Computers, skills and wages

Abstract

Computer technology is most prominently used by skilled, high-wage workers. This suggests that computer use requires skills to take full advantage of the possibilities, which are particularly present among relatively skilled workers. This article develops a simple technology adoption model showing that the decision to adopt computer technology depends on (i) the tasks to be performed, (ii) the level of skill or education and (iii) the level of wages. Applying this model to British data, it is shown that the effect of wages and particular tasks on computer adoption is larger than the effect of skills on adoption. The estimates suggest that in Britain computer use is likely to be a matter of cost efficiency and not so much of workers’ skills.

Acknowledgements

We would like to thank the referee of this journal, Daron Acemoglu, Josh Angrist, David Autor, Eli Berman, Clair Brown, Allard Bruinshoofd, Francesco Caselli, Eric Gould, Hans Heijke, Hugo Hollanders, Caroline Hoxby, Adriaan Kalwij, Lawrence Katz, Francis Kramarz, Alan Krueger, Erzo Luttmer, Markus Möbius, Richard Murnane, Jörn-Steffen Pischke, Arthur van Soest, Luc Soete, John Van Reenen, Bruce Weinberg, Finis Welch, Thomas Ziesemer and seminar participants at EALE, EEA, ECIS, Harvard, LSE, MERIT, CentER, the University of Oxford and ZEW for helpful comments on earlier drafts of this article. We are particularly grateful to Alan Felstead and Francis Green for providing the data and for comments on an earlier version of this article. This research has been supported by the Netherlands Organization for Scientific Research (NWO).

Notes

¹ The continuous fall in the prices of computer equipment has been documented and examined by Baker (1997), Caudill and Gropper (1997), Greenwood and Yorukoglu (1997), Autor et al. (1998), Filson (1998), Krusell et al. (2000), Jorgenson (2001) and Borghans and ter Weel (2007). In addition, Jorgenson (2001) provides figures showing a more than 10% annual decline in software prices since the 1970s. These two trends are likely to have amplified the computerization of the workplace.

² See Borghans and ter Weel (2004) for an elaborate version of the model presented here.

³ These costs can not only be thought of as maintenance, depreciation and operating costs, but also as costs of new software applications, hardware and technical assistance.

⁵ In Appendix Table B1 we report some descriptive statistics of the variables used in the analysis of this article.

⁴ Ashton et al. (1999) provide a description of the survey and questionnaire. See Dickerson and Green (2004) for a methodological paper on measuring job skills using survey data. They apply the same data as we do here to evaluate their approach.

⁶ We do not report the first-stage results in Table 2 because they do not yield additional insight. The F-test criterion of Staiger and Stock (1997) is easily passed for all three regressions reported in Table 2.

⁷ Appendix Table B1 lists the industries. We take agriculture, forestry, fishing, energy and water supply as the reference group.

⁸ The inclusion of occupational dummies into the regression leads to similar results.

⁹ For a number of tasks almost every respondent answered that these aspects were at least ‘not very important’. This led to numerical problems in the maximum likelihood estimation. For this reason, tasks which were reported by at least 95% of the workers to be part of their job have been excluded from the estimation.

¹⁰ Similar to the results reported in Table 1, the point estimates and marginal effects of education are substantial but not significant. Relative to a worker without a degree, a worker with some sort of degree has a larger probability to use a computer at work. Also, being female still exerts a strong and positive impact on the likelihood of using a computer at work.