Linear Models - Lecture 18 - Diagnostics for Predictor and Residuals

Описание: In this video, we introduce to the viewer several informal methods to study to appropriateness or aptness of a regression model to a given data set. These methods are primarily based and rely on residual analysis, an idea presented in the first few minutes of this lecture. This informal graph approach with residual analysis is sufficient to detect many practically common departures of the normal error model's assumptions. Each type of departure: Nonlinearity, Nonconstancy of errors, Nonindependence of errors, Nonnormality of errors, the presence of outliers, and Omission of important predictor variables, are all and each discussed in turn in this lecture.
One last important remark is in order: This lecture enables the viewer to intuit and deduct several common departures from normality of the simple linear regression model with normal errors in practical applications, and it should primarily serve to do just that, namely give intuition, no more no less.