Job quality and automation: Do more automatable occupations have less job satisfaction and health?

Abstract

Social scientists have made predictions about the automation of jobs, as well as the negative consequences that technology has on job quality, but not how these phenomena are connected. Are occupations with a higher susceptibility of automation associated with lower job quality? This study involves an empirical examination of automation-related measures from Frey and Osborne and applies them to job quality variables (job satisfaction and self-rated health) drawn from the US General Social Survey (GSS), Quality of Working Life and Work Orientation Panel from five different waves (2002 to 2018 every 4 years; N = 7240). The finding is that highly automatable occupations have lower level of job satisfaction and health and, hence, less job quality. This has implications on the future of work, which could be but not necessarily characterised by fewer bad quality jobs.

Keywords

Automation technology job quality occupations job satisfaction health

Introduction

Social scientists have made predictions about the magnitude of impending automation of jobs, which raised concern among policymakers, media and the public about technological unemployment (Collins, 2013; Brynjolfsson & McAfee, 2014; Ford, 2015). Several empirical studies have attempted to quantify the types of occupations that are susceptible to future automation (Frey & Osborne, 2013, 2017; Arntz et al., 2016; Webb, 2020). Labour process research finds that technology and automation in capitalism impact control over work, surveillance, deskilling and other facets of job quality (Braverman, 1974; Noble, 1984; Zuboff, 1988; Thomas 1994), which is confirmed in more recent studies (Levy, 2015; Rosenblat, 2018; Griesbach et al., 2019). Furthermore, the historical arc of job quality indicates a substantial deterioration in employment standards in the last few decades giving rise to precarious work (Kalleberg & Vallas, 2018). Precarious work grows in a context where subcontracting and outsourcing are becoming a method of choice for employers (Weil, 2014; Vertesi et al., 2020), and labour unions recede as institutional pillars advocating for workers’ interest (Western & Rosenfeld, 2011; Rosenfeld, 2014).

Previous research is highly informative on job quality and technology. In this article, the author contends that there is a lack of studies on the job quality of occupations that are likely to be automated. Are occupations with a higher susceptibility of automation (high automatability) associated with lower job quality? The term automation is distinguished from ‘automatability’ in that the former is about general job displacement from technology and the latter refers to job displacement that is probabilistic. This study involves an empirical examination using computerisation as automatability measures (Frey & Osborne, 2013, 2017). The author applies this automatability measure to job quality variables in the US General Social Survey (GSS), Quality of Working Life and Work Orientation Panel from five different waves (2002 to 2018 every 4 years; N = 7240). There are a variety of ways in which job quality is measured. In this study, the author measures job quality by the level of job satisfaction and self-rated health.

The overall finding is that the highly automatable occupations exhibit lower levels of job quality. In other words, occupations that are most likely to be automated are less satisfactory and less healthy. However, there are some important qualifications: The potential replacement of low-quality jobs does not necessarily mean that they will be replaced by higher-quality jobs. An automatable occupation measured in this study is not an automated occupation as there are many factors that could influence automation and this study only uses a few indicators of job quality. The normative implications on the future of work and the limitations of this study are discussed in the conclusion.

Studies on automation

There has been increasing academic and public attention on automation replacing workers. Previous waves of automation vary by locality and industry using past data on computerisation, capital expenditure and robotisation (Kristal, 2013; Eden & Gaggl, 2018; Acemoglu & Restrepo, 2017). Another approach is to examine the forward-looking automatability of occupations based on pattern extraction of already filed patents (Webb, 2020), worker surveys on adult skills which are task-centric rather than occupation-centric (Arntz et al., 2016), and predictions of task automation mapped onto occupations following interviews/ surveys of computer scientists (Frey & Osborne, 2013). The drawback of forward-looking studies is that automatable occupations do not equate to automated occupations. In other words, jobs that are technically automatable do not necessarily become automated, or at least not universally or simultaneously. In fact, the discourse of automation hides many ‘ghost workers’, whose manual labour is hidden by customers interfacing modern technologies and the internet (Gray & Suri, 2019). Furthermore, automation can be labour-complementary rather than displacing (Shestakofsky, 2017).

Despite this caveat, the advantage of looking to the future is that it helps researchers in determining whether automatable occupations will reduce the ‘bad jobs’ (Kalleberg, 2011) albeit with the important qualification that these are not automated occupations. Bad jobs are characterised by low wages, poor benefits and non-standard work, including part-time employment, day labour, on-call work, temporary help agency and contract employment, independent contracting and other self-employment (Kalleberg et al., 2000), but also covers broader measures like job satisfaction and health (Ritter & Anker, 2002; Burgard & Lin, 2013).

Researchers who have developed automatability measures have examined their link with income and education. Frey and Osborne (2013) find the higher the automatability of an occupation, the lower the income and education level. Using patent data, Webb (2020) finds that the impact of automatability varies by type: robots will displace low-wage, less educated workers. Computers/ software will replace middle-wage occupations that require less education. Artificial intelligence (AI) will replace high-wage and well-educated workers. These automatability studies are very informative but neglect qualitative factors of job quality, which are covered in this article.

