Research and development and labour productivity: do high-tech firms exhibit labour- or capital-saving technical change?

ABSTRACT

Employment and output in the advanced technology sectors have generally exhibited above-average growth for more than two decades. While this industry accounts for a relatively small share of total employment, the majority of private sector research and development (R&D) expenditures in the US is concentrated within seven sub-sectors. However, little attention has been paid as to whether high-tech productivity exhibits Hicksian capital or labour ‘savings’ bias or tendency to displace either factor input over time. Biased technical change can occur as economies transition between growth regimes. An augmented production function is employed to analyse the additional impact of R&D activity on firm-level labour productivity. A panel data set comprised of high-tech firms located across the advanced economies, China and India from 1990 to 2013 is used in the analysis. Labour-saving technical change was present across the advanced technology sectors and most countries. The expanded models of labour productivity that used fixed effects with lagged regressors confirmed the prior results as well as finding that R&D per employee, relative R&D intensity and firm market share contribute to firm-level labour productivity growth across countries and sectors. Additional support was found for diminishing returns to scale but not for R&D spillover effects.

KEYWORDS:

• Biased technical change
• high-tech sectors
• R&D expenditures

JEL CLASSIFICATION:

• O30
• G30
• C33

Acknowledgements

The author acknowledges support from the Faculty Development Funding Program, Ramapo College of New Jersey. He wishes to thank Kevin Ng, Dr Marie Duggan and an anonymous referee for their comments. .

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ The three-digit SIC code high-tech sectors identified in Brown, Fazzari, and Petersen (2009) include drugs (283), office and computing equipment (357), communications equipment (366), electronic components (367), measuring and controlling (scientific) instruments (382), medical instruments (384), and computer and data processing (737).

² Total factor productivity growth is that which exceeds the growth of capital and labour inputs.

³ Hall and Mairesse (1995, 268) observed that in their panel of French manufacturing companies ‘firms are clearly becoming more capital intensive over time, since employment is declining substantially over the whole period [1980–1987]’ while the capital-to-labour ratio was rising. Wakelin (2001, 1083) confirms a similar pattern for a panel of 170 UK manufacturing firms, where ‘deflated sales per employee rose by 2%, with a fall in employment of almost 7% and a rise in the real capital to labor ratio of 10% on average over the period [1988–1996].’

⁴ A positive rate of labour productivity growth necessarily implies that output growth exceeds employment growth.

⁵ Based on summations of firm-level output and employment across the seven high-tech sectors.

⁶ Temple (2006) summarizes methodological issues with using aggregate production functions to model growth and technical change.

⁷ The expression for Solow’s residual term, dQ/dt/Y in Equation (4) is used in Jones (1976) to denote the ‘proportional rate of growth of output which is attributable to technical progress rather than any increase in the factors employed’ (171).

⁸ ‘Divide both sides of the production function by output [Y], yielding the balance expression 1 = F(K_t/Y_t,L_t/Y_t; t). Capital accumulates and inherits the trend in output, so the capital–output ratio is constant in steady state. Labour does not inherit the trend in output; however, so L_t/Y_t, falls in steady state. To satisfy the balance equation, technical change must exactly offset the decline in L_t/Y_t. That is, technical change must be labour augmenting’ (Jones and Scrimgeour 2008, 182).

⁹ As Hall and Mairesse (1995, 274) observe, the employment coefficient, β in Equation (8), is equal to 1 minus the sum of the capital-to-labour ratio (α) plus the R&D to labour coefficient (γ) in Equation (9) or equivalently, 1 − α – γ + (η − 1) = η – α – γ = β. Thus, the coefficient, β on ln(L_ijt) in Equation (9) will be the estimate of (η − 1).

¹⁰ The Thomson Financial ‘Worldscope’ database includes employee wages under ‘selling and general administrative expenses’ with wages being the majority of these costs.

¹¹ PPIs for individual countries were obtained from the Federal Reserve Board ‘FRED’ database.

¹² The STATA 11 xtregar one-step procedure is employed to estimate ρ and create generalized least squares estimates. Baltagi and Wu (1999) proposed this method for correcting for serial correlation when using unbalanced panel data. Additional information on methods and formulas can be found in STATA (2009).

¹³ Using fixed-effects controls for the most granular measure of firm-level heterogeneity. Thus, it is not possible to include dummy variables because they would introduce perfect or almost perfect collinearity. The null hypothesis that all the fixed-effects coefficients are simultaneously equal to zero is tested with an F-statistic and reported in the column labelled ‘FE F-value’.

¹⁴ There are no published critical values for the Baltagi–Wu test for first-order serial correlation in unbalanced panels. Similar to the Bhargava, Franzini, and Narendranathan (1982), Durbin Watson statistic for balanced panels, Baltagi–Wu values close to 2.00 suggest the absence of serially correlated residuals.

Additional information

Funding

This work was supported by the Ramapo College of New Jersey [Faculty Development Funding Award].