Existing definitions of bad jobs

The concept of ‘bad jobs’ has garnered increasing interest in labour market research, yet its definition remains elusive and context-dependent. The conceptualisation of bad jobs in existing literature reveals diverse approaches to establishing definitional thresholds. At its most basic level, any classification of a bad job relies upon an implied spectrum of job quality and a threshold distinguishing bad jobs from ‘other jobs’. Wage-centric approaches derived thresholds from pre-existing income poverty measures or relative pay cut-offs (Acemoglu Reference Acemoglu2001; Lee and Sobeck Reference Lee and Sobeck2012; Fusaro and Shaefer Reference Fusaro and Shaefer2016; Branch and Hanley Reference Branch, Hanley, Kalleberg and Vallas2017). However, recognition of job quality’s multi-dimensional nature has led to more sophisticated approaches. Where multiple dimensions are considered, researchers have either established separate thresholds for each dimension or reduced multiple dimensions to a single index. The latter approach has employed various methodologies, including wellbeing-driven indices (Williams et al Reference Williams, Zhou and Min2020), averaging methods, and researcher-assigned normative weights. Also, the job-strain model from psychology computes the net sum of resources (positive) and demands (negative) (Bakker and Demerouti Reference Bakker and Demerouti2008; Eurofound 2021), while others have employed systematic approaches to measuring cumulative deprivations across multiple dimensions (González et al Reference González, Sehnbruch, Apablaza, Méndez Pineda and Arriagada2021).

The evolution of these definitional approaches reflects the growing complexity in understanding job quality. While the wage-centric perspective initially prevailed, scholarly understanding has evolved towards a more comprehensive and multidimensional approach. Kalleberg et al (Reference Kalleberg, Reskin and Hudson2000) broadened the definition of bad jobs to include both low wages and the absence of health insurance and pension benefits, acknowledging that monetary compensation alone insufficiently captures job quality’s full scope. Nonetheless, this definition remained primarily focused on material aspects of employment. Schmitt (Reference Schmitt2008) further advanced the discourse by proposing a multifaceted definition incorporating wages, health insurance, pension benefits, and paid time off. According to these criteria, employment is classified as ‘bad’ if it fails to meet certain thresholds in at least two of these four categories. This framework represented a shift towards recognising the multidimensional nature of job quality, although it remained confined to tangible and (easily) measurable aspects while overlooking broader intrinsic elements, such as work intensity, autonomy, and social environment.

More recent approaches have further broadened the concept of bad jobs, integrating non-wage dimensions including job security, work intensity, autonomy, and opportunities for skill development. Howell and Kalleberg (Reference Howell and Kalleberg2019) contend that bad jobs perform poorly across multiple dimensions of job quality, suggesting that employment should be evaluated based on its provision of security, work control, and opportunities for personal and professional development. Similarly, Sehnbruch et al (Reference Sehnbruch, Apablaza and Foster2024) build upon this multidimensional framework, drawing from OECD approaches to conceptualise poor-quality employment as encompassing various undesirable characteristics affecting workers’ wellbeing, including earnings, employment stability, security, and working conditions.

Despite these theoretical advances in understanding and defining ‘bad jobs’, many studies continue to employ the term loosely, often equating it with low job quality without applying precise or theoretically grounded definitions (Green and Lee Reference Green and Lee2024). This ambiguity complicates efforts to measure the distribution of bad jobs across different contexts. Green and Lee (Reference Green and Lee2024) emphasise that underpinning any classification of bad jobs is an implied spectrum of job quality and an associated threshold. Where multidimensional data exist, normative considerations might facilitate the delineation of ‘bad’ within each dimension. However, holistically defining a job as ‘bad’ presents challenges, requiring arbitrary cross-dimensional conditions.

This paper builds upon the methodological approach validated by Green and Lee (Reference Green and Lee2024) in the European context, which emphasises the intrinsic relationship between worker wellbeing and job quality. While their framework has demonstrated validity within European labour markets, its applicability to distinctly different institutional and cultural contexts remains untested. This paper makes a unique contribution by examining whether this wellbeing-based approach to defining bad jobs can be effectively applied to the South Korean context, characterised by different labour market institutions, cultural norms, and employment practices. By constructing an index that links job characteristics to worker wellbeing in the South Korean setting, this paper examines the universal applicability of this definitional framework. This moves beyond arbitrary thresholds towards a more robust definition of bad jobs, while testing its validity outside the European context. Such an approach is particularly vital as labour markets evolve, and novel employment arrangements challenge traditional job quality conceptualisations.

