Problems with the measurement of skill
HIGH INTERMEDIATE
3.3. Measuring skill
3.3.1. Proxy indicators
While good data have been available for many years to support the former kind of proxy, it begs many of the questions that need to be addressed when analysing patterns of skills demand and deployment. The most important of these concern issues of credentialism - whether some employers are imposing unnecessarily high qualifications requirements as a screening device – and overskilling, i.e. whether employees are required or enabled by their work to use all the skills they have been trained to exercise. On the supply side, a simple equation of skill with qualification also obscures the extent to which some employees develop the skills required to do their job without recourse to the formal qualifications system, or need to do substantial additional learning on top of their formal qualification in order to develop skills that can be practically applied in the workplace. Consequently, it sheds no light on the role of organisational learning or the generation of new knowledge in the production process. Perhaps most importantly for the present analysis, reliance on this proxy effectively locks the researcher into an alignment model and excludes any evidence on the contribution of skill deepening.
International comparisons on this basis may also be misleading because of significant inter-country differences in the skill content of qualifications at the same ostensible level in the same field. This applies with equal force to years of education, given such factors as different national structures of education and different national approaches to the allocation of formal learning between classroom-based and work-based components (e.g.
apprenticeship vs. secondary vocational schools).
A related approach which has been commonly taken since the 1990s – more often by implication than as a basis for formal analysis – is to associate training activity with skill: industries and firms which do a lot of formal training are assumed to be high-skill, and low- training ones to have low skill requirements. This implicit proxy has been common in discussion about skill ever since reliable data on training activity began to appear in the late 1980s and early 90s. It is intuitively credible as an argument, and indeed has lain at the core of the Low Skill Equilibrium argument since the original Finegold/Soskice article in 1988. As a proxy for skill, however, it is inappropriate even for that type of analysis, since the function of such analysis is precisely to establish whether low skill requirements are in fact the reason behind low training activity, meaning that a different indicator is needed for skills exercised. More generally, a different indicator is needed to permit analysis of any question of match between the level of training provision and the amount of skill required, most obviously in the case of skill shortages.
The practical difficulty with using this kind of proxy is that it is far easier to quantify formal, structured training than informal training. Very few official surveys of firm
of Training and Education (ABS 1993) was one of the few that did try to gather data about “on-the-job training”, albeit with the question worded in a way that picked up kinds of informal learning (e.g. teaching self, asking questions, watching other workers) which do not fall within the intuitive definition of training. However, even it could only ask reliable questions about whether a respondent had received such “training” and if so, which of four categories had been “most important” (Fraser 1996: 48). Since the proportion of employed respondents to the three surveys so far conducted in this series who reported such “on-the- job training” ranged from 65.5% to 70.7%, as opposed to between 29.8% and 44.6% who reported and could quantify their formal training, it is clear that omitting this category substantially understates the incidence of skilling activity in the workplace.
While some researchers in recent years have endeavoured to quantify and/or cost this missing element (Richardson 2004; Freyens 2006), they have only been able to reach an estimate by working backwards from firm-level data on turnover and labour costs. This equates broadly to the second type of proxy identified by Spenner in his 1990 article. The problem with the wage-effect proxy is that it involves arbitrarily ascribing a residual (usually in the guise of “unobserved heterogeneity”) whose causes are unknown and unidentifiable by econometric analysis to individual productive potential, which the researchers equate just as arbitrarily to skill. It also involves making many assumptions about the mechanism linking wage rates to individual productivity which cannot be empirically grounded and are often contradicted by common observation, e.g. about the speed with which wage rates adjust to changes in individual worker productivity and the degree of variation that exists between the wages of individual workers carrying out the same kind of job in the same firm. Above all, this proxy requires adjustment to compensate for variations in supply and demand; otherwise, as Form points out, horse teamsters and file clerks would need to be treated as unskilled simply because there is no longer any demand for their skills (1987: 31).
Felstead et al (2007: 3) identify three other kinds of proxy: proportion of the workforce in occupations classified reputationally as high-skilled, scores on literacy and numeracy tests, and workers’ self-assessment of their own skill levels. As noted earlier, Spenner treats the first as “non-measurement”. The earlier discussion on social construction illustrates why this sort of proxy creates almost identical problems of validity to those created by the qualifications proxy; but even if the initial ranking is empirically based, the skill content of occupations evolves over time, so that the equivalence needs to be constantly reviewed, resulting in an unstable metric (Felstead et al 2007: 7). International literacy surveys such as PISA and IALS are progressively expanding their scope to cover generic work-related cognitive competencies which provide an increasingly rich picture of some aspects of the capability of each nation’s workforce. However, such measures refer only to the embodied and potential aspects of skill and cast no direct light on how much of this capability is actually deployed for productive purposes. Self-report is perhaps better seen as a research method than a metric in its own right, and will be further discussed in that context below. In practice it is often difficult to avoid the use of such proxies altogether if there is a need to provide a full picture of, say, the skilfulness of an industry at a point in time, simply
because the data available in most countries are insufficient to cover all the dimensions of interest. The important things when it becomes necessary to resort to any such proxy are to use it in ways and for purposes which minimise its known potential to bring about
cancel out the error or at the least, make it more transparent; and if at all possible, to use proxies only as a supplement to a direct measure of at least one key dimension of skill which has been designed to capture the real construct of interest. This is the approach which has been taken in Chapter 8 of this thesis.