Effects of the housing price to income ratio on tenure choice in Taiwan: forecasting performance of the hierarchical generalized linear model and traditional binary logistic regression model

• Published in: Journal of housing and the built environment 2018-12, Vol.33 (4), p.675-694
• Main Authors: Lee, Chun-Chang, Liang, Chih-Min, Chen, Jian-Zheng, Tung, Cheng-Huang
• Format: Article
• Language: English
• Subjects: Academic achievement; Cities; Construction planning; Decisions; Empirical analysis; Forecasting; Generalized linear models; Geography; Households; Housing; Human Geography; Income; Landscape/Regional and Urban Planning; Leasing; Linear analysis; Loans; Meaning; Regression analysis; Regression models; Reliability; Reliability analysis; Robustness; Robustness (mathematics); Social Sciences; Statistical models; Tenure; Validity
• ISSN: 1566-4910; 1573-7772
• DOI: 10.1007/s10901-017-9572-3

This study examined factors that influence the tenure choices of households in different counties and cities of Taiwan. Data collected in the Housing Status Survey by the Construction and Planning Agency of the Ministry of the Interior were analyzed using hierarchical generalized linear modeling (HGLM). The study designated the household sector as a unit at level 1 and counties and cities as a unit at level 2, with the difference among the counties and cities accounting for 9% of the total variation in rental and purchase decisions. Based on the empirical results, tenure choice was positively and significantly affected by such level-1 factors as gender, age, educational level, area per capita, number of rooms per capita, private loans, and permanent income. The level-2 attribute variable, the housing price to income ratio, had a significant negative effect on tenure choice; a higher ratio of housing price to income resulted in a higher preference among households toward leasing in their lease-or-buy decisions. With regard to the forecast ability comparison, the hit rate of HGLM (90.10%) was higher than that of the binary logistic regression model (87.26%). In terms of the forecasting accuracy evaluated using four measures of association, HGLM outperformed the traditional binary logistic regression model. Based on tenfold cross-validation, HGLM also showed a better hit rate than the traditional binary logistic regression model, meaning that the evaluation results had both robustness and reliability.