相較於過去分析補習成效的文獻，本文主要有兩項差異，一是增加考慮高中畢業生的「起始能力」。這個變數以高中入學的基測成績表示。二是採用Heckman「兩階段自我選擇的模型估計法」。研究結果顯示家庭背景對於補習參與有正向影響；末考慮自我選擇的OLS估計會高估補習的功效。若針對補習者分析，在不考慮自我選擇的情況下，補習具顯著正向效果。但在控制起始能力後，補習雖仍有成效，然幅度大蝠減弱，顯著性也降低。若同時考慮自我選擇與起始能力，不論線性或允許遞減的補習成效都大幅下降，且都不顯著。最後，我們估算補習者若不補習、以及不補習者若補習的預期成績，結果顯示兩者成績都會退步，顯示兩者都已自我選擇各自較有利的學習途徑。總之，若不考慮自我選擇，不僅會高估補習者的補習效果，亦會誤以為不補習者若補習可提高學測成績。

Unlike existing studies on cram schools, the present study considers both the role of initial ability and corrects for self-selection bias. Initial ability is represented by the score of the basic competence test taken when the student enters senior high school. The present study finds that family background determines participation in cram schools, and the OLS results might overestimate the effect of cram schooling. If initial ability is not controlled, the participation in cram schooling increases the score on the general scholastic ability test (GSAT) for those who participate. After controlling for initial ability in the regression, cram schooling still increase the GSAT, but the magnitudes and degree of significance of the effect substantially decline When both self-selection and initial ability are considered, the effect of cram schooling is not significant, regardless of whether the effect of cram schooling is specified as linear or decreasing. Finally, both students who participate in cram schooling and those who do not earn higher scores on GSAT than if they had chosen differently. Without considering self-selection, the effects of cram schooling are overstated, leading to the incorrect conclusion that students who do not participate in cram schooling would enhance their GSAT score if they participated.