ICT and the returns to schooling and job-specific experience.

Kirby, Simon ; Riley, Rebecca

We use the United Kingdom Labour Force Survey to estimate the returns to schooling and job-specific experience in sixteen different industry sectors, over the period 1994-2001. Next, assuming skill levels are fixed, we assess the marginal effect on these returns of the capital intensity of production and the ICT intensity of capital. Our results indicate that in the UK, over the period 1994-2001, the rising ICT intensity of capital was associated with a rise in the return to schooling, and a reduction in the return to job-specific experience.

Keywords: Skill-biased technical change; return to human capital; information technology; experience

JEL Classifications: J30; J31; O30

Introduction

In this paper we investigate empirically the relationship between information and communication technologies (ICT) and the return to two different types of human capital--schooling and experience accumulated on the job with a specific employer. The latter of these is likely to be less transferable across different jobs or technologies. A number of studies have found that technical change and innovation during the 1980s and 1990s has been biased towards skilled labour, resulting in a rise in the skill premium for a given skill composition of employment. (1) Typically these studies analyse trends in wage bill shares or relative wages, for different occupation or qualification groupings, and how they relate to measures of technical change, determining the shift in the relative demand for skills that can be attributed to technology. While the complementarity of physical capital and recent technologies to human capital, measured by education or occupations, is essentially a stylised fact (Goldin and Katz, 1998), the complementarity of these factors of production to other types or measures of human capital has received less attention.

The association between ICT and the returns to schooling versus employer- or job-specific experience is important from several perspectives. For example, it may help explain changes in the intergenerational distribution of labour market outcomes. Older generations typically have more job-specific experience than younger generations. If, for example, ICT was associated with a reduction in the return to job-specific experience relative to schooling, it should also be related to a reduction in the incomes of older in comparison to younger generations, all else being equal. Indirectly, this effect may be exacerbated as older cohorts would have relatively more to lose from re-skilling themselves. In addition, generations close to retirement age may have less incentive to undertake certain types of human capital investment than younger generations. For example, the loss of wage income whilst acquiring further schooling has a relatively large effect on pension income for these generations and the time over which the returns to such an investment can be realised is relatively short. Indeed, one of the striking features of the UK labour market in the 1980s and 1990s has been the rise in disability benefit claims and early retirement of old working age men (Disney, 1999; Nickell and Quintini, 2002). More immediately, in a study of French firms, Aubert et al. (2006) find that the use of new technologies and associated innovative workplace practices has been biased against old workers. In a study of US firms over the period 1992-7, Abowd et al. (2007) suggest that firms using advanced technologies are more likely to use high-ability workers, and less likely to use old experienced workers.

These associations between ICT and the returns to the different aspects of human capital examined here may also prove informative about the process of technology adoption. Several studies have emphasised the importance of the learning process in adapting to new technologies and the way in which this affects the adjustment of the aggregate economy to the arrival of new technology. For instance, Greenwood and Yorukoglu (1997) suggest that if technical change is associated with significant diversion of skilled resources to learning, advances in technology may be associated with an initial slowdown in productivity growth. If more experienced workers have less incentive to upgrade their skills because switching to the use of new technology devalues job-specific experience, any initial adverse effects on productivity growth associated with learning may be more muted and technology absorption might occur more gradually (Helpman and Rangel, 1999).

We use the UK Labour Force Survey from 1994-2001 to estimate the return to schooling and tenure with current employer, a measure of job-specific experience, allowing the returns to vary by industry and year. Next, we regress the estimated returns on industry-year measures of capital intensity and the ICT share of the capital stock, the latter being a measure of new technology. Our results suggest that the rise in ICT has been associated with a bias towards skills acquired through schooling, and a bias against job-specific experience. These findings are consistent with the interpretation that the recent skill biased technologies may have contributed to the obsolescence of some job-specific skills.

