本文考慮設限資料的四種迴歸模型，並探討模型設定的涵義 （條件），模型的估 計方法，模型的比較與選擇方法。 我們採用預測能力的準則如平均平方誤差（ MSE ）及平 均絕對誤差（ MAE ），並考慮 Leung and Yu (1996) 的觀點來選擇模型。並以臺灣電子電 器業 R&D 資料進行實證應用與分析。 本文的主要結論分為二部份，一為計量模型之選擇， 結論為雙元方程式較單元方程式為佳， 而雙元方程式又以調正的 Tobit 模型較佳。 二為 R&D 模型之實證結果， 主要結論指出 R&D 密度受到廠商規模、集中度、外銷比重等因素的 影響，且驗證熊彼得的理論：「寡佔市場的廠商較具 R&D 活動的能力。」

　　　　　In this paper we consider four models for censored data and discuss model specification, estimation, comparison, and selection. In particular, we adopt MSE and MAE together with the method of Leung and Yu (1996) to select a proper model. These models are then used to analyze Taiwan R&D data. The main findings of the study are separated into two parts. The first part is about model selection. The empirical results show that the adjusted Tobit model are superior to the other three models. The second part concerns the determinants of R&D intensity. The main conclusions are that R&D activities are affected by firm size, concentration, and exports. Meanwhile, the empirical results confirm Schumpter's hypothesis, "Monopoly firms have high R&D intensity."