5. Decomposition of differences in the skills content of jobs
This section considers the underlying contribution to differences in the skills content of jobs between Scotland and the UK average arising from any differences in employment composition. In particular we examine the extent to which differences in the sectoral and occupational distribution of employment in Scotland as compared to the UK average may account for the differences in the skills content of jobs identified in the previous section.
It is certainly the case that there are considerable differences in the skills content of jobs between sectors and occupations. Tables 2a and 2b report the intensity rates for the broad skill measures by 1 digit industry ( SIC 1992) and 1 digit occupation ( SOC2000) respectively. Note that the industrial classification omits five Industries (namely: fishing; mining and quarrying; electricity, gas and water supply; private households with employed persons; extra-territorial organisations and bodies) because they had less than 100 sample observations across the whole. Differences in skills across industry appear sensible - education and health have long training and learning times as well as high qualification requirements. Respondents in the construction industry reported the longest learning time. At the other end of the scale, the hotel and restaurant sector has the lowest levels of skills content across all three dimensions of broad skills. Similarly, the observed differences in broad skills by occupation confirm our priors. Training time, learning time and required qualifications are high for professionals, and low for sales and, especially, elementary occupation workers. Skilled trades have long learning times - consistent with the sectoral evidence for construction which dominates this occupational category.
Differences in the intensity of the four measures of computing skills by sector and occupation are reported in Tables 3a and 3b respectively. Computer use across all four measures is highest in financial services and lowest in construction and hotels and restaurants in particular. These patterns are partly reflected in the occupational distribution of computer skills, although the most advanced usage is to be found amongst professionals, as expected, whereas administrative and clerical occupational groups have the greatest penetration.
Tables 4a and 4b present the differences in the intensity of generic skills by industry and occupation. Across all 13 measures of generic skills, there is wide variation by both sector and occupation. General patterns are consistent with more standard measures of skills - such as qualifications - and this gives support to the validity of the measure of generic skills derived from the skills surveys. Similar patterns are apparent in Tables 5a and 5b which summarise the high level generic skills indices by industry and occupation.
When we examine a particular skill by industry, a country may exhibit an aggregate of that skill above the UK national average for two distinct reasons. First, it may not differ much from the UK average for each industry, but the country may be specialised in high skill sectors. Secondly, most or all sectors in the country may have high skill jobs above the UK industrial averages, perhaps reflecting country-specific factors such as differences in infrastructure, physical capital or skills availability. Of course, there may be a combination of these two factors in operation. In order to establish the importance of each factor - sectoral specialisation in high skill sectors or overall high skills jobs in a country - to the aggregate country differential, we can use a modified shift-share analysis to decompose each country's difference from the UK national average into that due to the industry mix of the country and that due to the country-specific utilisation differential. A similar argument can be made with respect to differences in skills by occupation and the occupational composition of employment.
As shown in Box 3, the methodology of shift-share analysis as originally proposed by Dunn (1960) can be extended to the decomposition of aggregate skill differentials between countries. A country's skill differential from the UK national average can thus be decomposed into three separate components:
- a composition component;
- a utilisation component; and
- an interaction component.
These three components then can be added together to yield the overall aggregate country skill differential.
The composition component measures the contribution to the country's skill differential that accrues from its specific sectoral or occupational composition, assuming that the overall skills content of jobs in each country is equal to the national average. The composition component is therefore that part of the skills content differential that is the consequence of a country being specialised in the most or least skill-intensive sectors or occupations. Thus the composition component is positive if the country is specialised in high skill sectors or occupations and/or de-specialised in low skill sectors.
The utilisation component is the contribution to the country's skill differential that arises from sectoral or occupational differences in the skills content of jobs between the country and the UK national average, assuming the country's sectoral and occupational composition matches the national picture. Hence the utilisation component is positive if the skills content of jobs in that country are above average in most or all sectors and occupations - that is, if most or all sectors or occupations are above their national averages in the particular country.
The interaction component is the contribution to the country skill differential that derives from a country being specialised, relative to the national average, in sectors or occupations with jobs which have high or low skill content. This component can be interpreted as an indicator of the specialisation in each country in allocating employment to the sectors or occupations in which it has comparative advantage in skill content. It also measures the covariance between the composition component and utilisation component.
As demonstrated in Box 3, the overall gap between a country's aggregate skill content level and the national average can be additively decomposed into these three components, such that:
In order to gauge the relative contribution of each component to overall differential, Esteban (2000) suggests computing the relative weight of the variances of each component in the overall variance in differentials. Thus the overall variance in the country differentials can be written as the sum of the variances of each component, plus an additional term capturing the covariances between the components. In addition, we supplement this decomposition by presenting the overall regional differential and its constituent components expressed in percentage terms, so that the relative contribution of each component can be more easily assessed when comparing between different skill measures.
Figure 7 graphically presents the decomposition by industry of the skill differentials for Scotland for all 33 of the skill measures. As can be seen, in general, the contribution of the industrial composition to the overall skill differential from the UK national average is small. While it generally serves to reinforce the utilisation effect, in every case, it represents a rather smaller share than the contribution from the utilisation component. This implies that it is not the industrial composition of employment per se in Scotland that results in the lower skills content of jobs shown in Section 4, but rather it is lower skills content across most or all sectors that results in the lower rate. Figure 8 restricts attention to just the statistically significant differentials identified in Section 4. These confirm the above conclusion that it is the lower skills content of jobs within industries, rather than the industrial composition, that results in Scotland having a significantly lower level of computing skills, literacy and numeracy skills content than the UK average. Table 6 presents the relative contributions of each component - industry composition, utilisation and the interaction - to the overall skill differential, and this serves to confirm the graphical presentation in Figures 7 and 8 that the overall differential in the skills content of jobs between Scotland and the UK average is dominated by the lower skills content in Scotland for most of the dimensions of skills, rather than because of the sectoral composition of employment in Scotland.
Figures 9 and 10 graphically present a similar decomposition but by occupation rather than industrial sector. A similar picture emerges. Most of the differential in the skills content between Scotland and the UK average is due to lower occupational skills content in Scotland, rather than the occupational composition of employment, and this is certainly true for the significantly negative differentials for all four dimensions of computing skills and high level literacy and numeracy skills. Differences in the occupational composition of employment contribute relatively little to the overall differentials - whether significant or not - as shown in Table 7.
Thus, we can conclude that the overall lower skills content of jobs in Scotland does not appear to be due to the sectoral or occupational distribution of employment in Scotland, but is primarily due to the lower skills content, particularly lower computing skill content, within both industrial and occupational groups in Scotland. For those skills which are significantly less utilised in jobs in Scotland, it is the lower skills content within industry and occupation that results in the average skills content being lower in Scotland than the UK average, rather than because of a predominance of employment in sectors or occupations which use low levels of those skills.
In order to gauge the contribution of each component to the overall pattern in differentials across all four countries of the UK, Tables 8 and 9 present the variance contributions of each component to the overall variation in differentials for each skill when decomposed across sectors (Table 8) and occupational groups (Table 9). As can be seen, most of the variation between countries in each skill is due to variation in the utilisation rather than the composition of employment by sector or industry. Thus the patterns observed for Scotland, that the utilisation component dominates the overall differences from the UK average, are common across all four countries of the UK. The summary averages at the bottom of Tables 8 and 9 reveal that the differences in utilisation within sectors and occupations between countries is as large as the overall variation in skills between countries. Thus, it would appear that the utilisation effect dominates in the overall differentials between countries.