Basic skills and workplace learning: what do we actually know about their benefits?

Abstract

In this paper we review the literature on the impact of workplace basic skills training on individuals, as measured by their effects on wages and employment probability. In addition, we also examine studies on the returns to individuals of general training at the workplace. On the whole, the evidence suggests that better numeracy and literacy skills have a strong positive effect on individuals' earnings and employment stability, even when other relevant factors, such as qualifications levels, are taken into account. There is also good evidence to suggest that general training provided at the workplace has a positive impact on individuals' wages, particularly when this training is employer provided rather than off the job. However, the literature also suggests that improvement of basic skills levels in adults has very small or even no positive effects on wages and employment probability. We discuss the implications of these findings on the formation of government policy on basic skills provision. We also propose that there is a real need for more research in this area, not only in terms of longitudinal quantitative studies tracking the effects of basic skills programmes on firms and individuals but also in terms of detailed case studies focusing on specific training programmes and their impact at the level of the individual and firm.

Notes

Corresponding author. Bedford Group for Lifecourse and Statistical Studies, Institute of Education, 20 Bedford Way, London WC1H 0AL, UK. Email: [email protected]

See especially Rainbird et al. (2004); Evans et al. (2002); Billett (2001).

For the purposes of the analyses reported here, Dearden et al. have equated the IALS results to the NCDS ones, and through these to the current national levels in the United Kingdom. This was done in a purely practical way by the researchers: that is, they have set out to preserve the same basic distribution of attainment levels in the population. Levels here therefore refer to the five current national ones in the United Kingdom: Entry 1–3, Level 1 and Level 2. As an indication of the standards these refer to, Level 1 in literacy and numeracy corresponds to the reading and mathematics level expected from 11 year olds in the school National Curriculum; for full details see QCA (2000).

The population is partitioned differently for these two comparisons, since only about a fifth of the population, based on NCDS estimates, has literacy skills below Level 1, compared to about a half for numeracy.

For the NCDS, controls were family background (i.e. parents' educational level, social class, financial difficulties in family when child aged seven) plus various childhood variables (school type, parental interest in education) and ability at age seven. For IALS data, a more limited set of background variables was available/used.

These differences may be partly explained by the fact that the IALS covers the entire age range (16–64) and therefore has a lower aggregate level of employment than the NCDS sample, which is confined to a particular one‐year cohort.

• Raw employment effects associated with literacy and numeracy skills, by gender (Dearden et al., 2000)

  Raw wage effects associated with numeracy and literacy skills, by gender

The results reported are for analyses with controls for family background, parental interest, schooling, age 7 and 16 attainment, qualifications and ‘soft’ skills.

Including educational qualifications obtained up to age 33, early attainment scores on maths and reading tests at age 11, family background (such as social class and educational level of parents), type of school attended, and characteristics of the job in 2000 (public/private sector, size of organization, union membership). Potential endogeneity bias was allowed for by running equations in first differences as well as in levels.

Both the Jenkins and the Machin studies used data from the UK birth cohort studies.

Estimating the returns to training is complicated by the fact that the recipients of training may differ from non‐recipients with respect to unobserved variables. A standard method for dealing with this issue is to use a model in first differences, i.e. to focus on wage growth rather than wage levels (just as, in looking at the effects of basic skills acquisition, it is preferable to look at the effects of changes in skill). This is essentially what Blundell et al. do here, although they use a more general technique which they refer to as quasi‐differencing. Heckman selectivity corrections were used to allow for the endogeneity of the initial wage, as well as employment status and occupation.

The researchers also show that the skills enhancement which training delivers tends to depreciate over time. Returns to training courses taken before 1989 were substantially lower than those recent courses taken in 1989–1991.

Controls=ethnic background, early test scores (maths and reading at age 11), trade union membership, sector/firm type, marital status, regional unemployment rate, qualifications by 1981. They also controlled for endogeneity bias using a probit for selection into education/training.

For the current job, data were provided on several sorts of training including on‐the‐job training, courses within the organization, courses outside the organization, and the number of days of training. There was also information on training in their previous job and in earlier jobs. The sample consisted of approximately 2300 men and 1300 women. The potential endogeneity of training was dealt with both by using the Heckman two‐stage procedure and also by estimating equations in first differences (earnings growth).

All these estimates controlled for a range of individual characteristics including age, marital status, number of children, educational background, months unemployed and months out of the labour force, current job tenure as well as type of university attended, class of degree and further education since 1980.

For each country the logarithm of the gross hourly wage was regressed on explanatory variables including age, gender, hours of work, nature of contract, organization size, public‐sector employment status and industrial sector. Samples were confined to employees aged 25–54 years. The endogeneity of training was controlled for through Heckman's two‐stage procedure (i.e. including among the explanatory variables a term summarizing information obtained from a separate probit estimate of the likelihood of receiving training).

He used data from the 1983 Dutch Brabant Survey, which contained information on individuals who were in the sixth grade of primary school in 1952 (i.e. were then aged 11 or 12) and who were re‐interviewed in 1983. In 1952, data on social background and IQ were gathered; in 1983 the questions covered education after primary school, current job status, earnings, and vocational training. Groot used a sample of 1075 individuals who were wage earners in 1983. Wage regressions were estimated by both OLS and using a general switching regression model, designed to allow for self‐selection into certain types of training.

Clearly the disadvantage of the approach is the potential difficulty of knowing how far results can be generalized, but there are some advantages. One is that the training data should be more precise because they have been obtained from documentary records rather than the hazy recall of individuals responding to surveys. Secondly, there is no bias from differences in training varying across diverse firms. Bartel estimated a wage growth equation to eliminate person‐specific fixed effects; she used an incidence of training equation to control for the selection of certain types of individual into training programmes.

Booth and Satchell modelled the length of time (in months) until the respondents' first job came to an end using an event history model with competing risks. The data came from the fourth follow‐up of the NCDS sample that occurred in 1981 when cohort members were 23 years old. The competing‐risks framework was considered appropriate because a person's first job can end in several different ways: a voluntary quit into another job, a voluntary quit into unemployment, and involuntary termination of employment. Controls were for type of school attended, for test scores in maths and comprehension at age 16, for sector of employment (public or private) and for whether the person had a disability.

The data came from the 1986–1990 Social Change and Economic Life Initiative (SCELI), a major research project conducted within six local labour markets in Britain (Aberdeen, Coventry, Kirkcaldy, Northampton, Rochdale and Swindon). In each locality a random sample of about 1000 people aged 20–60 were interviewed about their work histories. Elias used data from one of the localities—Rochdale—only and modelled the probability of job termination on monthly work history data. Job termination in a person's work history involved a change of employer or a transition to a non‐employment state. The sample consisted of 171 males and 258 females. Controls included job tenure, the age of the respondent, part‐time work, trade union membership, large employer (500+ employees), and working in a managerial, technical or professional job.

Several different approaches were used. The main results were obtained from what they characterize as a ‘before and after’ approach in which the impact of training in an earlier period on the probability of moving jobs in the following year was assessed. More elaborate models were also tried, such as instrumental variables (to allow for situations in which the unobservable individual characteristics that explain likelihood of moving jobs are correlated with previous training), sequential models and models in which training and mobility were determined simultaneously.