The returns to schooling and job-specific experience across industry and year

Specification of the empirical model

We estimate the returns to schooling and job-specific experience across industries and years, within a standard model of earnings, by allowing the coefficients on these measures of skill to vary accordingly. We use a standard Mincer-type earnings function, augmented with quadratic terms in both schooling and tenure to capture non-linear returns, (2) of the form

ln ([Y.sub.i] = [alpha] + [[beta].sub.s] [S.sub.i] + [[beta].sub.S2] [S.sup.2.sub.i] + [[beta].sub.T] [T.sub.i] + [[beta].sub.T2] [T.sub.i.sup.2] + [gamma] [X.sub.i] + [[epsilon].sub.i] (1)

where In [Y.sub.i] is the log hourly wage, deflated to 2000 prices, (3) [alpha] is a constant term, [S.sub.i] is years of schooling, [T.sub.i] is years of job-specific experience or tenure, [X.sub.i] is a vector of explanatory variables and [[epsilon].sub.i] is an error term for individual i. We extend this standard earnings function to allow the returns to schooling and job-specific experience to vary both by industry and year, denoted by subscripts j and t respectively. We include industry and year-specific dummy variables, denoted as [[eta].subjt], to control for any industry-year specific effects on earnings that may bias our estimates of the industry and year-specific return to schooling and tenure. This could include industry and time-specific demand or supply shocks or composition effects. Thus our model becomes:

ln ([Y.sub.i] = [alpha] + [[beta].sub.Sjt] [S.sub.ijt] + [[beta].sub.S2jt] [S.sup.2.sub.ijt] + [[beta].sub.Tjt] [T.sub.ijt] + [[beta].sub.T2jt] [T.sub.ijt.sup.2] + [[gamma].sub.[gamma]jt] [[eta].sub.jt] + [gamma] [X.sub.i] [[epsilon].sub.i] (2)

We estimate (2) using OLS assuming that the error term is normally distributed around zero. There has been much discussion in the literature about the use of instrumental variables to control for ability bias (see Card, 1999; Harmon, Oosterbeek and Walker, 2003; for a review of the literature). There is evidence from a number of countries that the use of instrumental variables to estimate the returns to education produces larger estimates than the simple OLS approach, in contrast to expectations. However, as discussed in Harmon, Oosterbeek and Walker (2003) there are problems with finding instruments that are not only uncorrelated with wages, but that are actually correlated with schooling. It has also been suggested that estimates using existing instrumental variables may be in themselves biased upwards, and the effect of measurement error and ability bias on OLS estimates of returns to education cancel themselves out (see Harmon, Hogan and Walker, 2003).

Data

To estimate (2) we use data from the United Kingdom Labour Force Survey (LFS) from 1994 to 2001. The LFS is a quarterly sample survey of approximately 61,000 households across the United Kingdom with a 5-quarter rolling panel design. (4) For the purposes of this paper we use information at the individual level; this translates to all adults within the household. The LFS has included questions on earnings since the fourth quarter of 1992, but we restrict our analysis to the period from 1994 onwards to avoid problems with different industrial classifications. (5) The ICT data, discussed below, is available to 2001. Hence, we exclude LFS data collected after this date. From 1992 to the end of 1996 the earnings questions were asked of respondents only in the fifth and final survey wave. Since 1997 the earnings questions have been asked in both the first and fifth waves of the survey, effectively doubling the quarterly sample for earnings data. We have restricted our sample to wave 5 respondents to avoid issues of differential attrition bias over our sample period.

Our sample comprises employees of working age who are not in full-time education who have responded with a positive value for earnings and hours worked. (6) We have restricted our sample to those whose hourly earnings were greater than or equal to 1 [pounds sterling] and less than or equal to 100 [pounds sterling] in 2000 prices. (7) Wilkinson (1998) suggests that there may be an element of error in answers by LFS proxy respondents. He suggests an adjustment procedure for the earnings data from the LFS depending on whether the proxy respondent is a spouse or non-spouse proxy respondent to correct for this error. We have applied these adjustments to our measure of hourly earnings.

The characteristics of our sample are reported in table 1. Table 2 shows the distribution of the sample across industries at the start and end years of the sample and on average for the sample as a whole. We use each quarter of the LFS to boost annual sample sizes and to maximise the industry detail. This restricts the available control variables somewhat. For example, we are unable to control for individuals' union membership status as this is asked of respondents only in the autumn quarter of the LFS.

