The persistence of overskilling and its effects on wages

Overskilling is the phenomenon whereby a worker’s skills are underutilised in his or her job. Overskilled workers are employed, but they are underutilised and mismatched, in that their skills and abilities are not a good match with the requirements of the job. Overskilling can lead to decreased wages and job satisfaction, which suggests that the investment in skills for that individual has been somewhat wasted.

Overskilling mismatch has been shown to be persistent; that is, present overskilling mismatch increases the probability of future overskilling mismatch. However, the previous research showing this extends back only one year. This report examines the persistence of mismatch over a longer (up to three years) time period and its effect on wages.

An obvious explanation for the persistence of overskilling is that it reflects personal unobserved characteristics (such as the person having an inflated view of their own skills). This paper exploits longitudinal data to show that persistence is more than this, with the probability of being overskilled increasing if the individual has been overskilled in the previous period, after allowing for unobserved characteristics.

Key findings

• Overskilling is persistent: overskilling mismatch is common among those who have been overskilled in the past. Persistence varies by educational level, with its being lowest among university graduates and highest among VET diploma graduates and those who did not finish high school.

• The wages of university graduates are reduced by past overskilling, more so than for any other education level.

A possible reason for the second finding is that graduates tend to be in better-paid jobs and therefore there is more at stake for them. This observation is supported by the results of quantile regressions, which differentiate the impact of overskilling by whether an individual is at the top or the bottom of the earnings distribution. With the exception of certificate III and IV graduates, workers who are better paid among their peers are more likely to suffer higher wage penalties from being overskilled.

Readers may be interested in looking at earlier research reports on overskilling: The incidence and wage effects of overskilling among employed VET graduates available at this page and Over-skilling and job satisfaction in the Australian labour force available at this page .

Tom Karmel

Managing Director, NCVER

The purpose of this research is to examine the longer-run outcomes of employee overskilling as a form of labour market mismatch. The research focuses on full-time and part-time employees in Australia between the years 2001 and 2009. Overskilling mismatch has been shown in the literature to cause losses in wages and in job satisfaction, both of which provide direct and indirect indications of reduced productivity in Australian workplaces. Overskilling mismatch occurs when someone is in paid employment, but where their skills and abilities are not fully utilised. Overskilling mismatch and other forms of on-the-job mismatch extend the conventional (job search) concept of mismatch, where workers take time to find new jobs or are between jobs (often referred to as frictional unemployment or search unemployment). The two phenomena are clearly related, in that they can both be the manifestation of underutilisation of national human capital. Overskilled workers are employed, but are underutilised and mismatched, in that their skills and abilities are not a good match with the requirements of the job.

This project builds on several relevant recent research findings on overskilling mismatch. The first such finding is that most adverse labour market outcomes stem from overskilling mismatch, where general skills and abilities are underutilised, and not from over-education, where formal qualifications are underutilised. Hence we focus on overskilling mismatch. The second such finding is that overskilling has been shown to be self-persistent; that is, present overskilling mismatch begets future overskilling mismatch. However, existing results in the literature extend back one year only. The attribute of self-persistence is common among many adverse labour market outcomes, including the very similar phenomena of long-term unemployment and underemployment. The third relevant finding that we build on is that much evidence about mismatch suggests that the workings of mismatch in the labour market are related to the educational level of the worker. This is a theme that has been running through our stream of mismatch research: we find that the way human capital is fully or less than fully utilised is intimately related to the formal qualifications of the workers concerned.

We carried out multivariate panel regression analysis using the first nine waves of the Household, Income and Labour Dynamics in Australia (HILDA) Survey data to estimate the causal effect of past overskilling mismatch on present overskilling mismatch, and the causal effect of having been overskilled in the past on present wages. Our analysis confirms that there are extensive differences in educational level in the effects of past mismatch on present mismatch and on present wages. We carried out several cross-section estimations as part of our usual robustness investigation procedures; however, we depart from our past practice of reporting the comparison between cross-section and panel evidence on mismatch, as we consider that enough is now known on the subject of controlling for unobserved heterogeneity through panel estimation. Up-to-date applied economic and econometric evidence suggests that we should only be looking at cross-section results when there is a specific reason for doing so, recognising that they will not provide causal evidence and that they will typically contain biases generated by cross-section estimation.

We confirm previous results on self-persistence of overskilling mismatch and extend the analysis to incorporate the effect of overskilling that occurred up to five years in the past on present overskilling and present wages. As there is no such research in the national or international literature, we have been cautious and have experimented with many different models to discover the degree to which past overskilling shows self-persistence. We have found evidence that overskilling can be self-persistent for at least five years, but at the same time we have found evidence that the model that must be used for estimating self-persistence reaches its limits when we build five lags in the model specification and attempt to make estimations using a longitudinal dataset that covers nine waves. We therefore settle for the less ambitious but definitely more robust model which incorporates the overskilling that happened up to three years ago.