South Korean labour market

South Korea’s economic growth has been one of the most impressive in the world in the modern era. Since the mid-20th century, the country has experienced a transformation from an extremely poor agrarian economy to a highly industrialised nation, establishing itself as a major player in global trade and technology. This rapid growth has been accompanied by significant increases in real GDP per capita, driven largely by the country’s export-oriented industrial policy. South Korea’s GDP grew at an average annual rate of 7.02 per cent from 1961 to 2023 (World Bank 2024), making it one of the fastest-growing economies during this period.

This trend has been reflected in the nation’s labour market, which has demonstrated sustained growth in employment rates, while exhibiting remarkable resilience to major financial crises, such as the 1998 Asian Financial Crisis and the 2008 Global Financial Crisis (FRED 2024; Lee and Rhee 2012). Following these crises, South Korea demonstrated a remarkable ability to recover, with strong job creation and a significant reduction in unemployment rates. However, this rapid recovery was accompanied by substantial structural changes in the labour market (Lee and Mcnulty Reference Lee and McNulty2003). Most notably, employers seeking greater flexibility to manage economic uncertainties increasingly turned to non-standard forms of employment, including temporary, part-time, and contract-based work, to reduce labour costs and operational risks (Shin Reference Shin2013). These positions typically lacked the stability, security, and benefits associated with standard employment, raising significant concerns about job quality, particularly regarding job security and access to social protection (Kalleberg and Hewison Reference Kalleberg and Hewison2013).

This shift intensified following the 1998 Asian Financial Crisis, leading to the emergence of a dual labour market that operates along multiple dimensions (Grubb et al Reference Grubb, Lee and Tergeist2007). The primary division exists between secure/benefit-entitled permanent workers and precarious non-standard workers who face lower wages and limited benefits (Song Reference Song2020). This segmentation is further amplified by significant disparities between large conglomerates (chaebols) and small and medium-sized enterprises (SMEs). Chaebols, such as Samsung, Hyundai, and LG, have historically dominated the Korean economy, benefiting from government support and preferential access to resources. Their privileged position enables them to offer superior financial remuneration and job quality compared to SMEs, creating a two-tier employment structure (Ha and Lee Reference Ha and Lee2013). Unlike many European labour markets, where collective bargaining, regulatory standards, and sector-level agreements tend to moderate firm-size disparities, Korea’s institutional architecture allows these divides to persist and deepen. The resulting landscape is particularly challenging for non-standard workers in SMEs, who face a double disadvantage – both the precarity of their employment status and the limitations associated with working for smaller enterprises.

Growing public concern over these labour market inequalities, particularly regarding the treatment of non-standard workers and SME employees, created pressure for systemic change in the years prior to 2017. Then, with the election of a new reforming government, the labour market underwent major regulatory reforms starting in May 2017, with the Moon Jae-In administration’s focus on addressing these structural challenges. A cornerstone of these reforms was the reduction of maximum working hours from 68 to 52 per week (40 regular hours plus 12 overtime hours), aimed at curbing chronic overwork and improving work-life balance. The administration also implemented the first major revision of the Occupational Safety and Health Act in 29 years (Government Performance Evaluation Office 2021), strengthening workplace safety standards and promoting prevention-oriented management of industrial accidents. These reforms, part of a broader agenda that included expanding public-sector employment and raising the minimum wage, marked a decisive shift from previous market-oriented policies toward stronger worker protections.

Job quality in South Korea

Despite such labour reforms, gender-based wage inequality represents a persistent challenge in South Korea’s labour market, where women earn 31.1 per cent less than men (OECD 2023a). This pronounced wage differential stems from both cultural and institutional factors, including traditional gender norms surrounding work-family roles and persistent occupational segregation. The gap is further amplified by the disproportionate concentration of female workers in lower-paying positions and non-standard employment arrangements, particularly part-time and temporary work.

