Knowledge-based service economy and firm entry: an alternative to the knowledge spillover theory of entrepreneurship

The authors thank Prof. Haifeng Qian for sharing the BITS data.

Plummer and Acs (2014) use the SIC1987 industrial classification, whereas Fallah and co-authors (Fallah et al. 2014) use the NAICS2002 classification. Although there might be inconsistencies in the definitions of the high-tech sectors by these two studies, the major revamp of the way industries were classified between SIC and NAICS is likely to result in a considerably greater mismatch if instead of using the two studies we rely on a cross-walk between SIC and NAICS provided by the US Census Bureau. This is so because there is no way to map industries from SIC into NAICS definitions with sufficient precision, as often one industry in SIC corresponds to several industries in NAICS and vice versa.

For the years when SIC data are used for the dependent variable, the computer and electronic manufacturing industry is defined according to the US Census Bureau concordances and includes SIC3495 Household Audio and Video Equipment; SICSIC3571 Electronic Computers; SIC3572 Computer Terminals; SIC3577 Computer Peripheral Equipment, NEC; SIC3578 Calculating and Accounting Machinery; SIC3579 Office Machines, NEC; SIC3651 Household Audio and Video Equipment; SIC3652 Phonograph Records and Prerecorded Audio Tapes and Disks (reproduction of all other media except video); SIC3661 Telephone and Telegraph Apparatus; SIC3663 Radio and Television Broadcasting and Communications; SIC3669 Communications Equipment, NEC; SIC3671 Electron Tubes; SIC3672 Printed Circuit Boards; SIC3674 Semiconductors and Related Devices; SIC3675 Electronic Capacitors; SIC 3676 Electronic Resistors; SIC3677 Electronic Coils, Transformers, and Other Inductors; SIC3678 Electronic Connectors; SIC3679 Electronic Components, NEC; SIC3695 Magnetic and Optical Recording Media; SIC3812 Search, Detection, Navigation, Guidance, Aeronautical, and Nautical Systems and Instruments; SIC3822 Automatic Controls for Regulating Residential and Commercial Environments and Appliances; SIC3823 Industrial Instruments for Measurement, Display, and Control of Process Variables, and Related Products; SIC3824 Totalizing Fluid Meters and Counting Devices; SIC3825 Instruments for Measuring and Testing of Electricity and Electrical Signals; SIC3826 Laboratory Analytical Instruments; SIC3829 Measuring and Controlling Devices, NEC; SIC3842 Orthopedic, Prosthetic, and Surgical Appliances and Supplies (electronic hearing aids); SIC3844 X-Ray Apparatus and Tubes and Related Irradiation Apparatus; SIC3845 Electromedical and Electrotherapeutic Apparatus; SIC3873 Watches, Clocks, Clockwork Operated Devices, and Parts; SIC3915 Jewelers Findings and Materials and Lapidary Work (watch jewels); SIC7372 Prepackaged Software (mass reproduction of software); SIC7819 Services Allied to Motion Picture Production (reproduction of video).

Since the BITS data for the first 2 years of our research relies on SIC industry classification, it might be technically incorrect to refer to the computer and electronic manufacturing industry in our study as NAICS334. Although we try to spell out the industry name throughout the paper, in some cases for brevity, we default to the NAICS334 as a title for the industry.

Lagging by 3 or 5 years produces comparable results, but fewer models are identified. We, therefore, use 4 years to lag patent-related variables.

The measure is based on manufacturing because US PTO classifies utility patents into manufacturing industries only.

http://​www.​uspto.​gov/​web/​offices/​ac/​ido/​oeip/​taf/​naics/​stc_​naics_​faall/​usa_​stc_​naics_​fa.​htm.

Relying on a four-digit NAICS classification allows capturing the level of related variety in a region, which is important for knowledge flows, whereas using two-digit NAICS sectors would capture unrelated variety that is important for guarding regional economies from negative shocks (Frenken et al. 2007).

An earlier version of this manuscript included location quotient (LQ) as the second variable relevant to the industrial conditions. This variable was highly correlated with population density in all high-tech models introducing multicollinearity problems as evidenced by a VIF of around 20. The models that omit LQ performed better in terms of identification compared with the models without population density. The results were identical, and we, therefore, proceeded by dropping LQ from our specification.

As a sensitivity test for the effects of the Great Recession, we estimated the models separately for years 1997–2007 and 2007–2011. The estimation results for the former period were practically identical to the ones reported in this paper; in the case of the latter period, very few variables were significant and the strength of the instruments was considerably below the traditional cutoff value of 10 indicating weak instruments.

Evidence using other instrument sets is reported in the Sensitivity analysis section.

Estimated effects of patenting intensity on total business entry and business entry in high-tech sector and in high-tech nongoods-producing subsector becomes positive and statistically significant if the models exclude time fixed effects. In this specification, the negative effects in the high-tech goods-producing sector and in computer and electronic product manufacturing do not change.

We omit the term Lⁱ, a measure of labor shedding in industry i, as it relates more to necessity entrepreneurship, which is assumed away when a relationship between knowledge production and business entry in the high-tech sector is considered.