Article Text

Download PDFPDF

Short report
Poor health as a potential risk factor for job loss due to automation: the case of Norway
  1. Philipp Hessel1,2,
  2. Solveig Christiansen1,
  3. Vegard Skirbekk1,3
  1. 1 Department of Epidemiology, Norwegian Institute of Public Health, Oslo, Norway
  2. 2 Alberto Lleras Camargo School of Government, University of the Andes, Bogotá, Colombia
  3. 3 Columbia Aging Center, Mailman School of Public Health, Columbia University, Oslo, Norway
  1. Correspondence to Professor Vegard Skirbekk, Centre for Fertility and Health, Norwegian Institute of Public Health, Marcus Thranes Gate 6, 0473 Oslo, Norway; vegard.skirbekk{at}fhi.no

Abstract

Objective This study aimed to quantify the extent to which health characteristics of workers are related to the potential risk of experiencing job displacement due to automation.

Methods Linking the 2015 Norwegian Statistics on Income and Living Conditions survey (n=6393) with predicted probabilities of automation by occupation, we used Kruskal-Wallis tests and multivariate generalised linear models to assess the association between long-standing illnesses and risk of job automation.

Results Individuals with long-standing illnesses face substantially greater risks of losing their job due to automation. Whereas the average risk of job automation is 57% for men and 49% for women with long-standing illnesses, the risk is only 50% for men and 44% for women with limitations (p<0.001). Controlling for age, having a long-standing illness significantly increases the relative risk of facing job automation among men (risk ratio (RR) 1.13, 95% CI 1.09 to 1.19), as well as women (RR 1.11, 95% CI 1.05 to 1.17). While, among men, the association between long-standing illness and risk of job automation remains significant when controlling for education and income, it becomes insignificant among women.

Conclusions Individuals with poor health are likely to carry the highest burden of technological change in terms of worsening employment prospects because of working in occupations disproportionally more likely to be automated. Although the extent of technology-related job displacement will depend on several factors, given the far-reaching negative consequences of job loss on health and well-being, this process represents a significant challenge for public health and social equity.

  • unemployment
  • automatisation
  • Norway
  • job loss

Statistics from Altmetric.com

What this paper adds

  • Studies have estimated that about half of all jobs in many advanced economies could, technically, be automated in coming decades, eventually resulting in widespread unemployment and job insecurity.

  • Using data for Norway, we show that individuals with poor health face a disproportionally larger risk of facing job automation, which could significantly exacerbate existing health disadvantages in the future.

Introduction

Computers and machines have replaced large numbers of jobs, significantly shifting job opportunities and skill demand in favour of high-educated and high-skilled individuals.1 While most of those jobs replaced by computers or machines have been so-called routine jobs, including machine operators and bookkeepers, the spread of new technologies in professions such as law, financial services, education and medicine has intensified debates about the potential negative consequences of technological change, including mass unemployment, employment polarisation and economic inequality.

Although the precise effect of technological change on employment prospects is difficult to predict, studies have suggested that up to half of all jobs in Europe and the USA are at risk of disappearing due to automation.2–4 Given the far-reaching adverse consequences of job loss, unemployment and economic insecurity on physical, as well as mental, health,5 the possible loss of a significant number of jobs due to automation, particularly among lower educated workers, represents a major challenge for public health and social equity.

However, to date, we know little about the relationship between sociodemographic or health characteristics and the risk of job loss due to automation. While lower educated individuals are more likely to work in lower skilled occupations with, on average, higher risks of automation,2 no study has assessed the relationship between individual health and risk of job loss due to technological change. By using a linkage of the 2015 Norwegian Statistics on Income and Living Conditions (SILC) survey with predicted probabilities of automation, we address this gap by providing estimates of the risk of job automation related to different health conditions.

Method

Data

Individual-level data came from the 2015 Norwegian SILC survey, a representative survey of non-institutionalised individuals aged 15 and above, with the aim of providing data on sociodemographic conditions, including labour force participation and health. Since our aim was to assess the association between health characteristics and the risk of job displacement based on current occupation, for the purpose of this study, we only included individuals between ages 20 and 70 who were employed at the time of interview, thus excluding full-time students, homemakers, unemployed people or pensioners.

To derive an indicator of the probability of job automation, we matched individuals’ occupations recorded in SILC with detailed information on automation for 703 occupations provided by Frey and Osborne,2 widely considered as the most detailed study of its kind. To quantify the risk of automation by occupation, the authors used data from the US Department of Labor/Employment and Training Administration’s so-called Occupational Information Network (O*NET).6 Based on surveys among employers and employees, the latter collects detailed information on occupational requirements, skills and tasks, as well as the use of tools and technology, including their frequency and importance to each job.

Using this information, engineering experts then subjectively classified 70 occupations for which they felt highly confident to make an assessment according to whether they can be automated or not. Jobs were classified as either fully automatable, indicating whether a job can, in principle, principally be performed by state-of-the-art computer-controlled equipment, or fully non-automatable. Frey and Osborne then developed a machine-learning algorithm to calculate probabilities of automation for all 703 occupations included in O*NET. In a first step, the algorithm examined whether the subjective expert classifications were related to specific job-specific skills in the O*NET database, including manual dexterity, originality and social perceptiveness. Using the specific skills associated with jobs that were either considered fully automatable or non-automatable, the algorithm then examined the accuracy of the subjective assessments regarding automatability for all occupations, resulting in a probabilistic prediction for the risk of automation for each occupation.

