Circular iference in neuroscience – the dangers of 在神经科学–危险的循环推理.pptVIP

Suggestions Design study to have enough power to detect realistic correlations. (Need either anatomical restrictions or large numbers of subjects.) Consider studying trial-to-trial rather than subject-to-subject effects. Correct for multiple testing to avoid false positives. Avoid circularity: Use leave-one-subject out procedure to estimate regional cross-subject correlations. Report correlation estimates with error bars. * %% control variables nCond=7; % number of conditions nDataPerCond=1; % number of data points per condition nChannels=500; % number of channels (e.g. voxels or single cells) * Fig. 2 | The distribution of correlations found in social neuroscience is plausible, because small sample sizes yield large error margins. (a) If we had data for the entire population, we could determine the true population correlation exactly (i.e. down to measurement noise). An estimate from a limited sample, however, will vary depending on the particular points (e.g. subjects) sampled. This variability is larger for smaller samples. (For a sample of 2 points, the estimated correlation will be either 1 or -1 depending on whether the two points differ in the same or in opposite directions in the two variables correlated.) Here we simulated the probability distribution of sample correlations for each population correlation (-1 to 1). Two variables (which could correspond to brain and behavior) were simulated by adding a common Gaussian signal to two Gaussian noise vectors, while weighting signal and noise so as to exactly match a given correlation coefficient for a large sample. In order to obtain an estimate of the distribution of sample correlations, the data were repeatedly sampled at 16 randomly selected points (corresponding to 16 subjects). The gray surface shows the joint probability density function for population and sample correlations. The light blue lines show the marginal probability densities. The red line (when scaled to have an integral of 1) repre