非参数面板计数数据模型估计量的比较

Comparison of Estimators Under Nonparametric Panel Count Data Model

文章构建了面板计数数据模型的伪似然和全似然函数,发现在这两种似然下,对模型中非负的基线函数本身和其对数形式分别用样条估计而得到的估计量的准确性及计算时间存在很大区别;发现在编写较复杂的全似然函数代码时,通过对所有个体的样条基函数矩阵进行转换处理能够避免重复运算,从而大大减少计算时间。伪似然下对基线函数的对数用样条估计而得到的估计量最好。该方法应用到小儿哮喘研究后的结果揭示了IL5的非参数效应。

This paper constructs the pseudo-likelihood and full-likelihood functions under a nonparametric panel count data model, finding that there is a great difference in the accuracy and computation time of estimators for the non-negative baseline function and its logarithmic form in the model respectively estimated by spline under the two likelihoods, that the repetitive operation can be avoided by converting all individual spline basis function matrices when writing complex code of full likelihood functions, thus greatly reducing the calculation time, that using a spline to estimate the logarithm of the baseline function under pseudo-likelihood gives the best estimator, and that the nonparametric effect of IL5 is revealed when the method is applied to the study of pediatric asthma.