Clusters of flowstone ages are not supported by statistical evidence.
Clusters of flowstone ages are not supported by statistical evidence.
复制标题
流石年龄簇没有统计证据支持。
DOI：
10.1038/s41586-021-03586-0
复制
发表时间：
2021
期刊：
影响因子：
64.8
通讯作者：
Hopley P
中科院分区：
文献类型：
--
作者：
Hopley P
Pickering等人1在人类摇篮的29个流石沉积物中确定了6组U-Pb年龄，以证明南非的人类记录仅限于干燥的气候阶段。这六个聚类被识别为核密度估计中的峰值，该估计是使用我们中的一个开发的软件创建的2：不幸的是，该软件没有被适当地使用。核密度估计是数据的描述符，而不是统计工具-它没有参数或分布元素，并且数据在统计意义上不是“拟合”的。因此，核密度估计不能用作将29个分散值的小数据集细分为6个更小的聚类的理由。没有统计学证据表明数据包含一个以上的年龄组成部分，如Pickering等人所提出的。例如，流石数据通过标准Shapiro-Wilk正态性检验（P= 0.4）。我们不主张数据遵循正态分布;在这里，我们使用假设检验来论证，如果数据集太小而不能拒绝正态性，那么它太小而不能证明多模态。通过手动选择不适当的窄内核带宽（30，000年而不是默认的320，000年）和直方图箱宽度，作者已经创建了六个具有可疑科学价值的“幻影峰”。
Pickering et al. 1 identify six clusters of U–Pb dates among 29 flowstone deposits in the Cradle of Humankind to argue that the hominin record of South Africa is restricted to dry climate phases. The six clusters were identified as peaks in a kernel density estimate, which was created using software developed by one of us 2: unfortunately, this software has not been used appropriately. Kernel density estimation is a descriptor of data and not a statistical tool—it has no parametric or distributional element, and the data are not ‘fitted’in a statistical sense. Therefore, a kernel density estimate cannot be used as justification to subdivide the small dataset of 29 dispersed values into 6 even-smaller clusters. There is no statistical evidence that the data contain more than one age component, as was proposed by Pickering et al. 1. For example, the flowstone data pass a standard Shapiro–Wilk test for normality (P= 0.4). We do not claim that the data follow a normal distribution; here we use the hypothesis test to argue that if the dataset is too small to reject normality, then it is too small to prove multimodality. By manually selecting an inappropriately narrow kernel bandwidth (of 30,000 years instead of the default 320,000 years) and histogram bin width, the authors have created six ‘phantom peaks’ of questionable scientific value 3.