Job quality

Previous automatability studies do not address how automation is connected to broader sociological concerns around job quality and subjective well-being. Social scientists have varied emphasis on job quality with economists focusing on wages and fringe benefits (Freeman, 1981; Dale-Olson, 2006); psychologists on job satisfaction (Scarpello & Campbell, 1983; Judge et al., 2017) and sociologists on the organisation of work (Ruppaner et al., 2018). Kalleberg (2011, p. 5) defines job quality according to its economic dimensions (including pay and benefits), as well as non-economic dimensions (like control of work, intrinsic rewards and broad measures of satisfaction). Because automation is generally studied in the context of economic impacts on the workforce (Frey & Osborne, 2013; Acemoglu & Restrepo, 2017; Webb, 2020), the non-economic and social dimensions of automation need further elucidation. Furthermore, job satisfaction is only weakly correlated with wages (Clark, 2015), which suggests that social dimensions of job quality and their link to automation must be studied separately. Health is positively correlated to income, although with diminishing marginal returns at high levels of income (Ecob & Smith, 1999), which suggests that health is another independent feature of job quality. Job quality studies foreground job satisfaction (DeBustillo Llorente and Fernandez Macias, 2005; Brown et al., 2012), and physical well-being (Holman, 2013; Henseke, 2018) among others.

Automation and job quality

Some studies have examined the link between automation and job quality. Automation-related job losses have negative consequences on future job quality because more workers are shifted onto low-paid and less stable work (Schmidpeter & Winter-Ebmer, 2018; Shiller, 2019). The rise of information and communications technology (ICT), computerisation and the internet improved global connectivity, which benefited multinational corporations that offshore production to low-wage labour countries, which depressed domestic wage rises and weakened the influence of organised labour. Manufacturing, the heartland of unionisation, was especially impacted by technology-induced outsourcing (Cappelli, 1999; Kalleberg, 2011, p. 2). Globalisation implied the doubling of the global labour force, especially thanks to China, India and former Soviet states (Freeman, 2005). Even if jobs were not outsourced overseas, ICT reduced monitoring and managerial costs which enabled domestic outsourcing to poorly paid and unstable contract workers, for example, the franchise model (Weil, 2014). In the context of the current ‘lean and mean’ firm organisational model, outsourcing and automation are made possible by ICT and process automation technology (Baldwin, 2019). The spread of robotisation has reduced employment opportunities in regions where most robots have been present, whereby robots cluster in the manufacturing sector, where some high-quality jobs are still in existence (Acemoglu & Restrepo, 2017).

Workers become more fearful of job losses in workplaces with more advanced technology, suggesting less job security (Gallie et al., 2017; Dengler & Gundert, 2021). Increased technology use is associated with more stress and work intensification (Chesley, 2014). The use of people analytics and big data lower job quality by increasing managerial prerogatives (DeStefano, 2019). Automating trucker logs, for instance, have reduced the discretionary power of truck drivers (Levy, 2015). Automating the function of middle managers has increased the workloads for platform delivery drivers (Enriquez & Vertesi, 2021). Thus, the general view is that automation has an indirect (future job) or direct (more managerial power, more stress, less job security in the present job) negative effect on job quality. An important limitation in previous studies linking job quality with technology is that they are about technology changing the work experience rather than job displacement from automatability, which are covered in this article. However, these studies raise valid elements of job quality in relation to automation that are not part of this study.

A contrary perspective is that automation has had positive impacts on job quality if workers have their tasks and responsibilities expanded (Blauner 1964). Automation also positively impacts job quality if it is defined by the health status and job satisfaction of that occupation. One positive effect is the displacement of hazardous, dangerous, dull and unhealthy forms of labour (Wallén, 2008). For example, coal mining was the dominant industry in Appalachia, and even after the pit closures from automation and energy demand shift the workers continue to be plagued by health issues arising from exposure to coal dust (Suarthana et al., 2011). Mechanisation in factories has reduced repetitive work, which is an important source of worker dissatisfaction (Raffle, 1963). Automation tends to target already deskilled and repetitive labour (Attewell, 1987; Autor et al., 2003), which may be considered low-quality jobs because monotonous work is highly correlated with low job satisfaction (Melamed et al., 1995). Low-skilled routinised workers with high exposure to past robotisation are likely to report less job satisfaction, which might reflect fear of job loss due to technology (Schwabe & Castellacci, 2020). Occupations dominated by routine tasks have lower staff morale and hence less job quality (Hage & Aiken 1969) and given that routine work is more automatable (Frey & Osborne 2013) there is a potential negative relationship between routine work and job quality. Technophobes, or workers who fear technology, are more likely to report mental health issues (McClure, 2018). Hence, if job quality is operationalised by job satisfaction and health, past trends indicate that automated occupations tend to be low-quality. The contribution provided in this article is to examine whether future potential technological job displacements repeat the pattern of previous forms of automation in its tendency to replace low-quality jobs. The benefit of this approach is to examine the quality of jobs likely to be replaced, which will better help scholars to define the potential future direction of automation.

