Educational job mismatch, job satisfaction, on-the-job training, and employee quit behaviour: a dynamic analytical approach

ABSTRACT

This paper extends the literature on the consequences of over-education, in particular quit outcomes. It is the first study that explicitly tests the impact of job satisfaction and on-the-job training for workers in educational mismatched jobs and on quit behaviour using a longitudinal data set. Accounting for unobserved heterogeneity and endogeneity, the dynamic analytical framework examines labour market outcomes for job-mismatched workers. We find that over-education alone, or accompanied by skill under-utilization in combination with lower job satisfaction, increases the incidences of job quitting. Opportunities for training facilitate the retention of initially job-mismatched workers. These results have implications for interpreting mismatch data, retention, and resource allocation.

KEYWORDS:

• Over-education
• over-skilling
• job satisfaction
• on-the-job training
• turnover

JEL CLASSIFICATION:

• J24
• J28
• J31
• J63

Acknowledgment

This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the authors and should not be attributed to either FaHCSIA or the Melbourne Institute.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ This brief overview is not intended to be an exhaustive coverage of career theory or the related literature.

² Our sample includes all education groups to allow a larger sample and a realistic representation of the groups and markets that workers operate in. By controlling for ‘actual years of education’, and over- or under-education, we are also able to effectively separate different education groups in our model. A number of existing studies incorporate all education groups. Examples are for the labour market in Germany evaluated by Büchel and Mertens (2004); the US labour market examined by Rubb (2006); and the Australian labour market investigated by Linsley (2005). Linsley (2005)’s study used data from the 1997 wave of the Negotiating the Life Course (NLC) survey. It’s conclusion was disputed by Miller (2007), who argued that Linsley’s results come from a small sample size which limits the power of the tests undertaken, and he suggested that alternative datasets should be used to test career mobility theory in Australia. Our analysis provides new information with a more recent and larger data set, and econometric methods to address individual heterogeneity. This approach extends the analyses that focus on one education group (e.g. university graduates) to capture mismatches for the wider labour force. In this research, we also expand the Mavromaras et al. (2010, 2013) sample from graduates to the entire range of working-age employees.

³ This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the authors and should not be attributed to either FaHCSIA or the Melbourne Institute.

⁴ For example, in 2010 some questions were not asked in anticipation of some changes made in 2011. In addition, we had already tested the 2001–2009 data set for suitability and robustness in previous data suitability tests. Therefore, we decided to conduct the analysis with the 2001–2009 cohort as a sufficiently long period for the current analysis.

⁵ Because we focus on the comparison between workers who stay in their current job and workers who leave voluntarily, we exclude workers who leave involuntarily or leave for other reasons are not classified as voluntary quit.

⁶ In addition, self-employed workers and full-time or part-time students are excluded.

⁷ To avoid potential selection biases, we first tested the sample selection impacts for inclusion into the sample, and full-time employment, in terms of unobserved circumstances such as having young children, etc., or unobserved personal characteristics by employing a Heckman (1979) selection adjustment.

Following Cutillo and Di Pietro (2006) and Colin and Heywood (2007), we further applied a double selection probit model to estimate the incidence of over-education for full-time male workers. When sample selection correction for full-time employment was applied to our model, the ‘invers of mills ratio’ coefficient was very insignificant and the coefficients of interest were unchanged.

We also used an integrated approach by Wooldridge’s (2005) and the Mundlak (1978), which controls for the initial condition and unobserved effects – worker’s ability, quality, etc. The results were further not sensitive to the sample selection.

⁸ McGuinness and Wooden (2009) cross-tabulated the over-skilling variable with a measure of job complexity (responses to ‘my job is complex or difficult’, which was scored on the same 7-point scale used to measure over-skilling) to confirm that the more over-skilled the worker, the less difficult they consider their job to be. They further refined that either the severely (individuals with responses of 1, 2, or 3 on this scale) or moderately over-skilled (those with responses of 4 or 5) must not report high levels (a score of more than 5) of job complexity. They claimed that this association showed that the over-skilling responses would not be biased by respondents having incorporated non-labour-market-relevant skills and abilities into these responses.

⁹ Quit (voluntary leaving): (1) not satisfied with job, (2) to obtain a better job/just wanted a change/to start a new business, (3) retired/did not want to work any longer, (4) to study at home to look after children, house or someone else, (5) travel/have a holiday, (6) study/needed more time for study, (7) too much travel time/too far from public transport, (8) change of lifestyle, or (9) immigration. Involuntary leaving: (10) laid off, (11) no work available, (12) retrenched, or (13) redundant.

¹⁰ This variable reports the completion of job-related training that enhances skills for the current or future jobs. In our data set, the cost of training was covered by the employer for 76% of the respondents who received training.

¹¹ We use a likelihood ratio test to determine the better model between the DREP-MC and DREP. The statistic of the likelihood ratio test is$LR=2(ln(L_{DREP-MC})-ln(L_{DREP}))\text{~}{\chi }^2(m)$, where $ln(L_{DREP-MC})$ is the natural logarithm of the model’s likelihood for the less restrictive DREP-MC model, and $ln(L_{DREP})$is for the more restrictive DREP model. The statistic LR is distributed chi-squared with m degrees of freedom (i.e. the number of variables added to the model). The test statistics LR exceed the critical values when comparing between the DREP-MC and DREP models for Table 4, indicating that the null hypothesis of the DREP model is rejected, and that the DREP-MC model should be adopted for addressing the endogeneity issue.

¹² These results are available from the authors on request.