Chapter 8Multivariate Regression Analysis.PPT

Chapter 8Multivariate Regression Analysis 8.3 Multiple Regression with K Independent Variables 8.4 Significance tests of Parameters Interpreting Nested byx Interpreting ?* ANCOVA * * Population Regression Model For K independent variables, the population regression and prediction models are: The principles of bivariate regression can be generalized to a situation of several independent variables (predictors) of the dependent variable The sample prediction equation is: Predict number of children ever born (Y) to the 2008 GSS respondents (N=1,906) as a linear function of education (X1), occup’l prestige (X2), no. of siblings (X3), and age (X4): People with more education and higher-prestige jobs have fewer children, but older people and those raised in families with many siblings have more children. Use the equation to predict the expected number of kids by a person with X1 = 12; X2 = 40; X3 = 8; X4 = 55: 2.04 For X1 = 16; X2 = 70; X3 = 1; X4 = 25: 0.71 OLS Estimation of Coefficients As with bivariate regression, the computer uses Ordinary Least Squares methods to estimate the intercept (a), slopes (bYX), and multiple coefficient of determination (R2) from sample data. OLS estimators minimize the sum of squared errors for the linear prediction: min See SSDA#4 Boxes 8.2 and 8.3 for details of best linear unbiased estimator (BLUE) characteristics and the derivations of OLS estimators for the intercept a and slope b Nested Equations A set of nested regression equations successively adds more predictors to an equation to observe changes in their slopes with the dependent variable Predicting children ever born (Y) by adding education (X1); occupational prestige (X2); siblings (X3); age (X4). (Standard errors in parentheses) (1) (2) (3) (4) F-test for ?2 The hypothesis pair for the multiple coefficient of determination remains the same as in the bivariate case: But the F-test must also adjust the sample estimate of R2 for the df associated with the K predictors: As