Data

Dependent variables

Job quality involves both economic and non-economic elements. Given that economics studies cover the link between automation with wages and employment levels, the author focuses on the non-economic measures of job quality. Job quality variables are drawn from the 2002 to 2018 (5 waves every 4 years, 1448 respondents per wave, which is N = 7240) GSS which has a module called ‘Quality of Working Life (QWL)’. QWL module has been commonly used by social researchers to identify trends in job quality (Smith & DeJoy 2012). GSS Data Explorer allows for specific variables to be drawn from it.¹ The unit of analysis is respondents with a given occupation in the GSS.

Respondent's health in general (health1): One measure for job quality is the respondent's general health. While some respondents have poor health for non-work reasons, it is generally the result of job characteristics. Poor job security, high work demands or excessive stress at work contribute to poor health (László et al., 2010). Health is highly correlated with measures of job quality such as skill use, discretion, work intensity and pay and job insecurity (Henseke, 2018). Therefore, low self-rated health indicates poor job quality (Arends et al. 2017). The health measure is coded ordinally (0: ‘Poor’; 1: ‘Fair’; 2:‘Good’; 3:‘Very good’; 4:"Excellent")

Job satisfaction in general (satjob1): Another good measure for job quality is self-rated job satisfaction (Kalleberg, 2011; Clark, 2015). High levels of job satisfaction imply higher job quality (Cimete et al. 2003). Job satisfaction is coded ordinally (0:‘Not at all satisfied’; 1:‘Not too satisfied’; 2:‘Somewhat satisfied’; 3:‘Very satisfied’).

Explanatory variables

The main explanatory variable involves the future automatability of occupations and is drawn from Frey and Osborne (2013):² Among the 702 occupations, the range of automatability is between 0.0028 (recreational therapist) and 0.99 (for 12 occupations including telemarketers and title examiners). Frey and Osborne's automatability by occupation are publicly available with BLS Standard Occupational Classification (SOC) codes by Jonathan Kizer.³ Six hundred and fifty-nine SOC occupations in Frey/Osborne map onto 264 International Standard Classification of Occupations (ISCO) occupations found in GSS so the author took the mean value for each ISCO (corresponding to several SOC codes) before merging to GSS.

Frey and Osborne base their automatability score on the task model approach of Autor et al. (2003), which predicts that computers are more substitutable for human labour in routine relative to non-routine tasks. Computers substitute for routine labour and are complementarities for non-routine labour. They then take 70 of the 702 occupations from the O*Net database, which contains a list of certain tasks with three groups of so-called ‘computerisation bottleneck’ (i.e. difficulty in automating the task based on present technology), including perception and manipulation (which are listed in O*Net as finger dexterity, manual dexterity, cramped workspace/ awkward position), creative intelligence (originality, fine arts) and social intelligence (social perceptiveness, negotiation, persuasion, assisting and caring for others), also cf. Table 1 in Frey & Osborne (2013). If any of these 70 occupations exhibited any of these bottlenecks, they are classified as ‘not automatable’ (0), otherwise ‘automatable’ (1). To help them make the determination, they organised a workshop at the Oxford University Engineering Sciences Department, where robotics experts and engineers offered their input. With the 70 classified occupations at hand, they use a machine learning technique to classify the remaining 632 occupations. The most accurate classifier model used an exponentiated quadratic regression equation. Thus, their selected classification scheme is probabilistic. In Frey and Osborne's model, they add three bins for low, medium and high probability of automatability (with a threshold set at 0.3 and 0.7). They believe that automation will happen in two waves. ‘In the first wave, we find that most workers in transportation and logistics occupations, together with the bulk of office and administrative support workers, and labour in production occupations, are likely to be substituted by computer capital.’ (Frey and Osborne, 2013, p. 41) In the second wave, automation will depend on being able to remove the bottlenecks, including manual dexterity (currently medium automatability) and creative and social intelligence (currently low automatability).

Table 1.

Descriptive statistics.

Statistic	N	Mean	SD	Min	Max
Job satisfaction	7268	2.35	0.74	0.00	3.00
Self-rated health	7727	2.62	1.05	0.00	4.00
Income above $25K	8464	0.59	0.49	0.00	1.00
Black	13,542	0.15	0.35	0.00	1.00
Female	13,542	0.45	0.50	0.00	1.00
4Y college	13,522	0.30	0.46	0.00	1.00
Computerisation probability	13,423	0.51	0.34	0.004	0.99
Percentile software	13,357	0.45	0.26	0.01	1.00
Percentile robot	13,357	0.43	0.29	0.01	0.98
Percentile AI	13,357	0.46	0.27	0.01	1.00
Declining job outlook	13,428	2.40	1.51	0.00	5.00

For all tables: ***p < 0.001, **p < 0.01, *p < 0.05.

In the robustness checks and appendix, the author includes alternative automatability measures by Webb (2020) and employment projections by O*Net which are each regressed on the same dependent variables.

Exposure percentile for robots/software/AI automation (Webb, 2020): Occupations exposure to robot/software/AI automation is calculated from keywords found in patents. A higher exposure means higher automatability. Because raw percentile measures range between 1 and 100, while the dependent variables are coded from 0 to 4, the raw percentile measure does not capture effects well. Therefore, the author divides the raw percentile measure for Webb's three variables by 100, to restrict the range from 0.01 to 1. Webb's patent-generated automatability scores are publicly available via his website. Details on variable construction are further discussed in the appendix.