Years of schooling are measured as years spent in continuous full-time education. We proxy job-specific experience by tenure, defined as continuous years served with the current employer. It is possible that the return to schooling may be overestimated in our model. Years spent in continuous full-time education may not fully capture the amount of time spent in education. For example, a respondent may have taken a gap year in-between finishing secondary education and beginning tertiary education. Our proxy for job-specific experience also suffers some measurement error. For example, an individual may have done the same job with different employers, in which case our proxy underestimates job-specific experience. Also, an individual may have worked in different jobs with the same employer, in which case our proxy overestimates job-specific experience.

These human capital measures are interacted with industry and year to obtain estimates of industry-year specific returns to schooling and tenure, allowing us to assess the relationship between the capital intensity of production and the ICT intensity of capital on the one hand and the returns to schooling and tenure on the other, discussed in the next section.

Results

Table 3 gives OLS estimates of the model in (2), where j = 1-16 and t = 1-8. We control for a quadratic in potential experience (8) that varies by both industry and year as in the case of schooling and job-specific experience. Thus, we relax the assumption of constant returns across industry and year for each aspect of human capital included in the model. We also include sex, birth cohort (through nine cohort dummy variables), quarter, region of residence, (9) size of the establishment where the individual works, cohabiting status, full-time status (defined as greater than or equal to 30 hours per week, excluding overtime). We have included separate dummy variables for each sex that control for private sector employment. There is econometric evidence that women suffer a pay penalty when working in the private sector, whereas for men there is a pay penalty for working in the public sector (Anderson et al., 2001).

The estimates in table 3 are generally significantly different from zero and have the expected signs. The reported t-statistics are calculated using robust standard errors. The positive sign on the coefficient for being male is significant, as is the pay penalty for women compared to the pay premium for men in the private sector. We find quite a considerable pay premium for employees resident in London and the South East regions, as expected. The premium to working full-time is also very significant, with full-time employees earning just under 15 per cent more per hour than those working part-time.

The coefficients on the industry-year dummy variables and the industry-year-specific coefficients on schooling, job-specific experience and potential experience (and the squared terms in these) are not reported in table 3. Instead we plot the estimated marginal return to one additional year of schooling and tenure or job-specific experience in figures 1 and 2 respectively, together with their 95 per cent confidence intervals. These are evaluated at industry-year sample means, [[bar.S].sub.it] 1/[N.sub.jt][summation.sub.i] [S.sub.ijt] [[bar.T].sub.jt] = 1/[N.sub.jt] [summation.sub.i] [T.sub.ijt], where [N.sub.jt] denotes the number of individuals employed in industry j at time t. The standard errors used to calculate confidence intervals for the marginal returns estimates take into account parameter uncertainty only, treating the industry-year sample means as given. Letting [[[bar.[omega]].sup.S.sub.jt] denote the estimated return to schooling in industry j at time t we have

[FIGURES 1-2 OMITTED]

[[[bar.[omega]].sup.S.sub.jt] = [[??].sub.Sjt] + 2 [[??].sub.S2jt] [[bar].sub.jt] (3)

var ([[[bar.[omega]].sup.S.sub.jt]) = var ([[??].sub.Sjt]) + 4 [[??].sup.2.sub.jt] var ([[??].sub.S2jt]) (4) +4 [[bar.S].sub.jt] cov ([[??].sub.Sjt], [[??].sub.S2jt])

where the [??] are the estimated coefficients, [beta], from (2). Similar expressions can be derived for the marginal returns to job-specific experience. The quadratic term in both schooling and tenure in (2) complicates the marginal returns expression in (3) and its variance in (4), but, F-tests suggest that both the squared terms in schooling and tenure should be included.

Figure 1 shows the return to schooling over the period 1994 to 2001 for each of the sixteen industries analysed. A line has been drawn in the figures at zero returns to schooling to highlight those estimates that were not significantly different from zero at the 95 per cent level. It is clear from these illustrations that all our estimates of the return to schooling were significant at the 95 per cent level and vary across industries. Overall, the estimated returns to schooling within each industry appear reasonably stable over the sample period as a whole. The lowest returns to schooling are concentrated in the hotels & restaurants, construction and the manufacture of basic metals, and machinery industries. The returns to schooling are greatest in the manufacture of chemicals and allied products, the manufacture of electrical and optical equipment and business services industries. The transport sector also has a high return to schooling relative to the other industries, although the return does decline somewhat towards the end of the 1990s. Our estimates of the return to schooling in the manufacture of electrical and optical equipment and in communications industries are rising over the sample period.

