Two neurons have similar stimulus tuning and their trial-to-trial fluctuations are strongly correlated. Why can this correlation reduce the amount of independent information available to a population decoder?