Employment projections 2018–2028 by Occupation (O*Net) ⁴: O*Net lists all occupations with employment projections over a 10-year period, which are coded for their expected growth or decline trends. The codes include ‘Decline’, ‘Little or no change’, ‘Slower than average’ [growth], ‘Average’, ‘Faster than average’ and ‘Much faster than average’. The author codes declining numbers as 1, stable and increasing numbers as 0. The secondary measure is to code the variable ordinally with ‘Much faster than average’ coded as 0, ‘Faster than Average’ as 1, ‘Average’ as 2, ‘Slower than average’ as 3, ‘Little or no change’ as 4 and ‘Decline’ as 5. One could argue that there are a variety of reasons why occupations are declining aside from automation, including outsourcing and the decline in demand for the product that is sold by that occupation. Hicks and Devaraj (2015) point out that most of the manufacturing employment decline is linked to automation and corresponding rising domestic productivity rather than outsourcing to foreign cheap labour countries. In addition, some of the declining occupations (cashiers, machine operators, furnace operators, computer operators, etc.) in industries such as retail, manufacturing or electronics do not face declining but rather increasing demand in society (especially as the US population continues to increase). Thus, employment projections are a valid proxy for automation.

For unfilled remaining observations, the author manually fills the observations by matching the occupation name in the GSS and the automation dataset. This last step reduced the missing data for the three automation variables to less than 1%.

Control variables

The hypothesised negative relationship between job quality and automatability derives precisely from the link between low education and low job satisfaction (Vila & Garcia-Mora, 2005), although models with controls address whether this relationship would hold net of education. The link between low education, low income and high automatability have been made by Frey and Osborne (2013) and Webb (2020). In addition, Hegewisch et al. (2019) note that while 47% of workers are women, 58% of the automatable occupations are female. Cook et al. (2019) point out that Black workers concentrate on office support, food services and production work, in which more than one-third of the jobs can be displaced. The control variables are also in the GSS.

A year dummy for the variation in job quality over time (reference: 2002).

Educational attainment, Bachelor (educ) (0: 0–15 years of education, 1: 16 + years of education).

Respondent Income, above $25K annually (rincome) (0: less than 25K, 1: above 25K).

Race, black (race) (0: Non-Black, 1: Black).

Sex, female (sex) (0: Male, 1: Female).

All variables that are used in the study are found in the descriptive statistics as shown in Table 1. The lower N for the job quality variables compared to the other variables is explained by the GSS only asking a subset of all GSS respondents the quality of work panel. In the empirical analysis, the observations without job quality responses are dropped automatically.

Method

Given that the dependent variables take the form of the Likert scale, the author uses ordinal logistic regression to test the relationship between worker well-being and automatability.

Results

Using the Frey–Osborne computerisation probability measure, the author finds that highly automatable occupations have lower job satisfaction (JobSat) and worse health (Health), which holds both with and without controls (Table 2). Increasing the computerisation score by 0.1 (scale from 0.004 to 0.99) decreases the log odds of job satisfaction by 0.59 (max is 3) and log odds of self-rated health by 0.39 (max is 4). Expressed in odds ratios, the occupations with a respective increase in computerisation have only 0.55 as high job satisfaction and 0.67 as high self-rated health compared to low computerised occupations. Occupations likely to be automated have more negative job features because they are less healthy and offer a less satisfying experience. Thus, the effects of potential automation on job quality are in line with effects on job quality measured by conventional measures like education and income, where the lower-skilled and poor workers are targeted for automation (Frey, 2019). In other words, automation tends to affect poor quality jobs along both economic and non-economic dimensions.

Table 2.

Predicting job quality based on computerisation probability (Frey/Osborne)⁵.

	JobSat	Health	JobSat	Health
Computerisation Prob	−0.70 ***	−0.83 ***	−0.59 ***	−0.39 ***
	(0.07)	(0.06)	(0.08)	(0.07)
0\|1	−3.87 ***	−4.30 ***	−4.00 ***	−3.98 ***
	(0.08)	(0.09)	(0.12)	(0.13)
1\|2	−2.55 ***	−2.18 ***	−2.59 ***	−1.54 ***
	(0.05)	(0.05)	(0.10)	(0.09)
2\|3	−0.25 ***	−0.59 ***	−0.19 *	0.17 *
	(0.04)	(0.04)	(0.09)	(0.08)
3\|4		0.73 ***		1.54 ***
		(0.04)		(0.09)
2006 (ref: 2002)			0.09	0.43 ***
			(0.08)	(0.07)
2010			−0.12	0.25 ***
			(0.08)	(0.07)
2014			−0.11	0.17 *
			(0.08)	(0.08)
2018			0.08	0.24 **
			(0.08)	(0.08)
Income			0.20 ***	0.28 ***
			(0.06)	(0.05)
Black			−0.41 ***	−0.08
			(0.07)	(0.07)
Female			−0.16 **	−0.04
			(0.05)	(0.05)
College degree			−0.02	0.56 ***
			(0.06)	(0.05)
AIC	14460.12	21478.48	12091.60	17012.17
BIC	14487.66	21513.20	12172.25	17099.79
Log likelihood	−7226.06	−10734.24	−6033.80	−8493.08
Deviance	14452.12	21468.48	12067.60	16986.17
No. of obs.	7213	7667	6128	6249

The control variables show that only for the health outcome has there have been small improvements over time (no consistent change for job satisfaction). Higher-income jobs have higher levels of job satisfaction and health. Blacks and women work in occupations with less job satisfaction. College-educated workers have better health.

