Human capital formation and changes in low pay persistence

ABSTRACT

This study presents new empirical evidence on the role of time trends in low pay persistence. We utilize population-wide tax records to track monthly labour market trajectories of initially low-paid workers. By performing age- and qualification-specific regressions, we find that low pay persistence reduces with time. However, the magnitude is highly heterogeneous across workforce characteristics. For a qualified worker in their early 20s, the risk of staying on low-pay declines by, on average, 5–10% points after one year. For a worker in their 50s, persistence remains almost unchanged regardless of their qualification level. We conclude that policy initiatives need to be more nuanced than a simple one-size-fits-all approach by accounting for time trends in low-pay persistence.

KEYWORDS:

• Low pay
• human capital formation
• state dependence
• random-effects probit
• initial conditions

JEL CLASSIFICATION:

• J63
• J61
• J24

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ For example, Metcalf (1999, p. F49) noted: the ‘incidence of low pay is far higher among 18 to 20 year olds (…) than those aged 21+’.

² For further details please visit https://www.stats.govt.nz/integrated-data/integrated-data-infrastructure/and see the Disclaimer in the Appendix.

³ An employer-identifier enables us to determine the month-specific number of employees of an employer. To exclude self-employed, we drop individuals who received wages and salaries for a minimum of one month from an employer with only one employee on his payroll.

⁴ The relevance of the qualification level is not age-independent. Due to the lack of prolonged labour market experience, the qualification level might be of higher relevance for young workers and for older workers, the occupation might play a more important role. To have a consistent category across all age groups, we chose the highest qualification level as a proxy for human capital.

⁵ Some individuals either hold multiple jobs per month or transit between two jobs in a month.

⁶ There are several studies investigating the labour market prospects of low-paid women. However, all of these studies put particular emphasis on accounting for hours worked. For example, Mosthaf, Schank, and Schnabel (2014) differentiate between high-wage employment with 30 working hours or more, high-wage employment with less than 30 working hours, low-wage employment with 30 working hours or more, low-wage employment with less than 30 working hours; Fok, Scutella, and Wilkins (2015) define a person to be low-paid if their hourly rate of pay is less than 120% of the Australian hourly federal minimum wage (FMW) and their weekly earnings are also less than 120% of the weekly FMW; Knabe and Plum (2013) identify low-paid employment by calculating an hourly threshold wage; buddelmeyer2010low’s Buddelmeyer’s, Lee, and Wooden (2010) low-pay and high-pay indicators are based on a measure of the hourly wage in the main job. Mosthaf, Schank, and Schnabel (2014) exclude women from the analysis because the search intensity of non-employed women cannot be observed. Moreover, the author uses administrative data from the German Integrated Employment Biographies Sample (IEBS) and focuses on full-time jobs because working hours are only crudely measured.

⁷ A standard approach in the economic literature is to include time-varying covariates and to add their time-means (Mundlak 1978; Chamberlain 1984). This is not possible in our study as we only include variables that refer to the 2013 Census and labour market performance in 2012. However, we expect that some of the unobserved heterogeneity is captured by running age- and qualification-specific regressions.

⁸ A limitation of the model is the auxiliary distribution assumption on the distribution of the random-effects error term. Stewart (2007) has tested the robustness of his findings by applying a dynamic linear probability (DLP) model by using a Arellano and Bond (1991) GMM estimator. One conclusion is that the average partial effects of the lagged labour market positions (in his study: low pay and unemployment) are higher than in the case of the random-effects models. Own simulation also pointed at the same direction.

⁹ For example, $\mathrm{\Phi }(x_1+a)-\mathrm{\Phi }(x_1)<\mathrm{\Phi }(x_2+a)-\mathrm{\Phi }(x_2)$ if $x_1>x_2>0$ and $a>0$.