Specification testing for binary choice model via maximum score
提出一种Hausman型统计量，通过比较最大似然估计和最大得分估计来检验参数二元选择模型的设定错误，模拟显示该检验比传统信息矩阵检验有更好的尺寸性质，并对厚尾分布和异方差等常见设定错误具有合理功效。
This paper proposes a Hausman-type statistic to the test specification of a parametric binary choice model by comparing the maximum likelihood estimator and the maximum score estimator. Although the convergence rates are different, it is still meaningful to compare these estimators to detect misspecification of parametric models. A simulation study illustrates that the proposed test offers better size properties than the conventional information matrix test, and exhibits reasonable power against common forms of misspecification, such as heavy-tailed distributions and heteroskedasticity. • Proposes a Hausman-type specification test for parametric binary choice models. • Detects misspecification by comparing maximum likelihood and maximum score estimators. • Offers better size properties than the conventional information matrix test. • Exhibits reasonable power against heavy-tailed distributions and heteroskedasticity.