Robustness checks

The robustness checks include models with alternative automatability measures. Webb measures are based on pattern extraction from patents for software, robot and AI exposure. High software exposure occupations have lower rates of job satisfaction and are less healthy (online Appendix Table A1). Webb (2020) predicts software automation to target the middle-wage and those with lower levels of education. High robot exposure occupations are also less healthy and are less satisfying (online Appendix Table A2). Robot automation tends to target the less educated and low-wage workers. AI-automatable occupations results diverge from the other models. High AI exposure occupations have higher job satisfaction and are healthier without controls, but that effect disappears once accounting for the various controls (including income and education) (online Appendix Table A3). The occupational profile of workers in AI-automatable occupations is that they are more highly educated and highly compensated (Webb, 2020). This result suggests that artificial intelligence as a form of automation is much more wide-ranging than the other measures and if implemented would displace high-quality jobs but only when not accounting for income and education.

O*Net captures occupations that are likely to decline in the next decade (2018–2028). Declining occupations are associated with less satisfaction and less health, although the effect on health disappears when accounting for income and education (online Appendix Tables A4 and A5). This finding confirms the main results.

Discussion

This article has addressed the under-explored relationship between automatability and job quality. While the social science literature has focused on predicting which occupations are likely to be automated (Frey & Osborne, 2017; Webb, 2020), and the characteristics and evolution of job quality (Kalleberg, 2011; Weil, 2014), very few studies have tried to link these two phenomena. While some studies have examined the relationship between technology and job quality (Chesley, 2014; Levy, 2015; DeStefano, 2019; Dengler & Gundert, 2021) and automation and job quality more directly (Schmidpeter & Winter-Ebmer, 2018; Levy, 2015; Enriquez & Vertesi, 2021), they have not used survey data to establish the job quality of occupations that are likely to be automated. The finding is that automatable occupations correspond to lower job quality because they exhibit less job satisfaction, and workers in these occupations are less healthy. This relationship holds both with and without controls for income, education and demographic characteristics.

Automation, especially in the early phase, is most likely to target jobs that require less education and have lower incomes (Frey & Osborne, 2017; Webb, 2020), which, in turn, implies that lower quality ‘bad’ jobs are targeted by the potential automation wave. The results presented in this article hold up that relationship for subjective measures of job quality, even holding constant income and education. It also confirms the intuition that low-skilled routinised workers with high automation exposure report less job satisfaction due to fear of technological displacement (Schwabe & Castellacci, 2020). The Frey and Osborne data used in this study constructs automatability variables based on the routine task-based framework in Autor et al. (2003), indicating also that routine jobs are likely to exhibit lower job quality. The theoretical significance of this article is to contribute to the ongoing automation debate by examining non-economic elements of job quality in occupations that are most likely to be replaced by technological change. It is insufficient to focus automation research on the number of occupations likely to be automated, but it is relevant for researchers to understand whether the present trend towards automation affects the good or the bad jobs.

Aside from social scientific concerns, this article raises ethical considerations in the future of work. The engineers and roboticists, who develop automation technology, retain the belief that routine grunt work is becoming automated, allowing workers to take on more sophisticated and creative tasks (Halal et al., 2017). The implication is the optimistic vision that automation will displace the undesirable, poor quality jobs, while the new jobs will be desirable. The findings presented in this article only support the view that the replaceable jobs have less job satisfaction and are less healthy but does not say anything about the quality of future jobs. The fact that potential automation is removing low-quality jobs has important implications on workers in low-quality jobs like warehousing, especially given that current robotisation is not yet replacing warehouse work but resulting in work intensification (Ghaffary, 2019) and thereby keeping job quality low in existing jobs.

The replacement of low-quality jobs does not guarantee that new jobs that are created are high quality. The evolution of job quality from low regulation/unionisation and low quality to high regulation and high quality and the swing back to less regulation and de-unionisation since the 1970s (Levy & Temin, 2007; Kalleberg, 2011) suggests that generating high job quality in the form of worker power and control, which were not subject to analysis in the present study, is entirely a contingent institutional process and cannot be expected from managerial and engineering innovations alone. Even if one assumes that the job apocalypse will not happen, newly created occupations are not guaranteed to deliver high job quality unless pro-worker institutional designs are prioritised. Understanding the job quality implications of automatable occupations will help policymakers to find ways to generate new high-quality jobs. Furthermore, the fact that AI automation is the one area where high-quality jobs- not accounting for income and education- are targeted for automation suggests that long-term future automation may well remove more desirable forms of employment. This trend would make debates on universal basic income and shorter work weeks more relevant.

