7. Compared to modeling BMI linearly, do you think modeling BMI with a quadratic term, a spline, or categories is more or less robust to non-linearity? 4
8. Is the constant variance assumption met in this model? Describe how you reached your conclusions, i.e. why or why not? 9. Was the assumption of independence met in this model? Yes.Describe how you reached your conclusions, i.e. why or why not? 5
10. Use median regression to evaluate whether the findings for BMI are robust to outliers (continue to use “other” as the race reference category). Compare your results (BMI estimate only) to the standard regression model. Interpret the coefficient of BMI in the median regression. Do you think outliers for BMI are driving the association between BMI and systolic blood pressure? Please provide your SAS code and the SAS output of the models. See Appendix. 11. Is there evidence of an interaction between bmi and alcohol consumption? How did you test for the interaction? I added an interaction term for bmi and alcohol consumption, ran the model, and checked to see if p < 0.05 for the interaction term for the null hypothesis that β = 0.Write out your model. 9 *bmi*boozecat1 Please provide your SAS code and the SAS model output. See Appendix. 6
Appendix for SAS code and abridged SAS model output. Question 1 /* Q1 */ data HW2; set "P:\Spring 2014\EPI204\hw2.sas7bdat" ; bmi = wt/((height/ 100 )*(height/ 100 )); sex = sex - 1 ; boozecat = (booze = 0 ); run ; Question 2 /* Q2 */ proc glm data =hw2 ; class race / ref = last; model sbp = ageyrs bmi race sex smokever boozecat hdl / solution ; output out =hw2_2 predicted =phat residual =resids; run ; The SAS System The GLM Procedure Dependent Variable: sbp Source DF Sum of Squares Mean Square F Value Pr > F Model 8 1077805.962 134725.745 324.77 <.0001 Error 7478 3102111.733 414.832 Corrected Total 7486 4179917.695 R-Square Coeff Var
You've reached the end of your free preview.
Want to read all 13 pages?