[Usefulness of residuals in clinical research]

The simple linear regression analysis, multiple linear regression and logistic regression constitute powerful statistical analysis tools widely used in clinical research. These kinds of analyses are based upon mathematical models which at the same time are established on certain basic assumptions. The regression analysis assumptions are basically: a) that the model is really linear, b) that the distribution of data is normal (from a statistical point of view), c) that the variances of the employed data are homogeneous (homocedastics) and that the included data are independent. The regression diagnostic has become popular as a form to evaluate if the assumptions have been accomplished, one of its most important techniques is the residual analysis. A residual can be defined as the value which measures the distance between the regression line and the corresponding value of the variable "y". Among these kinds of residuals used to evaluate the assumptions of regression are: the crude residual, the standardized, of student and the jackknife. The most useful among them is the jackknife residual. The usefulness and limitations of the residuals in the evaluation of the regression analysis assumptions are described, basically referring to the identification and handling of extreme values (outliers).