[其它]国外心理学-设计和分析课件mreg.pdfVIP

Multiple Regression Psychology 3256 Introduction • Often we are interested in simple 1 to 1 variable relationships • but let’s say that r = .50 for some relationship r = cov /s s xy x y • • How much variance is accounted for by one variable in the other? How much indeed • well r deals with standard deviations, so square r 2 • r = .25 • so we have accounted for 25 percent of the variance • which means there is 75 percent left! ergo.. • There must be other variables that account for the rest of the variance • We deal with this by bringing them in to the model Pretty pictures X2 Y X1 In general, we have a model... ŷ = b +b x +b x +...+b x +e • 0 1 1 2 2 p-1 p-1 • We have p-1 predictor variables • This is for the data set itself, these are statistics, not parameters In the population... Y = β + β X + β X + ...+ β X + ε 0 1 1 2 2 p −1 p −1 εNID(0,σ 2 ) ε • ε is not a prediction error, it is individual € variation • Note it is Y not a predicted y hat What we get • if p-1 =1 we get a line • if p-1 =2 we get a surface or plane • if p-1 2 we get a hyperplane in hyperspace! • Best not to try and visualize a hyperplane.. And you thought you were done with ANOVA…. • We can find out if our regression model is significant with ANOVA • Variance due to regression (the model) • Variance due to residual Yes, ANOVA SV df SS MS