We use the multivariate regression results to predict the over-time effect of self-persistent overskilling mismatch on future overskilling by educational level. The comparisons we report are between people who differ only in terms of their overskilling mismatch self-persistence, with all their other characteristics set at the mean levels for people with the same educational level. It is important to note that this is an ‘other things equal’ comparison, which allows us to generalise the results to the population and which can only be achieved using multivariate regression. Reported results are accompanied by the level of their statistical significance.

We find that self-persistence in overskilling mismatch is very large among those who have been overskilled in the past and differs a great deal by educational level. A university graduate who was overskilled in all three past years has a 38% chance of being overskilled in the coming year. An equivalent university graduate who was not overskilled in any of the last three years has a chance of only 4.6% of becoming overskilled in the coming year! Note that this compares two university graduates with the same average graduate characteristics, their only difference being their past overskilling status. The difference of 33.4 percentage points is large. Overskilling mismatch self-persistence for workers who were overskilled in all three past years (and with their well-matched counterparts in brackets) has been estimated to be 59.8 (6.3)% for diploma vocational education and training (VET) graduates, 60.2 (8.4)% for certificate III and IV VET graduates, 56.3 (11)% for Year 12 school graduates, and 66.1 (12.9)% for those who did not complete school. These numbers show that overskilling mismatch is not only a self-perpetuating phenomenon, but it is also a phenomenon with a very strong ‘labour market memory’. The next step of the research was to establish the damage that overskilling mismatch inflicts on those who are persistently overskilled.

Previous research suggested strongly that the effect of present overskilling mismatch on present wages can be considerable and will typically hurt university graduates. VET graduates have on most occasions not been found to suffer a strong overskilling wage penalty. The evidence on Year 12 and less than Year 12 school graduates is both weak and mixed, principally because the wage distribution for these groups is more compressed, so that if there are to be any wage penalties, they will have to be small. The effect of past overskilling mismatch on present wages is estimated by education level. We find that the wages of university graduates are reduced by past overskilling. There is some mixed and weak evidence for diploma holders. The wages of certificate III and IV VET graduates are not influenced by past overskilling. There is a mixed picture for workers without post-school qualifications, with no discernible patterns. The only unambiguous result is therefore that the present wages of university graduates suffer from past overskilling, that this effect lasts for at least three years, and that it shows no sign of diminishing in strength over time.

In order to refine the wage results, we apply the method of quantile regression to estimate the effect of past overskilling on wages at different points of the wage distribution. Quantile regression allows us to examine whether the overskilling penalty is primarily borne by the best-paid- or the worst-paid. Especially regarding university graduates, we want to know if it is the better or the worse-paying graduate jobs that suffer more from overskilling penalties. Quantile regression estimates produce a clear result: they show that the university graduates who work in the better-paid jobs are the ones who suffer the highest overskilling self-persistence wage penalties. Quantile regression results also shed new light on the effect of overskilling on the wages of diploma VET graduates. By differentiating between the better- and worse-paying jobs, we see that workers at the top of the (diploma VET graduates) wage distribution appear to suffer overskilling wage penalties similar to those of university graduates, while those in the middle and the bottom of the distribution do not suffer any such overskilling wage penalties. Certificates III and IV VET graduates do not suffer any overskilling wage penalties. Finally, there is some evidence that among those with no post-school qualifications the better-paid suffer some discernible overskilling wage penalties. Given that workers without post-school education operate within a compressed wage distribution (especially from below through the minimum wage), this is a strong result.

The incorporation of quantile regression in the analysis allows a more general picture to emerge for all education levels, with the exception of certificates III and IV. We now find that, to a varying degree, workers who are better paid among their peers and become persistently overskilled are more likely to suffer a higher overskilling wage penalty. Although the effect is principally concentrated among the better-paid university graduates, it is worth noting that it is also present among the best-paid workers with no post-school qualifications. Quantile regression results offer further confirmation of the previously established evidence that the wages of certificates III and IV VET graduates are not influenced by the self-persistence of their overskilling mismatch.

The research suggests that overskilling can impose real costs on individuals, employers and the economy. To the degree that an overskilling wage penalty may reflect a productivity loss, its state persistence is shown to be strong; the associated wage penalties suggest that the implied productivity losses can also be large. The similarities with other forms of human capital underutilisation can be informative. In the same way that we cannot know whether frictional unemployment is at an optimal level, we cannot know whether skills underutilisation resulting from overskilling mismatch is at an optimal level. If we find that changing the circumstances surrounding underutilisation turns out to be more expensive than the losses (or foregone benefits) resulting from underutilisation itself, it would be sub-optimal to argue for policy intervention. As part of the discussion of the estimation results, we provide some calculations of the benefits that could result if we reduced the underutilisation resulting from overskilling mismatch. These calculations must be viewed as partial equilibrium results and, accordingly, treated with the necessary caution that partial equilibrium analysis warrants. Notwithstanding this caveat, given that the results we present here have been derived using a reliable nationally representative dataset and that they are based on robust econometric longitudinal analysis, they provide the best available evidence on this subject. The large estimates of the national cost of overskilling mismatch resulting from the estimated wage penalties indicate that the economy-wide losses due to overskilling should not be ignored.