Sector specificity of training

ABSTRACT

Human capital acquired through employer-provided training is pervasive in labour markets and an important determinant of post-schooling wage growth. The literature has focused on understanding under what conditions this training will lead to higher wages. Here I offer another aspect for consideration: Does the proximity of the current job to past jobs affect the return to training? In particular, I look at whether the wage return to training acquired at a previous employer depends upon the relative ‘proximity’ of the worker’s past employer to their current one, where proximity is measured by whether the worker received the training in the same industry or occupation as the current job. I investigate this using data from the British Household Panel Survey, with both fixed effects and instrumental variable estimation. The results suggest that both current and previous employer training spells are important determinants of wages, and both need to be included in wage regressions. Evidence further suggests that the return to formal training is highest if the training was received in a previous job that was in the same industry and occupation as the current one.

KEYWORDS:

• Training
• human capital
• wages
• occupational mobility

JEL CLASSIFICATION:

• J24
• J31
• J62

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/00036846.2023.2208334

Notes

¹ See Abraham and Farber (1987); Altonji and Shakotko (1987); Topel (1991); Altonji and Williams (2005); and Williams (2009).

² This decomposition is similar to one done for prior job experience suggested by Goldsmith and Veum (2002) and used by Williams (2009).

³ Hence I am excluding the supplemental European Common Household Panel (ECHP) of individuals that were surveyed between 1997 and 2001, and the Northern Ireland Extension Sample added in 2001. The ECHP sample was excluded because individuals were included for only a short period of time and because it included an oversample of ‘low-income’ households unlike the remainder of the data. The Northern Ireland sample was excluded because its coding of occupation differed from the other samples, and a consistent coding across the samples is required to be able to create measures of occupational experience and training.

⁴ Using only 1998–2000 data Booth and Bryan (2005) report results where they limit training to current employer training that was paid for by the employer or that was reported to have no fees. The obtain very similar training incidence levels to what is reported below.

⁵ When examined separately, training incidence and intensity for individuals and jobs in the Scotland and Wales extension samples is slightly smaller than in the original BHPS sample. This is unsurprising given the shorter number of years available in these extension samples.

⁶ There are a few individuals in the original BHPS sample who reside in Scotland or Wales, and some individuals in the Scotland extension sample that live in England, etc. Descriptive statistics and the subsequent analysis are based upon region of residence when the survey was taken and not based upon sample.

⁷ All results were obtained using the Stata packages reghdfe and ivreghdfe that run linear and instrumental variable regressions with many levels of fixed effects, by implementing the estimator described in Correia (2017).

⁸ The literature evaluating the effects of tenure and experience on wages has demonstrated that higher order polynomials or even non-parametric specifications are required to fully capture their impact on wages. See for example Topel (1991) and Altonji and Williams (2005). Since my focus is on estimating impact of formal training, I only used a quadratic in these measures.

⁹ The first stage F-tests for this estimator all exceed 100, and the Kleinbergen-Paap Wald rk F statistic is 42.6 suggesting that the wage equation is weakly identified.

¹⁰ In the sample that includes both part and full time workers, only 2.16% of the males reported working less than 30 hours, while 35.59% of females reported working less than 30 hours. Unsurprisingly, the hours distribution for females is bimodal when both part and full time workers on included, however the hours distribution (and the log wage distribution) have substantial overlap when only full time workers are used.