We present a method for calculating confidence intervals around error consistency values and revisit earlier work using this metric, showing that many differences found between DNN models of vision are statistically insignificant.