Decoding was performed separately for three subgroups of MT neurons based on their peak response ratios: lowest third (n=33), middle third (n=32), and highest third (n=32) of peak response ratio values. In addition, decoding was performed for mixed subgroups of the same size (n=32) that were selected randomly from the population. (A) Decoder performance for each subgroup based on peak response ratio. For each subgroup, 500 decoding simulations were performed (gray dots) using different samples of correlated noise, as described in Methods. For the mixed subgroup, each of the 500 simulations also involved drawing (without replacement) a new random subset of 32 neurons from the population. Black filled symbols denote mean proportion correct across the 500 simulations, and error bars represent 95% CIs. (B) Predicted detection probability (DPpred) for each neuron in each subgroup (gray symbols), along with mean DPpred values (black symbols, error bars denote s.e.m). For the mixed subgroup, data are shown for all 97 neurons since each neuron was included in many different subsamplings. In this case, gray data points represent the mean DPpred value for all random subsamplings (165, on average) that included each neuron. Median DPpred was not significantly different from 0.5 for the lowest third subgroup (p=0.59, n=33), was marginally significant for the middle third subgroup (p=0.043, n=32), and was significantly greater than 0.5 for the highest third (p=2.2×10–5, n=32) and mixed (p=3.2×10–5, n=97) subgroups (signed-rank tests comparing median DPpred with 0.5).