户籍歧视导致的收入差距依然存在吗——基于机器学习方法的再讨论  

Does the Income Gap Caused by Hukou Discrimination Still Exist?A Re-discussion Based on Machine Learning Methods

作　　者：江求川[1] 鲁元平[2,3] JIANG Qiuchuan;LU Yuanping

机构地区：[1]郑州大学商学院 [2]中南财经政法大学财政税务学院 [3]中南财经政法大学收入分配与现代财政学科创新引智基地

出　　处：《中国农村经济》2024年第9期101-120,共20页Chinese Rural Economy

基　　金：国家社会科学基金一般项目“共同富裕目标下代际收入流动阻碍因素与提升路径研究”(编号:22BJL074)的资助。

摘　　要：中国劳动力市场供需结构变化促使农民工与城镇职工间收入差距呈现新态势,有关户籍歧视是否依然存在的争论越来越多。本文用双重去偏机器学习方法重新检验农民工与城镇职工收入差距中的户籍歧视现象。经验分析表明:第一,迁移溢价干扰了对户籍歧视的识别,考虑迁移溢价因素后户籍歧视现象更加明显;第二,用双重去偏机器学习方法选择更加符合条件独立性假设要求的模型后,农业户籍对劳动者小时工资收入、全年总收入和全年工资收入均有负面影响,且对小时工资收入的负面影响更为显著;第三,经双重去偏机器学习修正后的Oaxaca-Blinder分解结果表明,农民工和城镇职工收入差距中大约有8%~15%属于户籍歧视;第四,Oster检验证实双重去偏机器学习的估计相较OLS估计更加可靠,不同机器学习算法下的双重去偏机器学习估计与Lewbel工具变量估计也表明本文结论是稳健的。Changes in the supply-demand structure of China's labor market have led to a new trend in the income gap between migrant workers and urban employees,sparking increasing debates on whether Hukou-based discrimination stil exists.This paper re-examines the phenomenon of Hukou-based discrimination in the income gap between migrant workers and urban employees using a doubly debiased machine learning approach.The empirical analysis reveals the following findings.(1)Migration premium interferes with the identification of Hukou-based discrimination,and the phenomenon of Hukou-based discrimination becomes more apparent after accounting for the factor of migration premium.(2)After applying the doubly debiased machine learning method to select models that better meet the conditional independence assumption,the agricultural household registration has a negative impact on the laborers'hourly wage income,the annual total income and the annual wage income,with a more significant negative effect on the hourly wage income.(3)The Oaxaca-Blinder decomposition,corrected by doubly debiased machine learning,indicates that approximately 8%to 15%of the income gap between migrant workers and urban employees can be attributed to Hukou-based discrimination.(4)The Oster test confirms that the estimation of doubly debiased machine learning is more reliable than the OLS estimation,and the doubly debiased machine learning estimation and Lewbel's instrumental variable estimation under different machine learning algorithms also demonstrate the robustness of the conclusions drawn in this paper.

关 键 词：户籍歧视 机器学习 收入差距 农民工 

分 类 号：F241.2[经济管理—劳动经济]