Checking the Normality Assumption in Multiple Regression with Excel 2007

For our inference procedures (t-test, F-tests, confidence intervals, prediction intervals) of multiple regression to be accurate, we assume the population errors come from a normal curve. Thus, if our residuals look like they come from a normal curve, we believe the population errors do, too. Make a dotplot of standardized residuals if you have fairly small sample sizes (say less than 40 observations), and make a histogram of standardized residuals if you have sample sizes that are large (say at least 40). The more mound-shaped the distribution, the more faith we have that the points came from a normal curve.