Unraveling the determinants of overemployment and underemployment among older workers in Japan: A machine learning approach

Overemployment and underemployment being widely existing phenomena, much less is known about their determinants for older workers. We innovatively employ machine learning methods to pin down important factors driving overemployment and underemployment among older workers in Japan. The results suggested that those with better economic conditions, worse health, less family support, or unfavorable job characteristics are more likely to report overemployment, whereas age, less disposable income, shorter current work hours, temporary nature of the job, and low job and pay satisfaction are predictive to underemployment. K-means Cluster analysis further shows that reasons for being work hour mismatched can be highly heterogeneous within the overemployed and underemployed groups. Subgroup analyses indicate room for pro-work policies among 65+ workers facing financial stress and lacking family support, female workers with unstable jobs and low spousal income, and salaried workers not working enough hours. Our study sheds light on strategies for fully utilizing the human capital of older workers.