Figure 2 shows the return to job-specific experience over the period 1994 to 2001 for each of the sixteen industries analysed. It is clear that the return to an additional year of experience accumulated with the same employer is lower relative to an additional year of schooling. In figure 2 a horizontal line has been produced for zero returns to job-specific experience. Non-significant coefficients are apparent only in the hotels and restaurants industry, where the estimated return is not statistically different from zero in several of the sample years. The manufacturing of chemical and allied products industries exhibits the highest returns to job-specific experience, while the construction and transport manufacturing industries exhibit the lowest returns after the hotels and restaurants industry. The return to tenure appears to be declining over the sample period in some industries, most notably in the business services sector, the communications sector, and the manufacture of electrical and optical equipment. These are the same industries that exhibit either relatively high marginal returns to schooling and/or where the estimated return to schooling is rising over the sample period.

ICT and the returns to schooling and job-specific experience

Specification issues

To assess the potential effects of ICT on the return to schooling and job-specific experience we regress the industry and year-specific estimates of the return to schooling and tenure obtained from model (2) on measures of the capital intensity of production and the ICT intensity of capital. As specified in equation (3) the return to human capital depends on the level of human capital. To avoid the potential endogeneity problem that arises if human capital levels determine ICT adoption we regress on ICT intensity the [??] terms obtained by estimating equation (2), rather than the derived [bar.[omega]] terms in equation (3). Thus we estimate:

[[??].sub.jt] = [lambda]' [Z.sub.jt] + [u.sub.jt], (5)

where j=1-16 and t = 1-8, [[??].sub.'jt] = [[[??].sub.S], [[??].sub.S2], [[??].sub.T], [[??].sub.T2]].sub.jt], i.e. a vector of the schooling and tenure coefficients from the initial wage equation (2), [lambda]' is a (4xK) matrix of coefficients to be estimated, and Zjt is a K-dimensional vector of regressors. Regressors include capital and technology intensity measures that vary by industry and year (capital stock--output ratio and ICT capital--total capital stock ratio), industry dummy and year dummy variables. The last term in equation (5) is a vector of error terms [u'.sub.jt] = [[[u.sub.S], [u.sub.S2], [u.sub.T], [u.sub.T2]].sub.jt]. To correct for the heteroscedastic error structure that results from variation in industry size we weight industry observations within each year by the industry share of total employment as reported in table 2. Each year is given the same sample weight. We estimate this system using SUR, which, given the weighting procedure just described is more efficient than least squares.

With the specification of equation (5) the link at the margin between [Z.sub.k] and the return to schooling, given the level of schooling can be derived as:

[[??].sub.k,S] = [[??].sub.k,S] + 2 [bar.S] [[??].sub.k,S2] (6)

var([[??].sub.k,S]) = var([[??].sub.k,S]) + 4 [[bar.S].sup.2] var([[??].sub.k,S2]) +4 [bar.S] cov ([[??].sub.k,S], [[??].sub.k,S2])

where [[lambda].sub.k,S] ([[lambda].sub.k,S2]) denotes the coefficient on [Z.sub.k] in the equation for [[??].sub.S] ([[??].sub.S2]). The marginal effect of [Z.sub.k] on the return to job-specific experience in this model, given the level of job-specific experience, can be written similarly (replacing S with T throughout).

Data

We use the National Institute Sectoral Productivity (NISEC) dataset for measures of capital stock levels and UK National Statistics data for levels of gross value-added to construct capital stock--output ratios and ICT-total capital stock ratios for each of the sixteen industries and eight years in our sample. (10) The NISEC capital stock data are derived using National Statistics investment data, used to create National Statistics estimates of capital stocks by industry. However, the capital stock data from National Statistics do not include a separate measure of ICT capital. The NISEC data contain measures of ICT capital (computers, software and other ICT technology) constructed using asset specific depreciation rates. Non-ICT capital includes structures, vehicles and non-ICT equipment. The capital stock data in the NISEC dataset have been produced up to 2001 and are in constant 1995 volumes. Output data are obtained from the Blue Book, and we deflate these data to 1995 prices, to construct capital-output ratios.

