Chapter 4: Research method
4.3 Variable measurement
Profit
R&D efforts usually result in product or process improvements (Tsai and Wang 2005). For R&D activity in a profit-making organization, the outcomes are accomplishments such as cost reduction and sales improvement (Brown and Svenson 1988). Thus, this study uses profit, measured as earning before interest and tax, to proxy financial performance.
R&D output
R&D outputs are intangible assets that can be labeled as the firm’s “knowledge stock” (Hall, Jaffe, and Trajtenberg 2005). Empirical testing requires an observable proxy for R&D outputs. Pakes and Griliches (1984) emphasize that there is quite a strong relationship between the number of patents and R&D, indicating that patents are good indicators of unobserved R&D outputs. There is also a considerable amount of studies using the number of a firm’s patents as a major performance indicator (e.g.
Scherer 1982; Griliches 1984; Jaffe 1986; Ernst 2001; Bottazzi and Peri 2003; Feeny and Rogers 2003; Hagedoorn and Cloodt 2003; Lin and Chen 2005; Tsai and Wang 2005). Therefore, in this study I take the number of patents as an indicator of R&D output quantity.
The value of patent counts as a proxy for R&D output is limited by the very large variability in the importance of individual patents, rendering patent counts as a noisy indicator of R&D outputs (e.g. Bottazzi and Peri 2003; Hagedoorn and Cloodt 2003;
Hall et al. 2005). Griliches (1990) also points out that the inventions that are patented differ greatly in ‘quality’. According to Hall et al. (2005), an increase of one citation per patent is related to an increase of 3—4% in market value. Therefore, more valuable patents are cited more frequently. In addition, claims are the parts of a patent that define the boundaries and legal basis of patent protection. Having more claims allows firms to have legal title to different aspects of invention. Several prior literatures (e.g. Trajtenberg 1989; Deng, Lev, and Narin 1999; Harhoff, Narin, Scherer, and Vopel 1999; Hirschey, Richardson, and Scholz 2001; Lanjouw and Schankerman 2004; Scotchmer 2005) also indicate that patents with higher average citation and
number of citations and claims as an indicator of R&D output quality.
I heed Balkin, Markman, and Gomez-Mejia’s (2000) recommendation that a composite measure be used to capture broad aspects of innovation activities more accurately. In addition, the correlation between the number of patents, citations, and claims is quite high (from 0.7846 to 0.9622), indicating several problems with multicollinearity, so I use Principal Component Analysis (PCA) to generate the single factor of R&D outputs which explains 94.45% of the observed variation.
R&D investment
The level of R&D investments is the most extensively used proxy for the level of innovative effort. Its advantages are that it is a relatively well understood term and it provides a dollar figure for use in analysis (Rogers 1998). In accordance with previous research (e.g. Shrader 2001; Hall and Bagchi-Sen 2002; Sakakibara 2002;
Hernan et al. 2003; Negassi 2004; Huang and Liu 2005; Lin and Chen 2005; Liu, Lin, and Chin 2005; Yu,Chiao, and Chen 2005), this study adopts R&D expenditures to proxy R&D investments.
Independent Variables
R&D cooperation type (Vertical cooperation, Horizontal cooperation, Generalized cooperation, R&D competition)
Survey data are the primary sources for the prior R&D cooperation research (e.g.
