台灣高中職學生打工行為對於學業成就之影響－工具變數法之應用

Abstract

本文欲探討台灣高中職(五專)學生於高中時期的打工行為，對於其大學入學考試成績的影響。打工行為是人力資本累積的途徑之一，長期而言，人力資本的累積對於未來的薪資平均而言存在正向的報酬；然而，在台灣的升學制度下，多數高中職(五專)學生必須面臨各種形式的大學入學考試以當作分發大學校系的標準，因此，在時間稟賦的限制下，在投入打工時間的同時，也必須犧牲相對應的學習課業時間，由此觀點可知，在學打工行為雖然能夠累積人力資本，卻也需付出一定程度的學業機會成本。 然而，在學時期打工決策的選擇存在內生性問題，這是因為個人的能力存在異質性，導致誤差項與解釋變數相關，因此，若不考慮其內生性，估計出來的係數將不具有一致性進而產生偏誤。本文利用中央研究院社會學研究所的年度計劃──「青少年成長歷程研究」為資料來源，並嘗試利用兩階段工具變數法修正其內生性問題，以獲得理論上的一致性估計量。 研究結果顯示，於第一階段的估計中，不論是採用受訪者的學校型態(IV1)、受訪者居住地的職業學校家數(IV2)、受訪者家中子女數目(IV3)、受訪者有無可支配的零用錢(IV4)當作工具變數，除了IV4對於內生變數(打工決策)不具有解釋力外，其餘皆具有顯著的解釋力；於第二階段的估計中，與不考慮內生性的OLS法互相比較得知，不考慮內生性的OLS模型低估了打工決策對於學業成就的負面影響，意即在考慮內生性下的工具變數法模型，其打工決策對於學業成就存在更嚴重的負面影響。 最後，本文透過第一階段弱工具變數檢定的F值、過度認定限制檢定的P值以及第二階段的調正後判定係數(Adj R-squared)當作篩選標準，在各種不同工具變數的組合下，獲得最佳的工具變數組合IV1+IV2。

The objective of this study is to investigate the effects of Taiwanese high school students’ part-time job on their performance of the college entrance examination. Taking a part-time job is one of the best ways for students to accumulate human capital and it is beneficial for their future salary in the long run. However, current Taiwanese education system puts massive pressure on high school students, who have to invest most of their time in study in order to gain admission to an ideal college. Time, accordingly, is the most valuable asset for high school students but working part-time significantly reduces their study time. As a result, a loss of study time becomes the student’s major opportunity cost of working a part-time job. Nevertheless, unobserved individual ability heterogeneity could lead to endogeneity, and by neglecting the endogeneity, the inconsistency of the estimated coefficient would result in bias. This research is based on the Annual Taiwan Youth Project, conducted by the Institute of Social Sciences of the Academia Sinica. Moreover, a two-stage instrumental variable approach was implemented to correct the endogeneity and obtain a consistent estimator, theoretically. The result of first stage indicates that, except whether survey participant has disposable income (IV4), the following instrumental variables - school type of survey participants (IV1), number of vocational schools on survey participant’s location (IV2) and number of siblings of survey participants (IV3) - have significant influence on endogeneity. The second stage suggests that the negative effect of doing part-time jobs on academic performance is underestimated by the OLS model, without considering endogeneity. In other words, doing part-time job causes a larger significant negative effect on academic performance when we use an instrumental variable model, considering endogeneity. Last but not least, by using a weak instrument test, an over-identification test and adjusted R-squared as filters, the best instrumental variable combination, IV1+IV2, was acquired.