Table 5: Best statistics for each simulation scenario with three replicates. Here “model” corresponds to the true model used for simulation. Statistics were differentiated in their ability to identify the top 1%, 2% and 10% of genes with large signals (labeled Signal) and large signal-to-noise ratios (S/N). The rows labeled “MM” correspond to parameter estimation using the method of moments. Results using the actual true parameter values (labeled “Truth”) are also given. Instances where the best statistic did not matching the appropriate true model and goal are highlighted in red. Instances where the best statistic did not match the model but was consistent with goal are highlighted in blue. The results highlight the importance of matching the statistic to the goal; in only one case did a T statistic out-perform other statistics to identify high signal changes while in no cases did a fold change out-perform other statistics to identify high SNRs.