Kaiser 2002a; Caloghirou et al. 2003; Chang 2003; Belderbos et al. 2004). According to Hagedoorn and Schakenraad (1994), in this study I use archival data and content analysis from 1998 to 2001 to measure the variable of R&D cooperation. Four sources were used to measure each company’s R&D collaboration. First, I collect collaboration data from the database of the Industrial Technology Development Alliance Program (ITDAP) of the Ministry of Economic Affairs, Department of Industrial Technology in Taiwan. Second, according to the database of USPTO, if the company has co-assignee(s) for the same patent rights, then I identify these assignees with R&D collaboration relationships. Third, companies that belong to the same business group are regarded as R&D collaborative companies.21
21 This assumption may not hold because companies that belong to the same business group are not necessary to cooperation on R&D. Therefore, I also exclude this data source and to examine the
Finally, I collect collaboration related news from DigiTimes daily. DigiTimes daily consists of sections devoted to news concerning computers and peripherals, semiconductors, optoelectronics, IT, communications, networking and software, and is the most professional and popular newspaper in Taiwan’s high-technology industry.22 The following key words are used to obtain R&D collaboration and strategy alliance data: collaboration: strategy alliance, joint development, joint research, research alliance, strategic collaboration, and cooperative alliance, etc. There were 30,258 related news items during 1998-2001. Then, I analyze and code the news that relates to my sample companies (604 listed companies in Taiwan’s high technology industry). Finally, to make sure that the classification procedure is reliable, I invited two specialists (Scott Lin, the manager of DigiTime daily, and Walter Huang, the business director of ZuKen Taiwan Inc.) to review the code process. I then revised the coding text according to their suggestions.
I use the following principle for constructing the database of R&D cooperation:
Firstly, I exclude the R&D cooperation news with possibility. For example,
“…A company indicates that it will not rule out the possibility of integrating vertically with other key computer component companies. To enhance market competition ability, the related R&D cooperation or strategic alliances are all under discussion. It is still not the right time to make a public announcement….” (April 17, 2001)
Secondly, news content can be used to determine the types of R&D cooperation, including vertical R&D cooperation (cooperation between suppliers or customers), horizontal R&D cooperation (cooperation with competitors), and generalized R&D cooperation (cooperation with competitors and vertical industries simultaneously). For example, the following news item is regarded as vertical R&D cooperation:
“Lucent and Winbond have agreed to jointly develop stand-alone and embedded flash memory products using CHISEL for a period of two years. It is expected that both Lucent and Winbond will offer products
after excluding the data of business group.
22 According to my interview with several practitioners, e.g. Victor Tsan, the general director of Market Intelligence Center, Institute of Information Industry, and Walter Huang, the business direct of Zuken Taiwan Inc, they all agree that DigiTimes is the most reliable professional newspaper in Taiwan
that use the CHISEL technology.” (December 13, 1998)
The following news item is regarded as horizontal R&D cooperation:
“Accton yesterday confirmed that it will invest about US$15 million in a joint venture with one large American network company to develop an internet audio chip….” (December 6, 1999)
If the same company has both vertical R&D cooperation and horizontal R&D cooperation during the same year, then I define it as generalized R&D cooperation. In addition, I exclude duplicate news, and also consistently apply a single classification principle.
R&D cooperation intensity
Most of the research uses a dummy variable to proxy R&D cooperation (e.g.
Hagedoorn and Schakenraad 1994; Shrader 2001; Sakakibara 2002; Caloghirou et al.
2003; Belderbos et al. 2004). However, dummy variables cannot represent the frequency and intensity of R&D cooperation for each firm. Therefore, I measure R&D cooperative activity in two ways. The first measure, R&D cooperation type, focuses on the event of cooperation formation, that is, whether a firm engaged in cooperation or not during a given period. Based on prior literature (e.g. Belderbos et al. 2004), the R&D cooperation type variable is taken as 1 indicating that a firm has at least one vertical R&D cooperation (cooperation between suppliers or customers), horizontal R&D cooperation (cooperation with competitors), or generalized R&D cooperation (cooperation with competitors and vertical industries simultaneously) during 1998-2001, and 0 otherwise. No cooperation (R&D competition) is treated as a reference variable. The second variable, R&D cooperation intensity, measures the total amount of cooperation undertaken by a firm during a given period, which does reflect the intensity of cooperative activity (Park, Chen, and Gallagher 2002). I follow Stuart’s (2000) and Park et al.’s (2002) approach and use the number of R&D cooperation formed by a firm during 1998-2001 to proxy R&D cooperation intensity.