There are some important limitations in the study. First, an automatable occupation is not an automated occupation, emphasising the contingent and socially embedded process of automation (Shestakofsky, 2017; Gray & Suri, 2019). The automation data sources used in this study are either derived from roboticist expectations about occupations that are susceptible to automation, inferred from language found in patents or are a rough proxy to higher automatability based on future occupational trends. Even robotics and AI experts tend to qualify their automation prediction by being more conservative about the scale and speed of automation than the general public (Walsh, 2018). Furthermore, while many tasks in many jobs can be automated, only 5% of occupations could have all their tasks automated (Chui et al. 2015). Logically, it is possible for more automatable occupations to not become displaced at all or at a later point than occupations that are deemed less automatable. If that were the case, then the analysis provided in the present article is less useful. However, the author maintains that in the absence of better automatability data, the present approach is the best one available.

The second limitation is that the average occupational job quality may not correspond well to individual aspirations. For instance, while it is reasonable to assume that a lower satisfaction job corresponds to lowered job quality for many, some people may find meaning outside of work. Also, workers in what would normally be considered low-quality jobs (due to low pay, irregular work and difficult working conditions) can report high level of job satisfaction because their past job experience was even worse (Léné, 2019). This is especially true given that automation is measured at the occupation level, while job quality is measured at the individual level, thus creating a potential mismatch between individual experience and experiences shaped by membership in an occupation.

Furthermore, job satisfaction has been critiqued previously as a reflection of job quality (DeBustillo Lorente & Fernandez Macias, 2005). The potential displacement of less healthy jobs does not necessarily mean that the workers that become displaced have a positive health outcome or find healthier jobs: on the contrary, displaced workers’ health deteriorates (Patel et al., 2018). Hence, the results presented here must be interpreted cautiously given that satisfaction and health cover only part of job quality. Job quality also covers job insecurity, opportunities for career advancement, dirtiness of work, complexity of the job and control over the pace, timing and content of work (Leschke & Watt, 2008; Hackman et al., 1978; Cain & Treiman, 1981; Green et al., 2013). Future studies should explore broader measures of job quality in relation to automatability and the possibility that high-quality jobs on other dimensions are most likely to be automated. Via qualitative studies and better quantitative data, a deeper grasp of job quality at the workplace level and how it is impacted by the forces of technology and organisational change is eminently important.

Supplemental Material

sj-docx-1-jir-10.1177_00221856221129639 - Supplemental material for Job quality and automation: Do more automatable occupations have less job satisfaction and health?

Supplemental material, sj-docx-1-jir-10.1177_00221856221129639 for Job quality and automation: Do more automatable occupations have less job satisfaction and health? by Larry Liu in Journal of Industrial Relations

Supplemental Material

sj-csv-2-jir-10.1177_00221856221129639 - Supplemental material for Job quality and automation: Do more automatable occupations have less job satisfaction and health?

Supplemental material, sj-csv-2-jir-10.1177_00221856221129639 for Job quality and automation: Do more automatable occupations have less job satisfaction and health? by Larry Liu in Journal of Industrial Relations

Footnotes

Acknowledgements

The author would like to thank the feedback from Miguel Centeno, Janet Vertesi, Adam Goldstein and the participants of the Labor and Employment Relations Association Conference, 2021, including Tingting Zhang, Andrew Weaver, Suyeon Kang, Turner Cotterman, Mitchell Small, Erica Fuchs and Adam Seth Litwin.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article

ORCID iD

Larry Liu

Supplemental material

Supplemental material for this article is available online.

Notes

Biographical note

Larry Liu is a PhD student at Princeton University Department of Sociology. He researches the future of work and automation within labor, economic and political sociology. He published in Social Forces, Socio-Economic Review and Ethnic and Racial Studies, among others.

References

Acemoglu

Restrepo

(2017) Robots and jobs: Evidence from US labour markets. National Bureau of Economic Research Working Paper, w23285.

Arends

Prinz

Abma

(2017) Job quality, health and at-work productivity. OECD Social, Employment and Migration Working Paper, No. 195.

Arntz

Gregory

Zierahn

(2016) The risk of automation for jobs in OECD countries: A comparative analysis. OECD Social, Employment and Migration Working Paper, No. 189.

Attewell

(1987) The deskilling controversy. Work and Occupations 14(3): 323–346.

Autor

Levy

Murnane

(2003) The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics 118(4): 1279–1333.

Baldwin

(2019) The globotics upheaval: globalization, robotics, and the future of work. New York: Oxford University Press.

Blauner

(1964) Alienation and freedom: The factory worker and his industry. Chicago: Chicago University Press.

Braverman

(1974) Labour and monopoly capital. New York: Monthly Review Press.

Brown

Charlwood

Spencer

(2012) Not all that it might seem: Why job satisfaction is worth studying despite it being a poor summary measure of job quality. Work, Employment and Society 26(6): 1007–1018.

10.

Brynjolfsson

McAfee

(2014) The second machine age: work, progress, and prosperity in a time of brilliant technologies. New York and London: W.W. Norton & Company.

11.

Burgard

Lin

(2013) Bad jobs, bad health? How work and working conditions contribute to health disparities. American Behavioral Scientist 57(8): 1105–1127.

