Multiple regression basics & more First, here are the minimal things
Transcription
Multiple regression basics & more First, here are the minimal things
Multiple regression basics & more First, here are the minimal things to do in a multiple regression analysis. You should probably also look at scatterplots of each X-Y pair, screen for bad data, etc.) 1. 2. 3. 4. 5. 6. 7. Start! Analyze > Regression > Linear Enter the Dependent variable Enter the Independent(s) in the same box for simultaneous regression Ask for Statistics: Estimates, Model Fit, Descriptives, Part and partial correlations, Collinearity diagnostics Save diagnostics: Cook's, Leverage (or Mahalanobis), Studentized deleted, Standardized DfBeta(s), Standardized DfFit generate Plots: ZPRED on the x-axis and ZRESID on the y-axis (i.e., a residuals plot), with a Histogram and a Normal Probability Plot Click OK. Add a LOESS curve to the residuals plot. You'll get lots of output! Basic simultaneous multiple regression Using the hregress.sav data one more time, regress timedrs simultaneously on phyheal, menheal, stress, and esteem following all of the steps above (you can skip the diagnostics), then answer the following questions (in an email or a Word document that you send to me). 1. 2. 3. 4. 5. 6. Is the model as a whole significant? Report F for the model along with R2. Which predictors are significant? Report the partial slope for each predictor along with the t-test and the associated pvalue. How much variance is accounted for by each significant predictor? Report sr2 (the squared semipartial correlation) for each significant predictor. Does there appear to be a problem with collinearity? Report the predictor with the lowest tolerance along with its tolerance value. How does the normality assumption look? Check the P-P plot (or histogram) and state how bad or good things look. How does homoscedasticity look? Check the residuals plot and state how bad or good things look. Transform (using a natural logarithm transformation) the timedrs variable to get rid of the skew using the syntax below: COMPUTE logtimedrs = LN(timedrs+1) . EXECUTE . 7. Now regress this new variable simultaneously on phyheal, menheal, stress, and esteem. Do any of your answers to 1 through 6 above change? If so, which, and how do they change? More about tolerance 8. 9. Regress menheal simultaneously on phyheal, stress, and esteem. Compute and report 1 – R2. How does your answer to #8 compare to the tolerance value from #4 above? Based on your answer to the previous question, what is tolerance? ΔR2 and the F-change statistic Regress timedrs (the untransformed version) on phyheal. Report R2. Regress timedrs simultaneously on phyheal, menheal, stress, and esteem. Report R2. What is the change in R2 from #10 to #11? Now do a sequential regression of timedrs on phyheal in Block 1 and menheal, stress, and esteem in Block 2. Ask SPSS for R squared change statistics. How does R Square Change in the Model Summary table for Model 2 compare to your answer to #12? 14. Is the change in R2 significant? Report F for the change in R2. 10. 11. 12. 13.