Intangible investment in people and productivity

Abstract

Organizational activity, information and communication technology work, and research and development (R&D) can be classified as work that creates intangible capital. We measure the returns to these three types of labor input by accounting for differences in their productivity compared with other labor inputs using Finnish firm-level data from 1998 to 2008. We apply a novel idea to use hiring as one proxy for productivity and demand shocks. We find that organizational workers increase total factor productivity and improve the profitability of high-productivity firms. R&D workers account for a large share of intangible capital; however, the returns to R&D are low. Investments in organizational competence are more likely to result in more rapid productivity growth. Firms with performance-related pay or domestically owned firms with extensive foreign activities have been among the highest performers with respect to the use of organizational work.

Appendices

Appendix 1: production function

We assume a constant-returns-to-scale production function in which labor input is quality adjusted:

$$Y = b_{0} (QL)^{{b_{1} }} K^{{b_{2} }} \exp (\varepsilon ) $$

(4)

where Y is the value added; QL is the quality-adjusted labor input (L is the total number of employees, and Q is the labor quality index); K is the net plant, property, and equipment; and ɛ is an error term. For the labor input, we separate the organizational (OC), R&D, and ICT workers, and the other workers serve as the reference group. Following the approach used by Griliches (1967) and Hellerstein et al. (1999) in another setting, the quality-adjusted labor input is written as follows:

$$\begin{aligned} QL &= a_{OC} OC + a_{RND} RND + a_{ICT} ICT + \left( {L - OC - RND - ICT} \right) \\ &= L\left[ {1 + \left( {a_{OC} - 1} \right)\frac{OC}{L} + \left( {a_{RND} - 1} \right)\frac{RND}{L} + \left( {a_{ICT} - 1} \right)\frac{ICT}{L}} \right] \\ \end{aligned} $$

(5)

where OC, RND, and ICT are the total number of organizational, R&D, and ICT workers, respectively, in each firm. We allow the productivity of OC, R&D, and ICT workers to vary from that of the other workers by the factors a _OC , a _RND , and a _ICT , respectively. In logarithmic form, we can approximately write the following:

$$\ln \left[ {1 + \left( {a_{OC} - 1} \right)\frac{OC}{L} + \left( {a_{RND} - 1} \right)\frac{RND}{L} + \left( {a_{ICT} - 1} \right)\frac{ICT}{L}} \right] \approx \left( {a_{OC} - 1} \right)\frac{OC}{L} + \left( {a_{RND} - 1} \right)\frac{RND}{L} + \left( {a_{ICT} - 1} \right)\frac{ICT}{L} $$

(6)

because the second, third, and fourth terms in the brackets are not too far from zero. The reason for using the approximation (6) is that a functional form that is nonlinear in parameters and variables would not be compatible with the estimation approach (proxy variable estimation) that we use. Further, using a more general production function than the Cobb-Douglas form (4), such as the translog function, would also complicate the estimation, as the number of variables would increase.

The production function can be written in logarithmic form as follows:

$$\begin{aligned} \ln Y &= c_{0} + c_{OC} \frac{OC}{L} + c_{RND} \frac{RND}{L} + c_{ICT} \frac{ICT}{L} \\ & \quad + \,b_{1} \ln L + b_{2} \ln K + \varepsilon \\ \end{aligned} $$

(7)

where c ₀ = lnb ₀, c _OC  = b ₁(1 − a _OC ), c _RND  = b ₁(1 − a _RND ), and c _ICT  = b ₁(1 − a _ICT ). Equation (7) can now be expressed in terms of TFP as follows:

$$\begin{aligned} \ln TFP &= c_{0} + c_{OC} \frac{OC}{L} + c_{RND} \frac{RND}{L} + c_{ICT} \frac{ICT}{L} + \varepsilon \\ &= c_{0} + c^{\prime } X + \varepsilon \\ \end{aligned} $$

(8)

where

$$\ln TFP = \ln Y - b_{1} \ln L - b_{2} \ln K $$

(9)

and where, in the right-hand side of (8), X = (OC/L, RND/L, ICT/L)′ is a vector of the ratios of the three types of employees to the total number of employees, and c is a vector of the respective productivity parameters (to be estimated subsequently). b ₁ is proxied by wage costs per value added at the two-digit industry level, and b ₂ = 1 − b ₁.

