4.3 Sequencing of upskilling and technological change
A related question is whether firms adopting new technologies already have more skilled workforces and higher productivity. Initially, studies such as Krueger (1993) (using cross-sectional data) found a strong correlation between wages and computer use by workers: that is, computer users were better paid than non-users. However, phases of technical change such as the use of computers may not have made the workers using them more productive, thereby causing the increase in wages. Rather, computer use may have been assigned to workers who are more able. Disentangling these two wage effects requires panel data that tracks the wage history of the same employee over time as technology changes. Both Hall and Kramarz (1998) and De la Fuente and Ciccone (2002) note that panel data points to unobserved ability being the primary cause of the wage premium for computer use. In general terms, these studies find that, once controls are made for unobserved individual characteristics (eg reliability, knowledge, trainability), the positive association between the increased use of new technologies and wages disappears. Contrasting evidence has been provided for New Zealand by Daldy and Gibson (2003) using computer training data in the 1996 Education and Training Survey. They found a significant (12%) wage effect for those who received such training, after controlling for a number of individual, occupation and industry effects.
Doms, Dunne and Troske (1997) used disaggregated US plant-level data that enabled them to match technology use in manufacturing plants in 1988 and 1993 with worker characteristics in 1990. Their data shows that “high tech” plants employ more skilled, higher paid, workers in both production and non-production activities. The fraction of workers employed in scientific, engineering, managerial and precision-craft occupations increases with the use of new technologies, while technologically advanced plants employed higher paid production, technical, clerical and sales workers. However, their data shows little correlation between the act of technology adoption and skill upgrading or wage increases. They find that plants adopting more new technologies had already employed a more skilled workforce (ie, a greater proportion with college degree or better), and have above average productivity in an earlier period. They see the correlation between new technology use and higher wages as reflecting the greater likelihood that plants with more educated and higher paid workforces will adopt more new technologies. Thus, when controls for worker characteristics were included, the wage premium for new technology usage dropped substantially.
Chennels and Van Reenen (1997) use British plant level data to investigate the relationship between the more intensive use of new technologies and wages. They use the 1984 and 1990 Workplace Industrial Relation Surveys, which contain considerable detail on workplace characteristics including the introduction of new technologies, but have limited data on worker skills. They find that plants that utilise technology more intensively pay higher wages. The question explored is the direction of causation – whether the use of new technologies causes higher pay, or whether those with higher pay facilitate technology adoption. They find that the second explanation is correct.
In a similar study, Entorf and Kramarz (1998)use longitudinal matched firm and worker data to look at whether French workers using a range of new technologies received higher pay before, during, or after the adoption of those technologies. The panel data they use for 1985 and 1986 contains data on the same workers and technology introduction by their employing firms. This allows for controls for worker and firm effects. They also find that workers using new technologies are paid more than non-user workers. Again, they find that new technologies are more likely to be used by abler and better paid workers, with the use of computer based technologies yielding a 16% wage advantage – which is of the same order as measured by Krueger (1993). This wage premium is able to be decomposed into 6% for workers with zero experience who begin using those technologies, and 10% due to experience. Once fixed worker characteristics are included, the 6% premium disappears, but a small part of the experience component remains. They suggest that firms select their ablest workers, who were already better paid, to use new technologies. As these workers learn and acquire experience, their wages further increase over time.
Entorf, Gollac and Kramarz (1999) use later observations from the same data set to further explore the effects of new technologies on wages and employment. They continue to find that workers using new technologies were already better paid before their introduction. Controlling for unobserved worker characteristics reduced the wage premium for computer use from 15-20% to below 2%.
