Evidence on new technologies and wage inequality in France

Abstract

Using individual data from the French Labour Force Survey and the Complementary Survey on Working Conditions for 1998, we analyse earnings inequalities along the wage distribution between workers using novel Information and Communication Technologies (ICT) at their job and those not using them. We estimate quantile regressions with technological dummies and carry out a decomposition analysis, both at the aggregate level and by occupations. At the aggregate level, most of the wage gap between both populations is explained by the divergence in their labour characteristics. In jobs where ICT are not very diffused, the technological premium is larger than in jobs characterized by a large presence of novel technologies. Whereas in the former type of jobs, the technological premium is mainly justified by a divergence in the labour market characteristics between ICT users and nonusers, in positions characterized by a wide presence of novel technologies the technological premium responds rather to a divergence in the returns to identical characteristics.

Acknowledgements

We would like to thank seminar participants at the Université du Maine, the TEMA seminar of Paris I and the EALE Conference 2006. We are especially indebted to two anonymous referees, David Margolis, Jean-Marc Robin and Catherine Sofer for their very useful comments and suggestions. Any remaining errors are ours.

Notes

¹ Lemieux (2006) attributes to compositional effects the rise in the variance of unobserved skills observed in the US over the past 30 years. Autor et al.'s (2005) findings can though be reconciled with Lemieux's findings if the study is focused on the actual net rise in estimated residual inequality from 1988 to 2003.

² We choose to exclude all individuals who are not full-time workers so as to avoid preference issues problems, and also because for partial time workers it is not always a choice to be at part time. We also eliminate all individuals with missing observations.

³ A shortcoming of these data is that they concern individuals. Not working with plant data limits the possibility of our econometric study to capture the positive productivity spillovers that someone not using ICT, but being employed in a plant where everybody uses them, may receive.

⁴ Around 59% of the formers and 51% of low paid workers using a connected computer at their job receives also internal orders through a computer.

⁵ For the rest of the occupations, there are not very significant differences in terms of diploma between ICT-users and nonusers, apart from the baccalaureate and undergraduate diploma in the employees’ case.

⁶ See Koenker and Bassett (1978) for a detailed description of the quantile regressions.

⁷ Evidently, occupations are differentiated by their ICT content. Therefore, not controlling for them in the regression would leave a lot of noise and would lead to an overestimation of the impact of ICT. We include occupations in the aggregate analysis and implement afterwards a detailed study by occupations.

⁸ At the mean of the sample (Ordinary Least Squares (OLS) regression), the technological premium amounts to 7.3%.

⁹ At the top of the distribution (i.e. the 90th percentile), the premium is slightly higher with MODERN3 than with MODERN1, while there is no difference between MODERN1 and MODERN2.

¹⁰ Results from the quantile regressions with interacted variables are available from the authors upon request.

¹¹ We assume that all workers have the labour market characteristics of the modern workforce and capture the gap attributable to a divergence in the rewards to these characteristics. Alternatively, we can do the same decomposition assuming that workers have the labour market characteristics of the traditional workforce. We did not find any significant differences between both definitions with the data at hand.

¹² For further insights on the selectivity issue, see in particular Krueger (1993), DiNardo and Pischke (1997), and or Lee and Kim (2004). In these studies, there is no appropriate instrumental variable for computer use that would allow a direct estimation of the technological premium. To address the possibility of unobserved heterogeneity, DiNardo and Pischke (1997) introduce additional control variables like parental background and achievement scores in their wage regressions. They also point out the necessity to include a detailed set of occupation dummies, as we do in this section.

¹³ See in particular Section IV (Entorf et al., 1999). As these authors use panel data, they are able to control for unobserved heterogeneity through the use of fixed effect regressions.

¹⁴ The exact specification for the Probit model accounts for gender, age (25 and less, 26–35, 36–45, 46–55, more than 55), seniority (5 and less, 6–10, 11–15, 16–20, more than 20), education (no degree BEPC, CAPBEP, baccalaureate, undergraduate, graduate-postgraduate), French citizenship, firm's size (1–19 employees, 20–49, 50–99, 100–499, 500–1000, more than 1000), and two interacted variables resulting from multiplying seniority respectively by gender and French citizenship. For the manager group, we also add the product of the age and gender variables. Detailed results are available upon request.

¹⁵ The gap between the exogenous premium and the ATT estimate is in fact very low for each occupation, less than 1% on average. The largest difference (1.5%) is observed among low-skilled workers.

¹⁶ Nevertheless, we find that results provided by MODERN1 and MODERN2 are fairly similar, and it is rather the use of a third technology (MODERN3) that constitutes a threshold value in our findings.

¹⁷ We find a kind of U-shaped evolution only for intermediate professions.