The difference between a datum and the value predicted for it by a model. In linear regression of a variable plotted on the vertical axis onto a variable plotted on the horizontal axis, a residual is the "vertical" distance from a datum to the line. Residuals can be positive (if the datum is above the line) or negative (if the datum is below the line). Plots of residuals can reveal computational errors in linear regression, as well as conditions under which linear regression is inappropriate, such as nonlinearity and heteroscedasticity. If linear regression is performed properly, the sum of the residuals from the regression line must be zero; otherwise, there is a computational error somewhere.