根據Gary Cox提出的「M + l法則」（M表示選區的應選席次數），在單記非讓渡投票制下，每個選區的選票會集中在M+1位候選人身上。但相關的實證結果卻顯示選票並非完全匯集於M+1位候選人身上，而是大多有程度不等的偏離。其主要原因在於，經驗世界中的政黨必須克服協調問題才能提名最適人數並成功配票，而選民則需要足夠的資訊才能決定如何策略性投票。因此，本文認為當理論預期的結果與實際的觀察有所偏差時，並不宜直接推翻理論模型，而應該嘗試尋找能夠解釋偏差的系統性因素以及檢討實證分析中的變數測量是否具有效度。針對台灣1989年至2004年六屆立委選舉的例子，研究結果發現選區規模、政黨提名策略、棄保是否成功、選民做策略性投票所需的資訊是否充分與學習效果都和有效候選人數偏離M+1法則的程度有系統性的關係。也就是說，雖然僅有在四分之一的區域立委選區中，有效候選人數等於應選席次數加一，但M+1法則事實上仍是受到經驗證據支持的。 According to “the M+1 rule” proposed by Gary Cox, under the SNTV system the number of effective candidates tends to be limited into M+1 (M stands for district magnitudes). In fact, empirical studies show that the votes are not necessarily concentrated on M+1 candidates, since assumptions of the M+1 rule may be not always true in the real world. Parties can nominate candidates rationally and distribute the votes equally under certain circumstances, but they sometimes can not overcome the problems of coordination. Voters tend to vote strategically as long as they have perfect information. But the information is by no means costless in the real world. Therefore, when the deviations between the theoretical expectations and the empirical observations occur, it does not necessarily mean that the theory or model is false. The theory still stands true if the deviations can be explained systematically.The dependent variable of the study is the difference between the numbers of effective candidates and the numbers predicted by “the M+1 rule.” Using the aggregate data of 167 districts of the Legislative Yuan Elections in Taiwan from 1989 to 2004, this paper finds that the district magnitude, party nomination strategies, successful vote distributions within parties’ candidates, the quality of voters’ information and the learning effects are systematically correlated with the extent to which the effective numbers of candidates deviate from the M+1 rule. Overall, although the numbers of effective candidates are equal to M+1 in only a quarter of districts, the M+1 rule is supported by the empirical evidence of Taiwan.