Absorptive capacity
Zahra and George (2002) highlight four distinct but complementary capabilities that compose a firm’s absorptive capacity: acquisition, assimilation, transformation, and exploitation. To account for differing abilities of firms to internalize other firms’
including the share of employees with Ph. D. and master’s degrees to proxy absorptive capacity.
Knowledge spillover
The earliest and simplest formulation of firm i’s knowledge spillovers is given by:
∑
≠= N
i j
j
i RD
SP ,
where SPi is the level of spillovers enjoyed by firm i; RDj is the investments in R&D by firm j; and N denotes the number of firms inside firm i’s industry (Kaiser 2002b).
However, it is not necessary that every firm can gain from other firms’ R&D investments. In this study, I argue that the strategy alliance usually includes knowledge sharing and technique exchange with each other. Therefore, knowledge spillovers of firm i are higher if the number of strategy alliance (including sales alliance, production alliance, and joint venture) inside or outside firm i’s industry (horizontal alliance or vertical alliance) is higher:24,25
∑
≠= N
i j
j
i SA
SP
WhereSA is the number of strategy alliance by firm j. j Uncertainty
Uncertainty is the degree of accuracy with which one can predict the future.
Where there is less variance, there is more certainty (Tosi, Aldag, Storey 1973).
Former literatures use standard deviation or coefficient of variation to measure uncertainty (e.g. Tosi et al 1973; Snyder and Glueck 1982; Kothari 2002). However, these measures do not consider the ordering of the data points and measure only their dispersion from the mean. The measures are unable to detect variation from a time trend. Hence, regression approach is superior to the above measurements. In this study, I apply Dess and Beard’s (1984) approach to measuring uncertainty which is obtained when each dependent variable (sales, employees, and R&D and capital expenditures)
25 I have separated knowledge spillovers into vertical spillovers (the number of strategy alliance with suppliers and buyers) and horizontal spillovers (the number of strategy alliance with competitors).
(44)
(28)
is regressed on time over the period 1998-2001. Four volatility measures are calculated: the standard error of the regression slope coefficient of sales divided by mean value of sales is used as a measure of market volatility; the standard error of the regression slope coefficient of earning before income and tax (EBIT) divided by mean value of EBIT is used as a measure of profit volatility; the standard error of the regression slope coefficient of employees divided by mean value of employees is used as a measure of employment volatility; the standard error of the regression slope coefficient of R&D and capital expenditures divided by mean value of the R&D and capital expenditures is used as a measure of technology volatility. All of these four can refer to market uncertainty. Finally, I use Principal Component Analysis (PCA) to extract the single factor of uncertainty. The principal component derived form PCA explains 73.66% of the observed variation.
Control Variables
To avoid the impact caused by other variables that are absent from my model, this study refers to prior research and chooses firm characteristics (including sales growth, capital structure, and firm size), industry characteristics (industry segments and industry effect) as control variables.
Sales growth
Higher sales means that the profitability of a firm is better. Therefore, I measure sales growth as the change in sales revenue to this period from last period and scale it by net sales revenue from the last period (e.g. Capon, Farley, and Hoenig 1990; Yu et al. 2005; Huang and Liu 2005).
Capital structure
Capital structure reflects the operation risk of a firm and is deemed as the important decisive factor of financial performance. Therefore, I use the ratio of total liabilities to total assets to proxy capital structure (e.g. Capon et al. 1990; Said et al.
2003; Yu et al. 2005; Huang and Liu 2005).
