Technology and Skill: An Analysis of Within and Between Firm Differences

1. Technology and Skill:An Analysis of Within and Between Firm Differences John Abowd, John Haltiwanger, Julia Lane, Kevin McKinney, and Kristin Sandusky

2. Outline of Talk • Skill-biased technical change • Our research and objectives • Measuring human capital • The demand for human capital • Cross sectional results • Partial adjustment results • Conclusions

3. Skill-biased Technical Change

4. Capital-labor substitution • Labor is differentiated by skill class • High skill • Low skill • Capital is differentiated by investment type • Information technology • Other capital • Information technology and high-skill workers are demand complements • Information technology and low-skill workers are demand substitutes

5. Factor price equalization • US comparative advantage in producing IT and high-skill intensive goods • ROW comparative advantage in producing IT-using and low-skill intensive goods • Factor price equalization via trade reducing the demand for low skill workers and increasing the demand for high skill workers

6. Macroeconomic evidence • Hypothesis originally due to Zvi Griliches, who almost certainly would have attributed it to one of the fathers of microeconomics • Berman, Bound, Griliches (1994) • increased use of non-production workers within manufacturing industries directly related to the increased IT investment and R&D. • Very little of the increase was associated with increased demand for goods produced by non-production worker intensive manufacturing industries (evidence against factor price equalization)

7. Microeconomic evidence • Ichniowski, Shaw and Prennushi (1997) • combination of “high-performance” HRM practices, which included selection and training of skilled workers, complementary with the successful adoption of IT • Bresnahan, Brynjolfsson and Hitt (2002) • increased use of IT directly related to increased demand for skilled employees • Hellerstein, Neumark, and Troske (1999) • capital and skilled labor complements in main analysis (Table 3), but substitutes in other specifications (Table 4)

8. Our Research Objectives

9. Objectives • Measure human capital employed by the business • Exploit the linked employer-employee data • Gather facts: characterize changing distribution of human capital • Within-firm changes • Firm displacement (entry and exit) • Explore why patterns exist • Theory: derived demand for human capital is a function of technology • Measure technology changes and relate to changes in demand for human capital

10. Measuring Human Capital

11. Motivation • Distinguish among similar businesses using the human capital of the employees • Normal measures: employment and wages, sometimes hours • Our measures: a variety of skill indices based on the portable part of the individual's wage rate • Use the differences in the human capital input to help explain differences in the outcomes

12. Theoretical Framework • The general human capital of an employee is represented by h, which is estimated from the portable part of the individual’s wage rate. • The firm-specific part of the wage rate is used to model compensation design issues. • The un-normalized distribution f(h) measures the firm’s human capital choices. • We estimate the normalized distribution of human capital, g(h). • For details see Abowd, Lengermann and McKinney (2003).

13. Measuring Human Capital: Data • State UI wage records and ES-202 • Universal for 3 states (among the seven listed in ALM) • Longitudinal (cover 1990-2003) • Permits linkage of employees and firms • Links to economic data • Annual Survey of Manufacturers (Manufacturing) • Business Expenditure Survey (Non-manufacturing) • Economic Census (1992 and 1997) • Business Register (1992 and 1997)

14. Measuring of Human Capital: Estimation • We use a decomposition of the log real annualized full-time, full-year wage rate (ln w) into person and firm effects. • The person effect is θ. • The firm effect is ψ, where J(i,t) is the employer of i at t. • Continuous, time-varying effects are in xβ, where some of the x variables are human capital measures (labor force experience) and some correct for differential quality in our measure of full-time, full-year wage rate.

15. Human Capital: Individual Measure • Individual human capital, h, is the part associated with the person effect and the measurable time-varying personal characteristics (labor force experience). • Our human capital measure is not a simple ranking by wage rate because of the removal of the firm effect and residual. • Firm human capital measures, H, are based on statistics computed from the distribution of g(h).

16. Human Capital: Distribution • Use the entire workforce present at the establishment at date t in firm j • Take the kernel density estimator of the distribution of hijt • Calculate the proportion of employment in any interval using Gjt(h)

17. Establishment Human Capital Measures • Using gjt(h) measure • Proportion of employment in each quartile of the h distribution (1992 basis) • Separate measure for person effect • Separate measure for experience effect

18. The Demand for Human Capital and Technology

19. Basic Approach to Demand for Human Capital • Production relationship at firm level as function of skill composition for firm j with technology Z: • Treating Z as quasi-fixed, cost minimization (Shepherd’s lemma) yields for workers of type s (where S is share of type s workers):

20. Demand for Human Capital: Basic Features • The demand for workers of type s by a particular firm depends on: • the type of technology adopted (Z) • managerial/entrepreneurial ability • Vintage • Location • Physical and intangible capital • the nature of the firm-worker type complementarities, • the scale of operations • the relative shadow wages

21. Empirical Specification Model 1: Levels Model 2: Partial Adjustment

22. Construction of Linked Data • Human capital file containing worker and firm identifiers, detailed worker characteristics • Business file containing firm identifiers and detailed business characteristics. • These two files linked by employer identifiers to form a business-level file. • Unit of business observation is the most detailed disaggregation available of EIN, State, 2-digit SIC, and county (pseudo-establishment)

23. Weights, Selection, and Other Issues • The sampling frames of the ASM and BES make dynamic analysis difficult • We correct for differential sampling of large and small establishments using special weights • We correct for differential exit using a selection equation • Not all measures are available every in both Censuses • There is no good correction for this

24. Construction of Technology Measures • Data for the manufacturing sector for the 1992 and 1997 Annual Survey of Manufacturers (ASM). • For services, wholesale trade and retail trade we use data from the Business Expenditure Survey (BES). • In the majority of ASM cases, we are able to link the two files by EIN, State, 2-digit SIC (SIC2), and county. • In the BES, there is no state county level detail and the survey is conducted using more aggregated business units (EIN, 2-digit SIC or Enterprise, 2-digit SIC)

25. Technology Measures • Technology Measures • Computer Investment/Total Investment (ASM, BES, 1992 only)  • Spending on Computer Software and Data Processing Services/Sales (ASM, BES, 1992 and 1997) • Inventory/Sales (higher inventories indirect indicator of lack of technology; ASM, BES, 1992 and 1997) • Traditional Technology Measures • Average Beginning and Ending Assets/Employment (ASM 1992 and 1997, BES 1992) • Firm Effect from Wage Equation • Potential proxy for “unmeasured” technology and other things

32. Summary of Findings • There is a strong positive empirical relationship between technology and skill in a cross-sectional analysis of firms. • Technology interacts with different components of skill quite differently: firms that use technology are more likely to use high ability workers, but less likely to use high experience workers. • The partial adjustment analysis supports these conclusions.