Very long working hours remain a defining feature of South Korea’s labour market, with workers consistently recording some of the longest hours among OECD nations (Hijzen and Thewissen Reference Hijzen and Thewissen2020), even after substantive reductions throughout three decades. In 2021, the average South Korean worker logged 1,908 hours annually, significantly exceeding the OECD average of 1,687 hours (OECD 2023b). Multiple factors contribute to this phenomenon, including a deeply embedded corporate culture, economic growth imperatives, and organisational expectations regarding employee dedication. Research indicates that extended hours significantly impact worker wellbeing in South Korea, leading to increased burnout, elevated stress levels and deteriorating work-life balance (Park et al Reference Park, Kook, Seok, Lee, Lim, Cho and Oh2020; Baek and Yoon Reference Baek and Yoon2024), just as they have been found to do elsewhere (Virtanen et al Reference Virtanen, Stansfeld, Fuhrer, Ferrie and Kivimäki2012). The situation proves particularly challenging for non-standard workers who, despite working comparable hours, often lack access to proportionate compensation and benefits. Moreover, many South Korean workers remain in positions where their schedules are rigidly controlled by employers (Kim and Min Reference Kim and Min2023). This lack of flexibility can lead to decreased decision latitude and autonomy, important components of job quality that are associated with improved employee satisfaction and wellbeing. Comparative studies have shown that the levels of Working Time Quality and Skills and Discretion are notably worse in South Korea than in either the United States or Europe (Aleksynska et al Reference Aleksynska, Berg, Foden, Johnston, Parent-Thirion and Vanderlyeyden2019). Such findings suggest that South Korean workers often navigate environments characterised by limited autonomy and greater restrictions on their working conditions, thus increasing the likelihood of encountering bad jobs.

Other indicators of job quality also remain troubling. For example, Lee (Reference Lee2018) found that between 2010 and 2017, job security declined significantly, even though workers gained more short-term flexibility in their working hours. This shift in job security suggests a broader trend towards more precarious employment conditions. Furthermore, increased exposure to workplace hazards, such as vibrations, low temperatures, chemicals, and infectious materials, has also been documented, raising concerns about the physical safety of workers (Kim et al Reference Kim, Chung, Kim, Noh and Choi2015).

Taking stock of such trends in multiple dimensions, Lee and Green (Reference Lee and Green2025) find significant temporal variations in average job quality across multiple dimensions between 2006 and 2020. Their analysis shows a concerning deterioration in four key indices: Prospects, Skills and Discretion, Work Intensity and Physical Environment. Conversely, improvements were observed in Earnings, Social Environment and Working Time Quality during this period.

The evolving approaches to defining bad jobs, noted above, provide a useful framework for examining labour market conditions across different national contexts. With its context of high economic growth, a dualistic labour market and variable trends across different dimensions of job quality, South Korea offers a compelling case study for examining the trend and distribution of bad jobs. An evidence-based approach should enable policymakers to focus most effectively on sectors and populations where bad jobs are most prevalent.

Data set

The study deploys the KWCS, collected by the Occupational Safety and Health Research Institute (OSHRI), which benchmarked and adopted similar questionnaires from the European Working Conditions Survey (EWCS). The purpose of the KWCS is to contribute to any relevant policies on the quality of work and employment issues. The KWCS is a repeated cross-sectional survey, which has been conducted every three years since 2006 with some amendments. Since launching the first KWCS in 2006, the OSHRI has conducted a total of seven surveys (as of 2025) in 2006, 2010, 2011, 2014, 2017, 2020, and 2023, targeting employed people aged over 15 years. Of these seven waves, this study utilises four most recent surveys (2014, 2017, 2020, and 2023) due to measurement consistency requirements for job quality dimensions, yielding a final analytical sample of 88,333 observations. Trained interviewers visited each household and administered the questionnaire through Computer Assisted Personal Interviewing for all waves, but online or remote survey methods were also partially administered for the 6th wave (2020) due to the COVID-19 outbreak. The sample includes individuals who worked for pay or profit for at least one hour in the week when the survey data were collected. Clustered random sampling methods were used, and survey weights are included in all our analyses, yielding nationally representative estimates.

Empirical strategy

This study employs a two-stage empirical approach to examine the prevalence and distribution of bad jobs in South Korea.