Table 4 reports the change over our sample period and the sample mean of ICT intensity of capital (ICT share of the total capital stock) for each of the sixteen industries. ICT intensity varies across industries and has risen most markedly over the sample period in the business services sector, the communications sector, financial intermediation and the manufacture of electrical and optical equipment. The capital-output ratio varies mainly by industry and rather less over time.

Results

Estimates of the marginal effects of the capital intensity of production and the ICT intensity of capital on the returns to schooling and tenure, as given by equation (6), are reported in table 5. These are evaluated at sample mean values of human capital and are derived from the estimated model in equation (5) as described above. We show three separate models. All models include a constant term and control for the capital intensity of production. Model 2 includes industry dummies. Model 3 also includes year dummies. We have controlled for industry and year-specific effects in estimating equation (2), to correct for any bias in our estimates of the industry and year-specific returns to schooling and job-specific experience that could arise from industry and time-specific demand or supply shocks, or composition effects. Thus, one might argue that we do not need to include industry dummies and year dummies in estimating equation (5). However, the industry effects and time effects in equation (5) capture factors that directly affect the return to human capital, whereas the industry and time effects in equation (2) capture industry and time effects that affect earnings more generally. To illustrate the sensitivity of our findings we report the results of including industry dummies and industry and year dummies together.

The top half of table 5 shows our estimates of the partial derivative of schooling and tenure returns to the capital--output ratio, as well as the estimated difference between these. With the exception of one case, none of our models show a significant effect of the capital--output ratio on the returns to either schooling or tenure. The results from model 1, excluding industry and time dummies in the estimation of equation (5), point to some positive relationship between the return to tenure or job-specific experience and the capital intensity of production, albeit only significant at the 10 per cent level. Because the capital intensity of production varies primarily across industries, rather than time, it is perhaps not surprising that the inclusion of industry dummies in models 2 and 3 leaves us with a statistically insignificant relationship between the return to tenure and capital intensity.

All models suggest a positive and statistically significant relationship between the ICT intensity of capital and the return to schooling. Also, the models suggest a negative and statistically significant relationship between the ICT intensity of capital and the return to job-specific experience or tenure. The size of the impact varies across models, but the sign of the effect is robust to the inclusion of industry dummies, and the inclusion of industry and year dummies. The estimates imply that a 10 percentage point rise in the ratio of ICT capital to total capital is associated with a rise in schooling returns ranging from 0.45 percentage points to 0.93 percentage points. Similarly, these results imply that a 10 percentage point rise in the ratio of ICT capital to total capital is associated with a reduction in the returns to job-specific experience ranging from 0.37 percentage points to 0.07 percentage points. There is less variation across models in the magnitude of the relationship between ICT and the return to schooling measured relative to the return to job-specific experience.

Conclusions

We have attempted to provide more evidence on the nature of the relationship between new technologies such as ICT and human capital commonly discussed in the literature. Using pooled cross-sections of the UK LFS we have estimated the return to schooling and to job-specific experience (tenure). Our standard earnings function suggests the return to an extra year of schooling is greater relative to an extra year of job-specific experience. We find evidence of variations in the returns to these two skill measures across industries and the years of our sample. In particular, we find that those industries where the return to tenure appears to be declining over the sample period are the same industries where the return to schooling is relatively high and/or rising over the sample period.

Next, using data on capital stocks from the NISEC dataset we have been able to regress these skill returns on measures of capital and technology intensity. In line with the literature, we find evidence of ICT-skill complementarity, but only when skill is measured by schooling. Our results also indicate that ICT technologies are associated with a reduction in the return to job-specific skills, measured here as tenure with current employer. Thus it appears that the rising ICT intensity of capital has been associated with greater divergence in the premium paid to these two different components of human capital.

REFERENCES

Abowd, J., Haltiwanger, J., Lane, J., McKinney, K. and Sandusky, K. (2007), 'Technology and the demand for skill: an analysis of within and between firm differences', NBER Working Paper No. 13043.

Anderson, T., Forth, J., Metcalf, H. and Kirby, S. (2001), The Gender Pay Gap, Women and Equality Unit, Cabinet Office.

