名人堂入選模型之建立與其應用──以美國職棒與日本職棒為例

The establishment and application of a Hall of Fame induction model: Evidence from Major League Baseball and Nippon Professional Baseball

緒論：研究目的在透過美日職棒於2000至2024年名人堂票選結果建立預測模型，並將美日模型相互預測，再對台灣棒球名人堂的票選結果做檢測。方法：利用羅吉斯迴歸，以是否入選為依變項，生涯長度等七大類數據為自變項。結果：研究發現，所建立之美國與日本入選模型的正確分類率分別為98.48 %與97.79 %，皆有非常高的分類結果，相互預測的正確分類率平均僅降1.31 %，優於過往研究與從未對數據權重交代的Bill James名人堂監測器(HOF Monitor)。生涯選入明星賽與獎項累積次數兩變數為顯著預測變數，推論兩變數皆可反映球員當季的表現，為攻守數據的良好代理變數。中華職棒部分，張泰山等四人以美國名人堂預測會入選，張泰山以日本野球殿堂預測亦會入選。名門在本研究預測模型中並非顯著變數，顯示棒球名人堂的票選結果不受名門球隊有較高的曝光度影響，球員的入選主要還是依據優異的生涯表現作為評比標準。結論：本研究所建立之模型無論對於樣本或是未來的延續、國家間的相互預測與對臺灣的檢測皆有不凡的正確分類率，顯見本研究之穩定性、交互適用性與可延續性。若逐年迭代更新球員資料，將是一穩定可靠、容易操作且可長期使用的預測模型。於未來研究中，可以進一步探討影響棒球名人堂入選之其他不可量化之因素，或是加入賽伯計量學的進階數據，並且將模型之建立延伸至投手，涵蓋至美日臺所有棒球選手的入選預測。

Introduction: This study aimed to establish predictive models for the Hall of Fame voting results for Major League Baseball (MLB) and Nippon Professional Baseball (NPB) between 2000 and 2024 and to apply them to predict and examine the results of the Hall of Fame for Chinese Professional Baseball League (CPBL). Methods: In logistic regression models, we used Hall of Fame induction as the dependent variable and career length and six other career metrics as independent variables. Results: The prediction accuracy of the MLB and NPB models was 98.48% and 97.79%, respectively, both demonstrating high classification accuracy. When the models were cross-applied for prediction, the mean accuracy dropped by only 1.31%, outperforming previous studies and the Hall of Fame Monitor by Bill James, which does not disclose its data weighting. Among the variables, the number of All- Star appearances and the total awards were significant predictors, suggesting that these variables reflect a player’s performance during the season and serve as good proxy variables for offensive and defensive statistics. For the Chinese Professional Baseball League (CPBL), the MLB model predicted that four players, including Tai-San Chang, would be inducted, while the NPB model also predicted that Tai-San Chang would be inducted. Being on a prestigious team was not a significant variable in these prediction models, indicating that the Hall of Fame voting results are not influenced by the higher visibility of players from prestigious teams; instead, a player’s induction is primarily based on outstanding career performance. Conclusion: The models established in this study demonstrated exceptional classification accuracy for both the sample data and future predictions and cross-country predictions and assessments in CPBL, underscoring their stability, cross-applicability, and reliability. By updating player data annually, these models prove stable, reliable, easy to operate, and suitable for long-term use in prediction. Future research could explore other non-quantifiable factors influencing Hall of Fame induction, incorporate advanced sabermetrics, and extend the models to include pitchers, covering Hall of Fame predictions for players from MLB, NPB, and CPBL.