12.

Cain

Treiman

(1981) The dictionary of occupational titles as a source of occupational data. American Sociological Review 46(3): 253–278.

13.

Cappelli

(1999) The new deal at work: managing the market-driven workforce. Boston: Harvard Business School Press.

14.

Chesley

(2014) Information and communication technology use, work intensification and employee strain and distress. Work, Employment and Society 28(4): 589–610.

15.

Chui

Manyika

Miremadi

(2015) How many of your daily tasks could be automated? Harvard Business Review, December 14.

16.

Cimete

Nimet

Gülbahar

(2003) Quality of life and job satisfaction of nurses. Journal of Nursing Care Quality 18(2): 151–158.

17.

Clark

(2015) What makes a good job? Job quality and job satisfaction. IZA World of Labour, Article 215. DOI: 10.15185/izawol.215

18.

Collins

(2013) The end of middle-class work: No more escapes. In: Wallerstein

Collins

Mann

Derluguian

Calhoun

(eds) Does Capitalism Have a Future. Oxford: Oxford University Press, pp. 37–70.

19.

Cook

Pinder

Stewart III

, et al. (2019) The Future of Work in Black America. McKinsey & Company. https://www.mckinsey.com/featured-insights/future-of-work/the-future-of-work-in-black-america.

20.

Dale-Olsen

(2006) Wages, fringe benefits and worker turnover. Labour Economics 13(1): 87–105.

21.

DeBustillo Llorente

Fernandez Macias

(2005) Job satisfaction as an indicator of the quality of work. The Journal of Socio-Economics 34(5): 656–673.

22.

Dengler

Gundert

(2021) Digital transformation and subjective job insecurity in Germany. European Sociological Review 37(5): 799–817.

23.

DeStefano

(2019) ‘Negotiating the algorithm’: Automation, artificial intelligence and labour protection. Comparative Labour Law & Policy Journal 41(1): 15–46.

24.

Ecob

Davey Smith

(1999) Income and health: What is the nature of the relationship? Social Science & Medicine 48(5): 693–705.

25.

Eden

Gaggl

(2018) On the welfare implications of automation. Review of Economic Dynamics 29: 15–43.

26.

Enriquez

Vertesi

(2021) Managing Algorithms: partial automation of middle management and its implications for gig worker. Academy of Management Proceedings: (online). https://journals.aom.org/doi/abs/10.5465/AMBPP.2021.16560abstract

27.

Ford

(2015) Rise of the robots: technology and the threat of a jobless future. New York: Basic Books.

28.

Freeman

(2005) China, India and the doubling of the global labour force: who pays the price of globalization? The Asia-Pacific Journal 3(8), https://apjjf.org/-Richard-Freeman/1849/article.html.

29.

Freeman

(1981) The effect of unionism on fringe benefits. Industrial and Labour Relations Review 34(4): 489–509.

30.

Frey

(2019) The technology trap: capital, labour, and power in the age of automation. Princeton: Princeton University Press.

31.

Frey

Osborne

(2013) The future of employment. Oxford University, Martin School, Working Paper.

32.

Frey

Osborne

(2017) The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change 114: 254–280.

33.

Gallie

Felstead

Green

, et al. (2017) The hidden face of job insecurity. Work, Employment and Society 31(1): 36–53.

34.

Ghaffary

(2019) Robots aren’t taking warehouse employees’ jobs, they’re making their work harder. Vox.

35.

Gray

Suri

(2019) Ghost work: How to stop Silicon Valley from building a new global underclass. New York: Eamon Dolan Books.

36.

Green

Mostafa

Parent-Thirion

, et al. (2013) Is job quality becoming more unequal? Industrial and Labour Relations Review 66(4): 753–784.

37.

Griesbach

Reich

Elliott-Negri

, et al. (2019) Algorithmic control in platform food delivery work. Socius 5: 1–15.

38.

Hackman

Pearce

Wolfe

(1978) "Effects of changes in job characteristics on work attitudes and behaviors: A naturally occurring quasi-experiment. Organizational Behavior and Human Performance 21(3): 289–304.

39.

Hage

Aiken

(1969) Routine technology, social structure, and organization goals. Administrative Science Quarterly 14(3): 366–376.

40.

Halal

Kolber

Davies

, et al. (2017) Forecasts of AI and future jobs in 2030: Muddling through likely, with two alternative scenarios. Journal of Futures Studies 21(2): 83–96.

41.

Hegewisch

Childers

Hartmann

(2019) Women, automation, and the future of work. Institute for Women’s Policy Research. https://iwpr.org/wp-content/uploads/2019/03/Automation-Final-Report-1.pdf.

42.

Henseke

(2018) Good jobs, good pay, better health? The effects of job quality on health among older European workers. The European Journal of Health Economics 19(1): 59–73.

43.

Hicks

Devaraj

(2015) The myth and reality of manufacturing in America. Ball State University, Muncie, IN: Center for Business and Economic Research. https://conexus.cberdata.org/files/MfgReality.pdf.

44.

Holman

(2013) Job types and job quality in Europe. Human Relations 66(4): 475–502.

