詹姆斯·R·埃文斯-数据、模型与决策（第4版）evans_sda4_ppt_06-b.pptVIP

Chapter 6: Regression AnalysisPart 2: Multiple Regression6(2)-1Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall Model FormMultiple linear regression model:Y = b0 + b1 X1 + b2 X2 + ... + bk Xk + ePredicted model:Y = b0 + b1X1 + b2 X2 + ... + bk XkThe b’s are called partial regression coefficients. 6(2)-2Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall Example: Colleges and Universities Data6(2)-3Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall Interpreting ResultsRegression statistics similar to single independent variable caseR Square (coefficient of multiple determination)The value .534 indicates that about 53% of the variation in graduation rate can be explained by the variation in the independent variables.Adjusted R2 accounts for sample size and number of independent variables. It is useful for comparing models with different sets of independent variables.6(2)-4Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall ANOVA ResultsSignificance of regressionH0: b1 = b2 =… = bk = 0H1: at least one bj is not 0Note: df for residual is n – k – 1; df for regression is k 6(2)-5Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall Example Residual Plot6(2)-6Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall Test for Individual Coefficients?H0: bj = 0 vs. H1: bj ? 0 t = bj/standard error, with n – k – 1 dfConfidence intervals: bj ? tn-k-1 s.e.6(2)-7Copyright ? 2010 Pearson Education, Inc. ?Publishing as Prentice Hall MulticollinearityMulticollinearity – when two or more independent variables contain high levels of the same information.The independent variables predict each other better than the dependent variable, making it difficult to interpret the regression coefficients and lead to poor statistical conclusions. Effects: Estimates of the regression coefficients are unstable depending on which variables are present, signs may be opposite of expectations, and