Group testing is an efficient method for testing a large population to detect infected individuals. In this paper, we consider an efficient adaptive two stage group testing scheme. Using a straightforward analysis, we characterize the efficiency of several two stage group testing algorithms. We determine how to pick the parameters of the tests optimally for three schemes with different types of randomization, and show that the performance of two stage testing depends on the type of randomization employed. Seemingly similar randomization procedures lead to different expected number of tests to detect all infected individuals, we determine what kinds of randomization are necessary to achieve optimal performance. We further show that in the optimal setting, our testing scheme is robust to errors in the input parameters.
On optimal two-stage testing of multiple mediators
在高维环境中的中介分析通常涉及在大量测量变量中识别潜在的中介。为此,最近提出了一种称为ScreenMin的两步家庭错误率程序(Djordjilović et al. 2019)。在ScreenMin中,变量首先被筛选,只有那些通过筛选的变量才被测试。建议的阈值选择已被证明,以保证渐近familywise错误率。在这项工作中,我们研究了选择阈值对有限样本familywise错误率的影响。我们推导出一个功率最大化选择阈值,并表明它可以很好地近似于Wang et al.(2016)的自适应阈值。我们说明了调查程序的病例对照研究检查鱼类摄入量对结直肠腺瘤的风险的影响。
在高维环境中的中介分析通常涉及在大量测量变量中识别潜在的中介。为此,最近提出了一种称为ScreenMin的两步族错误率(FWER)程序(Djordjilović et al. 2019)。在ScreenMin中,变量首先被筛选,只有那些通过筛选的变量才被测试。建议的阈值选择已被证明,以保证渐近FWER。在这项工作中,我们调查的选择阈值上的有限样本FWER的影响。我们推导出功率最大化选择阈值,并表明它很好地近似于Wang et al.(2016)的自适应阈值。我们在模拟研究中研究了所提出的程序的性能,并将其应用于病例对照研究,研究鱼类摄入量对结直肠腺瘤风险的影响。