45.

Judge

Weiss

Kammeyer-Mueller

, et al. (2017) Job attitudes, job satisfaction, and job affect: A century of continuity and of change. Journal of Applied Psychology 102(3): 356–374.

46.

Kalleberg

(2011) Good jobs, bad jobs: the rise of polarized and precarious employment systems in the United States, 1970s-2000s. New York: Russell Sage Foundation.

47.

Kalleberg

Reskin

Hudson

(2000) Bad jobs in America: Standard and nonstandard employment relations and job quality in the United States. American Sociological Review 65(2): 256–278.

48.

Kalleberg

Vallas

(2018) Probing precarious work: Theory, research, and politics. Research in the Sociology of Work 31(1): 1–30.

49.

Kristal

(2013) The capitalist machine: Computerisation, workers’ power, and the decline in labour’s share within US industries. American Sociological Review 78(3): 361–389.

50.

László

Pikhart

Kopp

, et al. (2010) Job insecurity and health: A study of 16 European countries. Social Science & Medicine 70(6): 867–874.

51.

Leschke

Watt

(2008) Job quality in Europe. ETUI-REHS Research Department.

52.

Levy

Temin

(2007) Inequality and institutions in 20th century America. National Bureau of Economic Research, Working Paper No. w13106.

53.

Levy

KEC

(2015) The contexts of control: Information, power, and truck-driving work. The Information Society 31(2): 160–174.

54.

Léné

(2019) Job satisfaction and bad jobs: Why are cleaners so happy at work? Work, Employment and Society 33(4): 666–681.

55.

McClure

(2018) ‘You’re fired,’ says the robot: The rise of automation in the workplace, technophobes, and fears of unemployment. Social Science Computer Review 36(2): 139–156.

56.

Melamed

Ben-Avi

Luz

, et al. (1995) Objective and subjective work monotony: Effects on job satisfaction, psychological distress, and absenteeism in blue-collar workers. Journal of Applied Psychology 80(1): 29–42.

57.

Noble

(1984) Forces of production: A social history of industrial automation. New York: Knopf.

58.

Patel

Devaraj

Hicks

, et al. (2018) County-level job automation risk and health: Evidence from the United States. Social Science & Medicine 202: 54–60.

59.

Raffle

PAB

(1963) Automation and repetitive work. Their effect on health. Lancet (London, England) 281(7284): 733–737.

60.

Ritter

Anker

(2002) Good jobs, bad jobs: Workers’ evaluations in five countries. International Labour Review 141: 331–358.

61.

Rosenblat

(2018) Uberland: How algorithms are rewriting the rules of work. Berkeley: Univ of California Press.

62.

Rosenfeld

(2014) What unions no longer do. Cambridge: Harvard University Press.

63.

Ruppanner

Lee

Huffman

(2018) Do mothers benefit from flexible work? Cross-national evidence for work time, job quality, and satisfaction. International Journal of Sociology 48(2): 170–187.

64.

Scarpello

Campbell

(1983) Job satisfaction: Are all the parts there? Personnel Psychology 36(3): 577–600.

65.

Schmidpeter

Winter-Ebmer

(2018) How do automation and offshorability influence unemployment duration and subsequent job quality? CEPR Discussion Paper No. DP13112.

66.

Schwabe

Castellacci

(2020) Automation, workers’ skills and job satisfaction. Plos one 15(11): e0242929.

67.

Shestakofsky

(2017) Working algorithms: Software automation and the future of work. Work and Occupations 44(4): 376–423.

68.

Shiller

(2019) Narratives about technology-induced job degradations then and now. No. w25536. National Bureau of Economic Research.

69.

Smith

DeJoy

(2012) Occupational injury in America: An analysis of risk factors using data from the general social survey (GSS). Journal of Safety Research 43(1): 67–74.

70.

Suarthana

Laney

Storey

, et al. (2011) Coal workers’ pneumoconiosis in the United States: Regional differences 40 years after implementation of the 1969 federal coal mine health and safety act. Occupational and Environmental Medicine 68(12): 908–913.

71.

Thomas

(1994) What machines can’t do: Politics and technology in the industrial enterprise. Berkeley: University of California Press.

72.

Vertesi

Goldstein

Enriquez

, et al. (2020) Pre-Automation: Insourcing and automating the gig economy. Sociologica 14(3): 167–193.

73.

Vila

García-Mora

(2005) Education and the determinants of job satisfaction. Education Economics 13(4): 409–425.

74.

Wallén

(2008) The history of the industrial robot. Linköping: Linköping University Electronic Press.

75.

Walsh

(2018) Expert and non-expert opinion about technological unemployment. International Journal of Automation and Computing 15(5): 637–642.

76.

Webb

(2020) The impact of artificial intelligence on the labour market. Stanford University, Working Paper.

77.

Weil

(2014) The fissured workplace. Cambridge: Harvard University Press.

78.

Western

Rosenfeld

(2011) Unions, norms, and the rise in US wage inequality. American Sociological Review 76(4): 513–537.

79.

Zuboff

(1988) In the age of the smart machine: The future of work and power. New York: Basic Books.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

5.33 MB

0.03 MB