in other health issues, particularly chronic diseases (Table 2). Similarly, we found that the self-stated percentage productivity loss in our study closely resembled those stated in other studies. Another way of managing bias is to control for causal inferences. In our study we tested for the influence of factors known to affect productivity loss in other circumstances (for example, proportion of time working outside and physical exertion), and the results were as expected.
Another potential limitation was that the sampled year proved to have had record-breaking heat over a large proportion of the country during the Australian summer4. Although this may have caused people to have strong recollections of heat in the previous 12 months, the predictions of ongoing increases in average temperatures and heat wave frequency in Australia22 suggests that the conditions during 2013/14 may not be exceptional for long and that their consequences should be incorporated into heat management planning. However, the conclusions could be strengthened by repeating the survey over multiple years.
Last the study could have been strengthened by deeper analysis of the causes of presenteeism and absenteeism and the interactions with disease, sleep disruption or other mental/emotional impacts.
Parsons, K. Human Thermal Environments (CRC Press, 2014).
Kjellstrom, T., Kovats, R. S., Lloyd, S. J., Holt, T. & Tol, R. S. J.The direct impact of climate change on regional labor productivity. Arch. Environ. Occup. Health64, 217–227 (2009).
Stern, N.The structure of economic modeling of the potential impacts of climate change: Grafting gross underestimation of risk onto already narrow science models. J. Econ. Lit.51, 838–859 (2013).
Morabito, M., Cecchi, L., Crisci, A., Modesti, P. A. & Orlandini, S.Relationship between work-related accidents and hot weather conditions in Tuscany (central Italy). Ind. Health44, 458–464 (2006).
Kerstin K. Zander. Heat stress causes substantial labour productivity loss in Australia[J]. Nature Climate Change,2015-05-04,Volume:5:Pages:647;651 (2015).