Firm size
The effect of firm size on innovation is tied to the relative advantage of large/small firms during the process of innovation (Mazzucato 2000). Acs and Audretsch (1987) find that large firms tend to hold a relative innovative advantage in
a differentiated good, while small firms tend to own a relative advantage in industries that are highly innovative and utilize a large component of skilled labor. Research also points out that firm size may have an influence on performance (e.g. Ittner and Larker 1997; Bharadwaj, Bharadwaj, and Konsynski 1999). Therefore, consistent with prior literature, this study uses total asset to proxy firm size (e.g. Mazzucato 2000; Shrader 2001; Kaiser 2002a; Hernan et al. 2003; Matusik and Heeley 2005; Tsai 2005; Tsai and Wang 2005).
Industry segments (Upstream, Midstream, and Downstream)
I divide high-technology industry into three segments, including upstream, midstream, and downstream industry.26 The classification criteria are according to the high-technology industry reports issued by the Industrial Technology Research Institute (ITRI) and “Electronic Industry Connection Encyclopedia” issued by Get-Fortune Publishing Ltd. Industry segment variables are dummy variables taking the value of one if the firm is in upstream and midstream industry, respectively.
Downstream industry is treated as a reference variable.
Industry effect (Semiconductor, Optoelectronics, Telecommunications, Computer component, Computer peripheral, and system and equipment)
Prior literature suggests that sectoral differences can play a role in explaining the various outcomes of innovative performance (e.g. Griliches 1998; Ernst 2001;
Hagedoorn and Cloodt 2003). Therefore, to control for industry effects on R&D investments, R&D outputs, and financial performance, I use the industry effect measured as dummy variables. Taiwan’s high-technology industries include semiconductor, optoelectronics, telecommunications, computer component, computer peripheral, and system and equipment.27 Thus, industry effect variables are dummy variables taking the value of one if the firm is in optoelectronics, telecommunications, computer component, computer peripheral, and system and equipment industries, respectively. The semiconductor industry is treated as a reference group.
See Table 11 for the variable measurements of this study.
26 In Taiwan, most of the upstream firms are raw materiel and design companies, while most of the manufacturing companies belong to midstream industry. Downstream industry includes application industry that is much closer to final customers. For example, the level of labor division in IC design (upstream), IC manufacturing (midstream), and IC testing and package (downstream) specialization for
Table 11: Variable measurements for this study
Variables Variable Measurement
Dependent Variable
Profit Earnings before interest and tax R&D investment R&D expenditures
R&D output The number of patents is used to measure R&D output quantity. The number of citations and claims is used to measure R&D output quality. Principal Component Analysis (PCA) is used to extract the single factor of R&D outputs.
Independent Variables
R&D cooperation type28 The R&D cooperation type variable is taken as 1 if a firm has at least one vertical R&D cooperation (cooperation between suppliers or customers), horizontal R&D cooperation (cooperation with competitors), or generalized R&D cooperation (cooperation with competitors and vertical industries simultaneously) during 1998-2001, and 0 otherwise.
R&D cooperation intensity Total number of R&D cooperation formed by a firm during 1998-2001.
Absorptive capacity The ratio of Ph. D. and master degree employees to total employees.
Knowledge spillover
∑
≠= N
i j
j
i SA
SP
where SPi is the level of spillovers enjoyed by firm i;
SA is the number of strategy alliances by firm j; N j
denotes the number of firms inside firm i’s industry.
Variables Variable Measurement Uncertainty Standard error of the regression slope coefficient
divided by mean value (sales, profits, employees, and the sum of R&D expenditures and capital investment). Principal Component Analysis (PCA) is used to extract the single factor of uncertainty.
Control Variables Firm characteristic
Sales growth (Net operating sales this period – net operating sales last period) / Net operating sales last period
Capital structure Total Liabilities / Total assets
Firm size Total assets
Industry characteristic Industry segment29
Upstream Upstream industry = 1; others = 0.
Midstream Midstream industry = 1; others = 0.
Industry effect30
Optoelectronics Optoelectronics industry = 1; others = 0 Telecommunications Telecommunications industry =1; others = 0 Computer component Computer component industry = 1; others = 0 Computer peripheral Computer peripheral industry =1; others = 0 System and equipment Other industry = 1; others = 0