Chapter 11 REGRESSION
Multiple Regression Uses Explanation Prediction
Multiple Regression Based On: Correlations Characteristics of a straight line
Regression vs. Multiple Regression One independent variable vs. more than one independent variable One dependent variable
Multiple Regression TType of Data Required Independent variables CCategorical - can be coded for entry CContinuous - meet assumptions Dependent variable CContinuous - should be normally distributed
AAssumptions Sample representative of population Variables should have normal distribution Homoscedasticity Linear relationship between variables
Power analysis If sample size = number of variables, Rsquared will equal Generally need subjects per independent variable Less than 10 subjects per independent variable leads to serious error
Relationship of correlation to regression Perfect correlation? No correlation? Imperfect correlation?
Regression Equation Formula for a straight line Predicted score = constant plus regression weight times score Y’ = a + bX Y’ = a + b1X1 + b2X2 = B3X3
Regression Equation Predicted score Y’ Constant a value of Y when X = 0 point where regression line intercepts the Y axis regression coefficient/s b or beta rate of change in Y with a unit change in X measure of slope of regression line
Regression Equation Constant or a is based on means of variables involved Regression coeffients, b or beta, based on correlation between two variables
Regression coefficients The b-weights are based on raw scores Beta-weights are based on standardized scores and are partial correlation coefficients
Regression line Least squares “Line of best fit” Deviations around this line sum to zero Deviations are the differences between the actual and predicted scores
Computer Example What is the multiple correlation between a group of independent variables entered in three blocks and the dependent variable, total positive psychological attitudes? Block 1: Age and education Block 2: Smoking hx and exercise Block 3: Sat. with wt. and Health
SPSS - Multiple Regression ANALYZE Regression Linear Statistics Confidence intervals R squared change Descriptives Part and partial correlations Options Exclude cases pairwise
Dummy coding Uses 1s and 0s a = mean of dependent variable for group assigned 0s throughout b - tests the difference between the group assigned 1 on the variable and the group assigned 0s throughout
Dummy coding Vector 1 Republicans = 1 Democrats = 0 Independents = 0 Vector 2 Republicans = 0 Democrats = 1 Independents = 0
SPSS - creating dummy variables TRANSFORM Compute New variable with new value IF create conditional expression
Effect Coding Uses 1s, 0s, and -1s a = grand mean of dependent variable b tests the difference between the mean of the group assigned 1 and the grand mean the b-weights add up to zero
Effect Coding Vector 1 Republicans 1 Democrats 0 Independents -1 Vector 2 Republicans 0 Democrats 1 Independents -1
MULTIPLE REGRESSION Selecting Variables for the Equation Standard (ENTER) Hierarchical (Setwise) Stepwise Forward Backward Stepwise
Mediator and Moderator Variables
Mediator Variable A variable seen as “between” the independent and dependent variable. Tested with multiple regression/path analysis
Moderator Variable Affects the association between an independent and dependent variable. Test for an interaction between the moderator and another independent variable using hierarchical multiple regression.
Example from the literature