If compensation for intangible capital is based on productivity, then the influence of the worker proportions should have the same effect on production and average wages. To analyze the wage effects, we estimate a model for the average wage W:

$$\begin{aligned} W &= f_{0} + f_{OC} \frac{OC}{L} + f_{RND} \frac{RND}{L} + f_{ICT} \frac{ICT}{L} + \varepsilon \\ &= f_{0} + f^{\prime } X + \varepsilon \\ \end{aligned} $$

(10)

In Eq. (7), we observe that ∂Y/∂(OC/L) ≈ c _OC  = b ₁(a _OC  − 1), and in Eq. (8), we observe that ∂TFP/∂(OC/L) = c _OC ; thus, we can analyze the productivity effect from the TFP equation. However, because we need to compare (a _OC  − 1) to the coefficient of OC/L in the wage model (10), we must first divide the productivity model coefficient by the labor share coefficient b ₁. The wage-productivity gap is then defined as follows:

$$gap_{OC} = (c_{OC} /b_{1} ) - f_{OC} $$

(11)

where f _OC is the coefficient from the wage model (10). For R&D and ICT employees, the gaps are calculated in a similar manner. The total wage-productivity gap for a firm is defined as follows:

$$gap = (OC/L)gap_{OC} + (RND/L)gap_{RND} + (ICT/L)gap_{ICT} $$

(12)

The aggregate figures are calculated as firm size weighted averages.

Appendix 2: human capital

Human capital, lnHC _it , is based on the firm average of the person effects obtained from the estimation of wage models for individuals with separate person and firm effects (see Abowd and Kramarz 2005). The wage regression includes only time-varying person characteristics as deviations from their means. The dependent variable is the log of the wage ln(w _jit ) of person j working in firm i at time t, measured as a deviation from the individual mean, μ _wj . This variable is expressed as a function of individual heterogeneity, firm heterogeneity, and measured time-varying characteristics in the following equation:

$${ \ln }\left( {w_{jit} } \right) - \mu_{wj} \, = \, \theta_{j} + \psi_{I(j,t)} \, + \beta^{\prime } \left( {x_{jt} - \mu_{xj} } \right) + e_{jit} $$

(13)

where θ _j is the individual fixed effect (interpreted as time-invariant compensation for human capital) and ψ _I(j,t) is the firm fixed effect (captures the unmeasured employer heterogeneity), where I(j, t) indicates the employer of j at date t. β′(x _jt  − μ _xj ) indicates the compensation for time-varying HC, which is stated as a deviation from the individual mean, and e _jit is the error term. The time-variant variables x _jt are experience (age minus years of education minus age when entering school) and seniority (duration in one’s job measured in years). Individual heterogeneity, which is captured by the person-specific fixed effect, includes the returns to education, and the remaining part of the person-specific fixed effect is the proportion of wages that cannot be explained by observed characteristics (to the econometrician). The firm-level HC variable is measured as the average of the θ _j terms of the employees for each firm in each year. In their study, Abowd et al. (2002) develop a numerical solution to address a large set of firm dummies when evaluating both individual and firm fixed effects simultaneously. We use their method as applied to Stata by Ouazad (2008). The HC measure also includes abilities that are not reflected in education and work experience; thus, this measure provides a more valid indication of the abilities of less educated production workers (40 % of all employees in our data).

Appendix 3: occupational classifications of non-production workers

See Table 6.

Table 6 Occupational classifications

Appendix 4: descriptive statistics

See Tables 7 and 8.

Table 7 Variables in the productivity estimation

Table 8 Variables used for explaining productivity changes