Aubert, P., Caroli, E. and Roger, M. (2006), 'New technologies, organisation and age: firm-level evidence', Economic Journal, 116, F73-93.

Berman, E., Bound, J. and Griliches, Z. (1994), 'Changes in the demand for skilled labor within U.S. manufacturing: evidence from the Annual Survey of Manufactures', Quarterly Journal of Economics, 109, pp. 367-97

Betts, J. (1997), 'The skill bias of technological change in Canadian manufacturing industries', Review of Economics and Statistics, 79, pp. 146-50.

Bound, J. and Johnson, G. (1992), 'Changes in the structure of wages in the 1980s: an evaluation of alternative explanations', American Economic Review, 82, pp. 371-91.

Card, D. (1999), 'The causal effect of education on earnings', in Ashenfelter, O. and Card, D. (eds), The Handbook of Labor Economics, Volume 3A, Oxford, Elsevier.

Cingano, F. (2003), 'Returns to specific skills in industrial districts', Labour Economics, 10, 2, pp. 149-64.

Disney, R. (1999), 'Why have older men stopped working', in Gregg, P. and Wadsworth, J. (eds), The State of Working Britain, Manchester University Press.

Goldin, C. and Katz, L. (1998), 'The origins of technology-skill', Quarterly Journal of Economics, 113, pp. 693-732.

Green, F., Felstead, A. and Gallie, D. (2003), 'Computers and the changing skill-intensity of jobs', Applied Economics, 35, pp. 1561-76.

Greenwood, J. and Yorukoglu, M. (1997), '1974', Carnegie-Rochester Conference Series on Public Policy, 46, pp. 49-95.

Harmon, C., Hogan, V. and Walker, I. (2003), 'Dispersion in the economic return to schooling', Labour economics, 10, 2, pp. 205-14.

Harmon, C., Oosterbeek, H. and Walker, I (2003), 'The returns to education: microeconomics', Journal of Economic Surveys, 17, 2, pp. 115-55.

Haskel, J. and Heden, Y. (1999), 'Computers and the demand for skilled labour: industry- and establishment-level panel evidence for the UK', Economic Journal, 109, pp. C68-79.

Haskel, J. and Slaughter, M. (2001), 'Trade, technology and UK wage inequality', Economic Journal, III, pp. 163-87.

--(2002), 'Does the sector bias of skill-biased technological change explain changing skill premia?', European Economic Review, 46, pp. 1757-83.

Heckman, J., Lochner, L. and Todd, P. (2003), 'Fifty years of Mincer earnings regressions', NBER Working Paper No. 9732.

Helpman, E. and Rangel, A. (1999), 'Adjusting to a new technology: experience and training', Journal of Economic Growth, 4(4), pp. 359-83.

Machin, S. (1996), 'Changes in the relative demand for skills', in Booth, A. and Snower, D. (eds), Acquiring Skills: Market Failures, Their Symptoms and Policy Responses, Cambridge University Press

Machin, S. and Van Reenen, J. (1998), 'Technology and changes in the skill structure: evidence from seven OECD countries', Quarterly Journal of Economics, 113, pp. 1215-44.

Manning, A. (2003), Monopsony in Motion, Princeton University Press.

Nickell, S. and Quintini, G. (2002), 'The recent performance of the UK labour market', Oxford Review of Economic Policy, 18(2), pp. 202-20.

O'Mahony, M. and de Boer, W. (2002), 'Britain's relative productivity performance: updates to 1999 Final Report to DTI/Treasury/ONS', mimeograph.

Riley, R. and Young, G. (2007), 'Skill heterogeneity and equilibrium unemployment', Oxford Economic Papers, pp. doi: 10.1093/oep/ gpm011.

Wilkinson, D. (1998), 'Towards reconciliation of NES and LFS earnings data', Labour Market Trends, 106, 5, pp. 223-31.

NOTES

(1) These include, amongst others, Bound and Johnson (1992), Berman et al. (1994), Machin (1996), Betts (1997), Machin and Van Reenen (1998), Haskel and Heden (1999), Haskel and Slaughter (2001 and 2002), Green et al. (2003), and Riley and Young (2007).