Instruments

Probability of job automation

Using information from Frey and Osborne about the probability of automation for 703 occupations, we matched the six-digit Standard Occupational Classification codes used by the US Bureau of Labor Statistics with the International Standard Classification of Occupations codes used in SILC. This linkage thus allowed us to compute the probability of job automation, ranging from zero to one, for each individual included in SILC.

Health

Based on self-reports from SILC, we included a measure capturing whether respondents had any long-standing illness or health problem (including blindness or serious vision impairment, deafness or serious hearing impairment, difficulty with basic physical activities such as walking, climbing stairs, reaching, lifting, carrying, an intellectual disability, difficulty with learning, remembering or concentrating, psychological or emotional condition, or difficulty with pain, breathing or any other chronic illness or condition) that has been shown to be consistently associated with prevalence of chronic diseases as well as limiting long-standing ilnesses.7

Covariates

The multivariate analyses included controls for age, education and income.

Statistical analysis

We first used univariate Kruskal-Wallis tests to assess whether individuals’ risk of job automation differed significantly according to their health status, separately for men and women. We then used multivariate generalised linear models (GLM) regressing individuals’ risk of job automation on indicators of health, controlling for sociodemographic characteristics. We implemented GLMs using a binomial distribution and a log link with robust standard errors.

Results

Around 28.5% of men in our sample report having a long-standing illness, compared with around 37% among women (table 1). Overall, men have a significantly greater risk of experiencing job displacement due to automation (56% mean probability) compared with women (49% mean probability). The predicted mean probability of job automation in Norway is lower among managers (17%) and professionals (15%) than among clerical support workers (90%), plant or machine operators (78%) and those working in primary industries (75%) (online supplementary table 1).

Supplementary file 1

Table 1

Descriptive statistics and relative risk of job automation according to health characteristics among Norwegian men and women

Table 2 shows the results of a set of multivariate GLMs regressing individuals’ risk of job automation on indicators of health, controlling only for age. Results obtained by using a binomial distribution and a log link can be interpreted as risk ratios (RR). Suffering from a long-standing illness significantly increases the relative risk of experiencing job automation by 13% among men (RR 1.13, 95% CI 1.09 to 1.19) and 11% among women (RR 1.11, 95% CI 1.05 to 1.17). As when controlling for education and income, the relative risk of experiencing job automation associated with the prevalence of a long-standing illness among men is reduced, but remains significant (RR 1.05, 95% CI 1.01 to 1.09). However, among women, the association between long-standing illness and job automation risk is no longer not significant when controlling for education and income (RR 1.02, 95% CI 0.92 to 1.05).

Table 2

Relative risk of job automation according to health characteristics among Norwegian men and women

Discussion

The principal aim of this study was to assess the extent to which poor health is associated with the risk of experiencing job displacement due to automation. For this purpose, we linked individual-level data from the 2015 Norwegian SILC with detailed information on predicted probabilities of automation for each occupation. While about half of all jobs in Norway could be automated in the future, the results show that individuals with long-standing illnesses face a significantly larger risk of job loss due to automation compared with those in better health.

Norway is a highly technologically developed nation, with some of the highest labour costs in the world, particularly salaries for low-skilled labour. There are still many jobs in the low-skilled sector, yet Norway has a compressed salary structure with relatively high salaries, particularly for those in this segment of the workforce. Norway may, therefore, be one of the first countries to experience the consequences of mass-scale automation, due to the high level of cost-effectiveness of introducing robots to automate work processes.

While this is the first study to quantify differences in the risk of job automation in relation to health conditions, the results rely on expert assessment regarding the likelihood of certain jobs being automated. The approach of this study rests on the assumption that specific jobs are comparable between the USA and Norway. To empirically assess this assumption, we used individual-level data from the Organisation for Economic Co-operation and Development’s Survey of Adult Skills (Programme for the International Assessment of Adult Competencies) comparing frequency of skill use in several dimensions strongly associated with job automatability according to Frey and Osborne.2 As results in online supplementary table 2 suggest, after controlling for sociodemographic characteristics, there exist no significant differences between the USA and Norway regarding the use of key skills associated with job automatability. While the study by Frey and Osborne2 is among the most advanced studies of their kind to date, their assessment is based on the technical potential of a job being automated. In consequence, their approach does not take into account future developments in macroeconomic conditions, such as factor prices for labour and computers or the degree to which workers whose main tasks become automated can find new roles. Finally, there may be legal and regulatory barriers that may prevent or slow down the automation of certain tasks.

Although the findings are largely driven by correlations between multiple risk factors, including low education and income, poor health and lower occupational class, the possibility of a large number of jobs being displaced presents a key challenge for social welfare and health systems. Widespread technologically induced unemployment could possibly undermine meaning people attach to their lives, especially if not confined to the economic sphere.8 Threats of losing work through automation could potentially lead to trauma and fear.9 In particular, the fact that those individuals already suffering from poor health are significantly more likely to face job automation could considerably exacerbate existing health conditions, as well as inequalities. A first step to addressing the challenge posed by technological change is, thus, to acknowledge that the threat it represents for future employment prospects is disproportionally larger for less educated and healthy individuals.

Acknowledgments

This work was partly supported by the Research Council of Norway through its Centres of Excellence funding scheme, project number 262700.

References

Footnotes

  • Contributors VS had the idea for the paper. All authors designed the empirical analyses that were carried out by SC. All authors interpreted the results. PH wrote the first draft of the paper, which was subsequently revised by all authors.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.