Stage 1: Bad Job Threshold Identification

Rather than specify multiple normative criteria for each of several dimensions of job quality, we first combine these dimensions additively into a single index of job quality, then use this index’s relationship with well-being to guide us to a useful threshold between bad jobs and all other jobs. A non-parametric approach is applied to identify the optimal bad job threshold by examining the relationship between job quality and worker wellbeing, following Green and Lee’s (Reference Green and Lee2024) methodology. The job quality index is divided into deciles, as this offered greater granularity than quartiles or quintiles. The choice of deciles ensured sufficient precision in identifying non-linearities or ‘kinks’ in the relationship between job quality and wellbeing, while retaining an adequate number of observations per cell. The following model is estimated:

$$W{B_i} = \alpha + \mathop \sum \limits_{d=1}^{10} {\beta _d}J{Q_{di}} + {\varepsilon _i}$$

where $W{B_i}$ is the WHO-5 wellbeing index for individual $i$ , and $J{Q_{di}}$ are dummy variables for job quality deciles. The marginal effects ( ${\beta _{d+1}} - {\beta _d})$ capture the wellbeing gains associated with moving from decile $d$ to decile $d+1$ . Statistical significance testing determines whether the wellbeing improvement from escaping the bottom decile is significantly larger than improvements between other adjacent deciles, providing empirical justification for the bad job threshold.

Stage 2: Trend and Distributional Analysis

Having identified the bad job threshold, temporal trends and cross-sectional variation in bad job prevalence are examined using probit regression:

$$P\left( {{Y_{ij}}=1{\rm{|}}{X_{1,}}\;{X_{2,}} \ldots, \;{X_K}} \right) = {\rm{\Phi }}\left( {{\beta _0} + {\beta _1}Yea{r_i} + \mathop \sum \limits_{K=2}^n {\beta _k}{X_{Ki}} + {r_i}} \right)$$

where ${Y_{ij}}$ indicates whether a job is a bad job; ${\rm{\Phi }}$ denotes the cumulative standard normal distribution; $Yea{r_i}$ captures temporal trends; ${X_{Ki}}$ represents a vector of demographic and job characteristics, including industry, employment type, firm size, education, age, and gender; and ${r_i}$ is a normally distributed error term. Industry classification follows the Korean Standard Industrial Classification across 20 sectors, with all categories having adequate sample sizes for reliable inferences.

For comparative analysis, an identical specification is estimated using low wage employment (i.e., bottom quintile of earnings) as the dependent variable. This allows distinction between comprehensive job quality deprivation captured by the multidimensional bad jobs measure and purely monetary disadvantage reflected in low wages.

These two distinct approaches enable: a) establishment of an empirically and theoretically grounded definition of bad jobs based on worker wellbeing; b) tracking of changes in bad job prevalence over a period of significant economic and policy change; and c) identification of which worker groups and sectors face the highest risk of bad job employment, providing targeted insights for policy intervention.

Derivation of the bad jobs measure

Eurofound (2012) developed composite indices for seven theory-based dimensions of job quality, capturing various aspects of the work environment. These include Earnings, other extrinsic aspects (Prospects and Working Time Quality) and four intrinsic aspects (Skills and Discretion, Physical Environment, Social Environment and Work Intensity). Due to the general applicability of this approach and the close alignment between the EWCS and the KWCS, the methodology outlined in Eurofound (2012) has been followed. Accordingly, the same composite indicators for each of the seven job quality dimensions have been constructed using the KWCS. The job quality items and a brief description of each are presented in Table 1 below.

Table 1. Description of the job quality indices for South Korea

Note. Only the time-consistent items available in all dimensions and all waves from 2014 are used for the analyses. All measures apply to both employees and self-employed; for the Social Environment of self-employed with no employees, support measures were assigned a value of zero.

This analysis draws upon four waves of the KWCS (2014, 2017, 2020, and 2023), excluding Waves 1 (2006), 2 (2010), and 3 (2011) due to measurement inconsistencies in Prospects and Social Environment dimensions. A single job quality index was derived through principal component analysis, yielding a composite measure, defined as the first principal component, a weighted average of job quality dimensions. While this commonly used approach has limitations (see Green (Reference Green2021) for a discussion), its advantage is that it simplifies the challenge of specifying the threshold between bad jobs and all other jobs based on wellbeing differences, while incorporating all the multiple dimensions comprised in the concept of job quality. Bad jobs, defined in this way, can thereby capture jobs carrying the most severe overall deprivation. The resulting binary classification then enables both national and international comparisons. For the assessment of wellbeing, the WHO-5 Wellbeing Index, which has been available in the dataset since 2010, was employed. This index provides a validated and widely adopted metric for psychological wellbeing. The distribution of this job quality index follows an approximately normal distribution (see Figure A1 in Appendix).