(2) The importance of relaxing the assumption of linearity in schooling is discussed in Heckman et al. (2003). We follow Cingano (2003) by relaxing the assumption of linearity in tenure.

(3) The hourly wages were deflated to 2000 prices using the implied UK National Accounts consumption expenditure deflator.

(4) There is a sample of approximately 2,000 responding households in Northern Ireland that have not been included in this analysis. Thus we use only the 59,000 households from Great Britain as our sample.

(5) From Winter 1993 the LFS records industry using the Standard Industrial Classification 1992. Before then industry is classified by the Standard Industrial Classification 1980.

(6) Working age is defined as 16-64 for men and 16-59 for women.

(7) Consistent with Anderson et al. (2001) and Manning (2003).

(8) Defined as current age minus age left full-time education.

(9) The regional classification is based on the August 1998 definition of Government Office Regions. Residents of Northern Ireland were not included in the sample.

(10) For further details on the NISEC dataset refer to O'Mahony and de Boer (2002).

Simon Kirby and Rebecca Riley, National Institute of Economic and Social Research. e-mail: s.kirby@niesr.ac.uk, r.riley@niesr.ac.uk. We are grateful to Ray Barrell, Stephen Hall, participants at an ESRC user-group seminar, held at NIESR in October 2003, and participants at the conference of the Royal Economic Society, held in Swansea in April 2004, for helpful comments and suggestions and to Mary O'Mahony for assistance with the NISEC database. We gratefully acknowledge the financial support received from the ESRC under grant LI38250122. Disclaimer: Material from the Labour Force Survey is Crown Copyright; it has been made available by the Office for National Statistics (ONS) through the UK Data Archive (UKDA) and has been used by permission. Neither the ONS nor the UKDA bear any responsibility for the analysis or interpretation of the data reported here.

Table 1. Earnings function sample mean characteristics

Number of observations 184044

Log hourly wage 1.92
Years of continuous full-time education 11.96
Years of tenure with current employer 7.51
Potential experience 20.96
Male 0.59
Agriculture and non-manufacturing production
 (reference category) 0.03
Manufacturing: chemicals and allied products 0.04
Manufacturing: basic metals 0.03
Manufacturing: machinery 0.03
Manufacturing: electrical and optical equipment 0.04
Manufacturing: transport 0.03
Manufacturing: food, drink and tobacco 0.03
Manufacturing: other manufacturing 0.08
Construction 0.07
Wholesale and retail 0.20
Hotels and restaurants 0.05
Transport 0.06
Communications 0.03
Financial intermediation 0.07
Business services 0.13
Personal services 0.07
Year sample is 1994 (reference category) 0.13
Year sample is 1995 0.13
Year sample is 1996 0.13
Year sample is 1997 0.13
Year sample is 1998 0.13
Year sample is 1999 0.13
Year sample is 2000 0.12
Year sample is 2001 0.11
Resident in the North west (reference category) 0.05
North east 0.10
Yorkshire & Humberside 0.09
East Midlands 0.08
West Midlands 0.10
East 0.04
London 0.10
South east 0.21
South west 0.09
Wales 0.04
Scotland 0.09
<25 employees at workplace (reference category) 0.34
25-49 0.11
50 or more 0.53
Don't know but over 24 0.01
Male private sector employee 0.54
Female private sector employee 0.38
Cohabiting 0.60
Full-time hours 0.81

Table 2. Distribution of industries by year

 Total
 1994 2001 sample

Agriculture and non-
 manufacturing production 3.61 2.84 3.17
Manufacturing: chemicals and
 allied products 3.85 3.23 3.59
Manufacturing: basic metals 3.38 3.17 3.38
Manufacturing: machinery 3.18 2.77 2.94
Manufacturing: electrical and
 optical equipment 4.17 3.88 4.10
Manufacturing: transport 3.55 3.30 3.37
Manufacturing: food, drink and
 tobacco 3.03 2.51 2.98
Manufacturing: other
 manufacturing 9.04 7.03 8.16
Construction 6.12 7.39 6.66
Wholesale and retail 20.26 19.76 20.12
Hotels and restaurants 4.87 4.59 4.93
Transport 5.95 6.41 6.19
Communications 3.21 3.76 3.29
Financial intermediation 7.51 7.51 7.25
Business services 11.62 15.03 13.30
Personal Services 6.65 6.82 6.57

