COA development experts in recent years have given thought to the psychometric evaluation of instruments and their ability to detect meaningful differences between patient groups. The scoring of the instruments, however, has received less attention, with various approaches sometimes suggested without a clear preference or justification.  The score is ultimately used for evaluating patient outcomes and treatment efficacy and is what requires validation, so this seems like a significant omission.  We examine the traditionally accepted unweighted summary score approach and compare it to the more complex IRT weighted scoring to evaluate if the gain in precision justifies the increased scoring complexity.  Precision may differ depending on whether the score is close to the mean of the population or closer to the extreme ends of the distribution.  Simulated data are used for this comparison to evaluate if the precision of the scores differs depending on the location of the score and if the instrument is used for group comparisons versus individual diagnosis.  Additionally, we recognize that the reliability of a scale is likely to be variable across the range of its scores.  With that in mind, we consider an approach to comparing mean scores between groups that incorporates the standard error of each individual IRT score into the model.  By using the IRT standard errors, we can adjust for the different levels of uncertainty associated with ranges of scores along the scale, ultimately providing us greater confidence in the group comparison results.