臺大管理論叢 NTU Management Review VOL.29 NO.1

117 NTU Management Review Vol. 29 No. 1 Apr. 2019 4.1 Robustness Test We estimate the following equations by using the three-stage least square regression (Zellner and Theil, 1962) for the robustness test for two reasons: (1) some equations contain the endogenous variables; (2) the three-stage least square regression supports the iterated GLS estimation and linear constraints. EPOI = β 11 ME + β 12 TA + Control variables + ε 1 (1) EPOR= β 21 ME + β 22 TA + Control variables + ε 2 (2) ISTK = β 31 EPOI + β 32 EPOR + β 33 IV + β 34 EPOI×IV + β 35 EPOR × IV + ε 3 (3) ESTK = β 41 EPOI + β 42 EPOR + β 43 IV + β 44 EPOI×IV + β 45 EPOR × IV + ε 4 (4) The estimation results are shown in Table 5 and consistent with the results derived by path analysis although there are very slight differences between the results of Hypothesis 3a and 3b estimated by path analysis and by three-stage least square regression. For Hypothesis 3a, we still find that the inclinations of firms to choose the R&D collaborative partners from their internal stakeholders are not intensified; β 31 (0.326, p < 0.001) is still bigger than β 34 (0.004, p > 0.1), and this result is consistent to that of path analysis. For Hypothesis 3b, β 42 (0.401, p < 0.001) is smaller than β 45 (-0.149, p > 0.1), indicating that the inclinations of firms to choose external stakeholders for R&D collaboration are indeed mitigated though β 45 is not statistically significant; this estimation result remains consistent with the estimation by path analysis, supporting Hypothesis 3b. Therefore, we claim that our synthesis framework is robust.