Base 23115 20283 184044

Table 3. OLS earnings estimates

 Coefficient t-statistic

Constant -0.65 1.11
Private sector (male) 0.036 7.49
Private sector (female) -0.084 15.00
Cohabiting 0.061 27.06
Full-time hours 0.146 45.61
Female (reference category)
Male 0.079 11.63
Resident in the North west
 (reference category)
North east 0.032 6.67
Yorkshire & Humberside 0.011 2.28
East Midlands 0.022 4.30
West Midlands 0.015 3.15
East 0.047 7.83
London 0.239 45.95
South east 0.171 38.32
South west 0.031 6.06
Wales -0.021 3.72
Scotland 0.030 6.10
<25 employees at workplace
 (reference category)
25-49 employees 0.082 25.42
50 or more 0.154 67.01
Don't know, but over 24 0.081 7.04
Sample size 184044
Adjusted RZ 0.441
MSE 0.403

Notes: Reference category is female, born 1929-39, resident in North
west, <25 employees at workplace, quarter 1 in 1994 sample, employed
in the agriculture and non-manufacturing production sector;
industry-year dummy variables, cohort dummy variables, and sample
quarter included but not reported here. For the industry-year specific
coefficients on the schooling and tenure variables see figures 1 and 2;
reported t-statistics are calculated using robust standard errors.

Table 4. ICT intensity Zf capital

 Mean ICT/K Absolute
 ratio change in
 1994-2001 ICT/K
 1994-2001

Agri. and non-manuf. production 0.021 0.016
Manuf. chemicals and allied 0.057 0.050
Manuf. basic metals 0.044 0.046
Manuf. machinery 0.087 0.113
Manuf. electrical and optical 0.157 0.137
Manuf. transport 0.056 0.032
Manuf. food, drink and tobacco 0.043 0.021
Manuf. other manuf. 0.068 0.075
Construction 0.041 0.021
Wholesale and retail 0.093 0.094
Hotels and restaurants 0.008 0.005
Transport 0.036 0.063
Communications 0.368 0.364
Financial intermediation 0.176 0.198
Business services 0.139 0.175
Personal services 0.033 0.050

Table 5. Regression of schooling and tenure returns on capital and
ICT intensity

 The effect of capital on the
 return to schooling:

 95% confidence
 interval

Model 1 -0.0018 (0.69) -0.0069 0.0033
Model 2 -0.0032 (0.22) -0.0324 0.0259
Model 3 -0.0010 (0.06) -0.0301 0.0282

 The effect of ICT on the
 return to schooling:

 95% confidence
 interval

Model 1 0.0929 (3.96) 0.0468 0.139
Model 2 0.0448 (1.69) -0.0072 0.0968
Model 3 0.0799 (2.26) 0.0103 0.1495

 The effect of capital on the
 return to tenure:

 95% confidence
 interval

Model 1 0.0007 (1.63) -0.0002 0.0016
Model 2 -0.0011 (0.35) -0.0072 0.0051
Model 3 -0.0002 (0.07) -0.0064 0.0060

 The effect of ICT on the
 return to tenure:

 95% confidence
 interval

Model 1 -0.0067 (1.62) -0.0148 0.0014
Model 2 -0.0351 (6.28) -0.0461 -0.0241
Model 3 -0.0367 (4.87) -0.0515 -0.0219

 Difference:

 95% confidence
 interval

Model 1 -0.0025 (1.03) -0.0074 0.0023
Model 2 -0.0022 (0.14) -0.0326 0.0283
Model 3 -0.0007 (0.05) -0.0313 0.0298

 Difference:

 95% confidence
 interval

Model 1 0.0996 (4.49) 0.056 0.1431
Model 2 0.0799 (2.89) 0.0255 0.1342
Model 3 0.1166 (3.15) 0.0438 0.1894

Notes: t-statistics in parentheses; 512 observations (4 equations * 16
industries*8years); industries weighted by share of employees in year
total; years weighted equally. All models include a constant term.
Model 2 includes industry dummies. Model 3 includes industry and year
dummies. Effects evaluated at sample mean levels of schooling and
tenure as shown in equation (6).