Assessing education’s contribution to productivity using firm-level evidence

There is plenty of individual-level evidence, based on the estimation of Mincerian equations, showing that better-educated individuals earn more. This is usually interpreted as a proof that education raises labour productivity. Some macroeconomists, analysing cross-country time series, also support the idea that the continuous expansion of education has contributed positively to growth. Surprisingly, most economists with an interest in human capital have neglected the level of the firm to study the education-productivity-wage nexus. And the few published works considering firm-level evidence are lacking a proper strategy to cope with the endogeneity problem inherent to the estimation production and wage functions. The purpose of this paper is to aim at providing estimates of the causal effect of education on productivity and wage labour costs.

This paper taps into a rich, firm-level, Belgian panel database that contains information on productivity, labour cost and the workforce’s educational attainment to deliver estimates of the causal effect of education on productivity and wage/labour costs. Therefore, it exclusively resorts to within firm changes to deal with time-invariant heterogeneity bias. What is more, it addresses the risk of simultaneity bias (endogeneity of firms’ education-mix choices in the short run) using the structural approach suggested by Ackerberg et al. (2006), alongside more traditional system-GMM methods (Blundell and Bond, 1998) where lagged values of labour inputs are used as instruments.

Results suggest that human capital, in particular larger shares of university-educated workers inside firms, translate into significantly higher firm-level labour productivity, and that labour costs are relatively well aligned on education-driven labour productivity differences. In other words, the authors find evidence that the Mincerian relationship between education and individual wages is driven by a strong positive link between education and firm-level productivity.

Surprisingly, most economists with an interest in human capital have neglected the level of the firm to study the education-productivity-pay nexus. Other characteristics of the workforce, like gender or age have been much more investigated at the level of the firm by industrial or labour economists (Hellerstein et al., 1999; Aubert and Crépon, 2003; Hellerstein and Neumark, 2007; Vandenberghe, 2011a, b, 2012; Rigo et al., 2012; Dostie, 2011; van Ours and Stoeldraijer, 2011). At present, the small literature based on firm-level evidence provides some suggestive evidence of the link between education, productivity and pay at the level of firms. Examples are Hægeland and Klette (1999); Haltiwanger et al. (1999). Other notable papers examining a similar question are Galindo-Rueda and Haskel (2005), Prskawetz et al. (2007) and Turcotte and Whewell Rennison (2004). But, despite offering plausible and intuitive results, many of the above studies essentially rely on cross-sectional evidence and most of them do not tackle the two crucial aspects of the endogeneity problem affecting the estimation of production and wage functions (Griliches and Mairesse, 1995): first, heterogeneity bias (unobserved time-invariant determinants of firms’ productivity that may be correlated to the workforce structure) and second, simultaneity bias (endogeneity in input choice, in the short-run, that includes the workforce mix of the firm). While the authors know that labour productivity is highly heterogeneous across firms (Syverson, 2011), only Haltiwanger et al. (1999) control for firm level-unobservables using firm-fixed effects. The problem of simultaneity has also generally been overlooked. Certain short-term productivity shocks affecting the choice of labour inputs, can be anticipated by the firms and influence their employment decision and thus the workforce mix. Yet these shocks and the resulting decisions by firms’ manager are unobservable by the econometrician. Hægeland and Klette (1999) try to solve this problem by proxying productivity shocks with intermediate goods, but their methodology inspired by Levinsohn and Petrin (2003) suffers from serious identification issues due to collinearity between labour and intermediate goods (Ackerberg et al., 2006).