Review materials for Tressoldi et al 2014, version 2

Supplementary materials for my review of Tressoldi et al. 2014. "Brain-to-Brain (mind-to-mind) interaction at distance: a confirmatory study" version 2. StimulusTimecourses.png: Stimulus time courses were extracted from coincidences data files. Each row corresponds to the stimulus time course for one of the 20 pairs of participants. Upwards bumps in these plots indicate stimulus (signal) periods, the rest are baseline (silence) periods. The strong correlation between the different stimulus protocols is immediately obvious. For the majority (13) of pairs the first stimulus occurred after one minute. For the remainder (7) the first stimulus occurred after 2 minutes. In addition to the high predictability of these time courses, this plot also demonstrates that the intial silence period was not randomized to be 1, 2, or 3 minutes as claimed in the manuscript, but simply either 1 or 2 minutes. TemporalCorrelations.png: Temporal correlations between raw data randomly sampled into 50% training and 50% testing samples as done in the analysis by Tressoldi et al. Each panel shows a 2D density histogram. Hotter colours indicate a greater frequency of data points clustering in that location. These plots compare the randomly assigned "training samples" (x-axis) to the remaining "testing samples" (y-axis). The four columns correspond to EEG channels 1-4. The four rows are four repetitions in which a new set of training/test samples have been assigned randomly. For most plots the strong correlation between the randomly assigned samples is evident even on visual inspection. Even for the plots that seem to show less of a correlation (e.g. the fourth column) the Pearson's correlation is extremely significant. It should not be surprising that a powerful classifier can exploit such correlations to decode arbitrary stimulus labels provided the stimulus labels aren't too high frequency.本补充材料对应本人对Tressoldi等人2014年发表的《远距离脑-脑（心-心）交互：验证性研究》（版本2）的评审工作。 StimulusTimecourses.png（刺激时程图）：刺激时程数据提取自共现数据文件。每一行对应20对被试中其中一对的刺激时程曲线。图中向上的凸起代表刺激（信号）时段，其余时段均为基线（静默）时段。不同刺激范式间的强相关性一目了然。其中13对被试的首次刺激出现在实验开始1分钟后，剩余7对则在2分钟后出现首次刺激。除了这些时程具备高度可预测性外，该图还揭露了一处与手稿表述不符的细节：手稿声称初始静默时段会在1、2、3分钟范围内随机设置，但实际仅采用了1分钟或2分钟两种固定时长。 TemporalCorrelations.png（时序相关性图）：该图复现了Tressoldi等人的分析流程，即将原始数据随机划分为50%训练集与50%测试集后计算时序相关性。每个子图均为二维密度直方图，颜色越暖代表该区域的数据点聚集频率越高。此类图表将随机分配的“训练样本”（横轴）与剩余的“测试样本”（纵轴）进行对比。四列分别对应脑电图（EEG）通道1至4，四行则代表四次重复实验，每次实验均会重新随机分配训练/测试样本。多数图表中，随机划分的样本间存在强相关性，仅凭目视即可分辨。即便在相关性看似较弱的图表（如第四列）中，皮尔逊相关系数也呈现出极强的统计学显著性。若刺激标签的频率未过高，那么高性能分类器可利用此类相关性对任意刺激标签进行解码，这一结果并不令人意外。