本文研究并比较了基于MMAe赋权的模型平均下,Wald置信区间,MATA置信区间,及Bootstrap置信区间的覆盖率、区间长度、左右错误率等各方面的表现性能。不同信噪比水平下的模拟显示,Wald置信区间在低信噪比时有更好的覆盖率,高信噪比下,三...本文研究并比较了基于MMAe赋权的模型平均下,Wald置信区间,MATA置信区间,及Bootstrap置信区间的覆盖率、区间长度、左右错误率等各方面的表现性能。不同信噪比水平下的模拟显示,Wald置信区间在低信噪比时有更好的覆盖率,高信噪比下,三者覆盖率相近,MATA置信区间相对长度更短。在与MMA等其他模型平均方法的横向比较中,MMAe的赋权在Wald、MATA及Bootstrap三种置信区间的构建下,均比其他赋权方式在更低的样本量下更早达到名义覆盖率。最后在实例中考察MMAe在不同置信区间下的表现,与模拟表现一致。This article studies and compares the performance of models based on MMAe weighting in terms of coverage, interval length, left and right error rates under average, Wald confidence interval, MATA confidence interval, and Bootstrap confidence interval. Simulations at different levels of signal show that the Wald confidence interval has better coverage at low signal, while at high signal, the three’s confidence interval has similar coverage, and the MATA confidence interval is relatively shorter in length. In the horizontal comparison with other model averaging methods such as MMA, the weighting of MMAe reached nominal coverage earlier than other weighting methods at lower sample sizes under the construction of Wald, MATA, and Bootstrap confidence intervals. Finally, the performance of MMAe at different confidence intervals was examined in the example, which was consistent with the simulation results.展开更多
文摘本文研究并比较了基于MMAe赋权的模型平均下,Wald置信区间,MATA置信区间,及Bootstrap置信区间的覆盖率、区间长度、左右错误率等各方面的表现性能。不同信噪比水平下的模拟显示,Wald置信区间在低信噪比时有更好的覆盖率,高信噪比下,三者覆盖率相近,MATA置信区间相对长度更短。在与MMA等其他模型平均方法的横向比较中,MMAe的赋权在Wald、MATA及Bootstrap三种置信区间的构建下,均比其他赋权方式在更低的样本量下更早达到名义覆盖率。最后在实例中考察MMAe在不同置信区间下的表现,与模拟表现一致。This article studies and compares the performance of models based on MMAe weighting in terms of coverage, interval length, left and right error rates under average, Wald confidence interval, MATA confidence interval, and Bootstrap confidence interval. Simulations at different levels of signal show that the Wald confidence interval has better coverage at low signal, while at high signal, the three’s confidence interval has similar coverage, and the MATA confidence interval is relatively shorter in length. In the horizontal comparison with other model averaging methods such as MMA, the weighting of MMAe reached nominal coverage earlier than other weighting methods at lower sample sizes under the construction of Wald, MATA, and Bootstrap confidence intervals. Finally, the performance of MMAe at different confidence intervals was examined in the example, which was consistent with the simulation results.
