To measure the innovation capacity of regions this thesis, following Li (2009), employs the number of domestic patents as the proxy for commercially valuable innovation output. It is used as the dependent variable (DV) in the estimation.
Patent data are the favoured, and most commonly used, indicators in measuring innovation output in regional innovation studies. Although patent information is not perfect, it provides a fairly reliable measure of innovation activity (Acs, et al., 2002; Acs & Audretsch, 1989). Practically, patent statistics are easy to access, available from various patent databases. It is possible to use patent data for longitudinal analysis (Acs & Audretsch, 1989) and the dynamics of technological change (Acs, et al., 2002). It seems patent statistics offer the best available output indicator for innovation activities (Freeman, 2004).
Meanwhile, issues associated with equating patent counts with the level of innovation activity are widely documented in the literature (Acs, et al., 2002; Archambault, 2002; Basberg, 1987; Griliches, 1990; Hagedoorn & Cloodt, 2003; Mansfield, 1986; Pavitt, 1985; Trajtenbery, 1990). According to Griliches‘ (1990) argument not all innovations are patentable, and not all innovations are patented, which questions the representativeness of patents on innovations. Accordingly, some alternative indicators for innovation were used in some empirical studies, such as the number of new products (Fritsch, 2002), new product sales (Liu & White, 1997), and literature-based innovation counts (Acs, et al., 2002). However, these indicators have similar pitfalls as patents, for instance the measurement of economic value of innovations and the quality of innovations.
82
In contrast to some initial studies on IC (Furman, et al., 2002; Hu & Mathews, 2005), domestic patents rather than international patents were used. Although international patents are a good proxy for commercially relevant innovations, they do not reflect the entire spectrum of innovative activities in a country, especially a developing one (Krammer, 2009). Besides, domestic patents reduce the source bias by using international patents from two different databases such as in Furman‘s (2002) work, as the criteria for a patent to be granted in each database may differ. Moreover, domestic patents are more comparable because all the regions are subject to the same national patenting laws, go through the same patenting procedures, and pay the same cost. Therefore domestic patents are much more suitable for this study than international patents.
The State Intellectual Property Office (SIPO) has systematically collected domestic patents since 1985 when China‘s patent law came into force. According to patent law, domestic patents are classified into three categories: invention, utility model and design. Inventions represent the most technologically sophisticated innovation output, radical innovations such as a new products or new methods. Utility models are less innovative compared to inventions. They are incremental innovations, such as the structure change of a product. Designs mainly reflect superficial novelty, such as changes of the shape and color of a product. The basic condition for a patent to be granted is whether it differs from existing technologies and designs, both domestically and internationally, regardless of the patent type. For inventions and utility models, they have to be novel, inventive and practically applicable. As a result, the variation of patent quality is remarkable across the three categories and they
83
differ from each other in terms of novelty, economic value, technological importance, and resource commitment (Li, 2009).
In light of the characteristics of each patent type, it is appropriate to compare the regional capacity according to specific categories. Focusing on more technologically important innovations, this study considers inventions and utility models. As not all innovations are patentable, both total number of applications and granted patents are examined separately to investigate as broad a range of innovation as possible. Granted patents represent innovations with more commercial value than patent applications. Therefore, for specific categories only granted patents are included in this study. Hence, there are four DVs; overall applications, overall granted patents, granted invention patents, and granted utility model patents. The data are divided by the regional population to reduce the bias from regional size differences. Following Furman, et al. (2002) and other researchers (Hu & Mathews, 2005; Li, 2006, 2009), this research use the logarithmic transformation to ensure the distribution of each variable is approximately normal.
In this thesis, data are collected at the regional level. In terms of patent counts, regional means the location of the patent owner. Regional patent counts are the total number of patents that applied by or granted to the owners who are located in a specific region.
A time lag between explanatory variables and patent data (output) is required in the models. It takes time to transform innovation effort to innovation capacity, in other words, transform input to output, and also to process and approve patent applications. In this thesis, one year is taken as the average lag for applications and four years as
84
the lag for all granted patents. However, in reality it is difficult to decide how long it will take R&D efforts to become innovations. Besides, the time lag differs between invention and utility model innovation and it may take more time for invention than for utility models.
For applications, it is assumed it will take at least one year to transform innovation effort to output and to prepare the document for patenting. For the processing time of a patent application, there are different opinions. Cheung & Lin (2004) stated it usually takes the State Patent Office one to one and a half years for an invention patent application, about six months for an utility model patent, and even shorter for design patent. Li (2009) states it usually takes around three years for an invention patent and one year for a utility model patent. A recent study using patents filed at the Chinese State Intellectual Property Office (CSIPO) finds the average duration of invention patent examination is 4.71 years (Wagner & Liegsalz, 2011). In terms of these different opinions and findings, this study takes three years as the patent examination time, disregarding patent type. This amounts to four years lag between input and output for granted patents
To verify if the time lags give best fit, other time lags were checked by analysing a dataset with a fixed time period for IVs and different time lags for DVs using the method of panel data analysis (see results in appendix one). The results show for applications, zero, one year, and two year lags are equal best fit for the model, no matter which type of patent, as there are no big differences between R squares. There are slight differences between the effects of the same IV on different DVs. For granted patents, models with four and five year lags explain more variances than the
85
others for all three DVs. Meanwhile the key results are robust in these two models. Therefore, considering both the verified results and what has been used in previous research, the time lags chosen here are appropriate and reasonable.