672 International Conference On Applied Economics – ICOAE 2010 Figure 1: Conceptual diagram the Independent R&D R&D1 R&D3 the Mixed R&D the Only introducing external outcomes Note: R&D1 indicate that a firm researches and develops on her own. On the other hand, R&D3 indicate that a firm introduces external research outcomes. Figure 2: The diagram of relationship among hypotheses HS : Substitute Relationship × HC : Complement Relationship HS-1 : firm size↑ R&D1↑ HS-1‘ : firm size↑ independent R&D↑ HS-2 : firm size↓ R&D3↑ HS-2‘ : firm size↓ the only introduction of HC-1 = firm size↑ mixed R&D↑ external research outcomes↑

International Conference On Applied Economics – ICOAE 2010 673 4 Data We obtained data through an original questionnaire survey that had been delegated to an investigation company to carry out. Those surveyed comprised 20455 firms registered as Internet monitors throughout the country and covering all industries. In this paper, we used the data of 10731 of these firms (the response rate was 52.5%). Please refer to Saito and Sumikura (2009) for survey details. The consultation period spanned 20 days, from 17 December 2008 to 5 January 2009. Table 1 shows the definitions of variables and descriptive statistics of our data. We instructed the firms, ―We would like you to answer if there are relevant alternatives about your R&D and introduction of technology (multiple answers allowed).‖ We provided the following answers: 「(a) We currently research and develop inhouse.」(R&D1),「(b) We do not currently research and develop in-house, although we did once.」(R&D2),「(c) We currently introduce external research outcomes.」(R&D3),「(d) We do not currently introduce external research outcomes, although we did once.」(R&D4),「(e) We have utilized internal or external research outcomes for own products, processes, and the sale of services in some form.」(R&D5),「(f) We have utilized internal or external research outcomes for our own business in ways other than as products, processes, and the sale of services in some form.」(R&D6)「(g) We have neither researched and developed in-house nor introduced external research outcomes.」(R&D7). 5 Analysis First, we examine the relationship between the R&D in-house and the introduction of external research outcomes to examine whether HS is true or HC is. Here, we tested this using a bivariate probit model based on Green (2003). We suppose that we can describe the following using the latent variable Y , which is unobservable whether i researches and develops in-house now (R&D1); whether i introduces external research outcomes now (R&D3). Yi is supposed to be observable. For example, Yi is 1 if i researches and develops in-house now and is 0 otherwise. Equation (1) describes R&D1 as the dependent variable, and Equation (2) describes R&D3 as the dependent variable. y and * 1 * i * y 2 are latent variables that are unobservable. We X is i‘s attribute vector. This describes * * suppose that we can observe y 1 = 1 if y 1 > 1, and y 0 = 0 otherwise. y 2 is alike, too. i industrial variables, the number of employees, and area variables. β is the parameter. i is the error term. We used a robust standard error to respond to heteroskedasticity. We used the number of employees as proxy for firm size. Then, we used a logarithm of it to respond size of variance. Additionally, we also used the second-order term of ―Employees‖ to observe how the percentage of independent R&D or the introduction of external research outcomes increased as the number of employees increased. We used industry dummy variables (base line: manufacturing business) as control variables to consider the differences between industries. We used prefecture dummy variables (base line: Tokyo) and the number of scientists (logarithm) in each prefecture as area variables. This variable of scientists particularly indicates natural scientists. However, this does not identify whether they belong to private firms or universities/public research institutes. We think that accessibility to external research outcomes is also important when firms decide to introduce external research outcomes or not. Therefore, we introduced the number of scientists in the area to consider whether introducing external research outcomes depends on access to research outcomes. Then, we suppose the error term in equations (1) and (2) below. * ' * y X y 1 if y 0, 0 otherwise,(1) 1 * 2 1 1 1, 1 ' 2 2 2, 2 1 1, X 2 E 2 X1, X 2 y X y if y 0, 0 otherwise,(2) 1 X 0, 1 X1, X 2 Var 2 X1, X 2 1, X , X . E Var Cov 1, 2 1 2 1 * 2 We suppose that Equations (1) and (2) are simultaneously decided if there is a correlation between (1) and (2), that is, π= 0 is rejected. If this is the case, it was a good decision to use a bivariate probit model. Table 2 shows the result. We found that (1) and (2) are simultaneously decided, because π= 0 is rejected in any model. Additionally, we found that both have a positive relationship to the sign of π. Therefore, the R&D on a firm‘s own and the introduction of external research outcomes are not so much substitutive, but complementary. In fact, we support HC while we reject HS. Model [1] only controls the first-order term of ―Employees,‖ and [2] adds the second-order term. In Model [1], the first-order term of ―Employees‖ is positively significant to R&D1. That is, larger firms tend to research and develop in-house. Although we rejected HS itself, we could accept HS-1 regardless of the result. On the other hand, the first-order term of ―Employees‖ is not significant to R&D3. This suggests that the introduction of external research outcomes may not depend on firm size. Then, we cannot accept HS-2. However, controlling the second-order term of ―Employees‖ at the same time in Model [2], the first-order term of ―Employees‖ is negatively significant, and the second-order term of ―Employees‖ is positively significant to R&D1. In fact, this necessarily does not agree that the larger a firm‘s size is, the greater the chance it performs research and development in-house. Rather, the percentage of firms with the R&D in-house decreases until a certain level of employees, and increases beyond that level. That is, small firms and big firms research and develop in-house while medium firms do not. This suggests that there is a non-linear relationship between firm