Table 2 shows descriptive statistics for all job quality dimensions, which are normalised to within 0-100, the overall job quality index, and the WHO-5 Wellbeing index.

Table 2. Descriptive statistics

Note. Survey weights have been applied to all results. For the Earnings index, missing values were imputed through regression analysis, while in the Social Environment index, missing values for self-employed respondents were imputed with zeros, as these workers typically lack formal workplace social support. All job quality indices have been normalised to a scale of 0–100. The Work Intensity index was reverse-coded, with higher values representing more intensive working conditions. The analyses utilised data from 4 waves (2014, 2017, 2020, 2023). The single job quality index was generated utilising principal component analysis.

Figure 2 illustrates the incremental improvements in wellbeing associated with each decile increase in job quality, demonstrating the relationship between enhanced job quality and workers’ wellbeing.

Figure 1. The marginal effect on wellbeing of improving a decile of job quality. Note. The figure shows the marginal effects of improving one decile of job quality. The values are estimated using the following regression model: ${Y_i} = \alpha + {\beta _1}j{q_i} + {\varepsilon _i}$ where ${Y_i}$ indicates the WHO-5 wellbeing index, and $j{q_i}$ represents dummies for deciles of single job quality index generated by principal component analysis. The values shown in the figure indicate the difference between the coefficients for each job quality index decile. Survey weight is applied to the results.

The analysis reveals a pronounced U-shaped relationship between job quality and marginal effects, with distinct implications for identifying bad jobs in South Korea. The strongest marginal effects are observed in the lowest decile of job quality, where improvements yield the greatest impact on worker wellbeing. While the effect diminishes through middle deciles (3–9), reaching its lowest point around the 6th decile, it rises again for the highest quality jobs (decile 10). This non-linear pattern has two key implications.

First, this pattern enables us to define a ‘bad job’ as employment falling below the threshold, i.e., the bottom decile, where quality improvements yield distinctly larger wellbeing gains, following Green and Lee’s (Reference Green and Lee2024) approach in the European context. Rather than simply low-quality jobs, bad jobs represent positions where workers experience deprivation severe enough that quality improvements have exceptional wellbeing impact, providing an empirical, worker-centred threshold rather than arbitrary cut-offs. Statistical significance testing confirms this threshold: the wellbeing gap between the lowest and second-lowest decile (4.63 units) significantly exceeds the average gap between other deciles (2.08 units).

Notably, however, South Korea exhibits a more extended lower tail compared to European labour markets (see Green and Lee Reference Green and Lee2024), with elevated marginal effects beyond the first decile. While still maintaining a clear threshold for identifying the most severely deprived jobs, this pattern likely reflects Korea’s dual labour market structure, where the intersection of employment status (regular vs. non-regular) and firm type (chaebol vs. SME) creates multiple tiers of job quality disadvantage rather than the sharper binary distinction observed in European labour markets. Although meaningful improvements in wellbeing could be achieved by targeting both first and second deciles, our focus on the bottom decile reflects both the empirically largest marginal effects and the need to prioritise the most disadvantaged workers where policy resources are limited.

Second, the U-shaped pattern suggests that the top decile contains jobs with especially high job quality, with levels that significantly enhance individual wellbeing above those jobs in the second-highest decile. However, given the urgent policy need to address poor working conditions that significantly harm workers’ wellbeing, identifying and addressing bad jobs represents a more pressing social justice concern than promoting quality improvements throughout the spectrum. The sharp rise in wellbeing between the first and second decile could stem from either severely poor job quality at the bottom of this distribution or non-linearity in the relationship between job quality and wellbeing, or both (Warr Reference Warr2007; Green and Lee Reference Green and Lee2024). Regardless of the underlying mechanism, these findings strongly support targeting policy interventions at jobs below the 10th percentile threshold, where improvements in job quality yield the largest wellbeing gains.