ppt课件-nonlinearcorrelation-andersonfive.pptVIP

Correlation with a Non - Linear Emphasis Day 2 Correlation measures the strength of the linear association between 2 quantitative variables. Before you use correlation, you must check several conditions: Quantitative Variables Condition: Are both variables quantitative? Straight Enough Condition: Is the form of the scatterplot straight enough that a linear relationship makes sense? If the relationship is not linear, the correlation will be misleading. Outlier Condition: Outliers can distort the correlation dramatically. If an outlier is present it is often good to report the correlation with and without that point. A hidden variable that stands behind a relationship and determines it by simultaneously affecting the other two variables is called a lurking (confounding) variable. Scatterplots and correlation coefficients NEVER prove causation. Don’t ever assume the relationship is linear just because the correlation coefficient is high. In order to determine whether a relationship is linear or not linear, we must always look at the residual plot. Residuals A residual is the vertical distance between a data point and the graph of a regression equation. The Residual is positive if the data point is above the graph. negative if the data point is below the graph. Is 0 only when the graph passes through the data point. What should you look for to tell if it is not linear?...... Sometimes a high “r” value for linear regression is deceptive. You must look at the scatter plot AND you must look at the residual pattern it makes. If the residuals have a curved pattern then it is NOT linear. To prove linearity A scatterplot of the residuals vs. the x-values should be the most boring scatterplot you’ve ever seen. It shouldn’t have any interesting features, like a direction or a shape. It should stretch horizontally, with about the same amount of scatter throughout. It should show no bends. It should show no outliers. Some Non Linear Regression Shapes…… Positive Qu