In this study, poverty rate of Turkey on 12 statistical regions (NUTS – 1 level) and some determinants of this rate is modeled by a linear regression model. Average household size, unemployment rate, high school and university enrollment rates, median income and urbanization rate as determinants of poverty rate are used as explanatory variables of this model. It is observed that the ordinary least squares (OLS) produce unstable estimates since the design matrix X is subject to strong multicollinearity. In order to obtain stabilized parameter estimates, two biased estimation methods known in the literature, namely Ridge regression and generalized maximum entropy (GME), are used. Inequality and sign constraints that are required in the context of economic theory are used for the GME estimator. Estimators are compared by their efficiency with the estimated mean squared error values obtained by the bootstrap method.
Generalized maximum entropy Least squares Ridge regression Multicollinearity Bootstrap
Bölüm | Makaleler |
---|---|
Yazarlar | |
Yayımlanma Tarihi | 31 Ekim 2015 |
Gönderilme Tarihi | 16 Kasım 2017 |
Yayımlandığı Sayı | Yıl 2015 Cilt: 24 Sayı: 2 |