Homework 04

Homework 04 - due 10/18/2017

For Homework 04, you will be using the HELP dataset, learn more at:

• https://melindahiggins2000.github.io/N736Fall2017_HELPdataset/ &
• https://github.com/melindahiggins2000/N736Fall2017_HELPdataset

Complete the following:

1. Perform a Simple Linear Regression for:
   • OUTCOME variable indtot: “Inventory of Drug Use Consequences (InDue) total score - Baseline”
   • PREDICTOR variable mcs: “SF36 Mental Composite Score - Baseline”
   • decide if you want to transform either variable indtot or mcs and if so, what transformation you applied and why
2. Perform regression diagnostics:
   • check the normality of the residuals (histogram and Q-Q plots)
   • check for linearity - is there any systematic relationship between the residuals and the predicted (or fitted) values?
   • homoscedasticity - plot of standardized residuals versus fitted values - this is known as a “Scale-Location” graph.
   • check for outliers and data points with high leverage or influence: outliers are often identified with standardized residuals > 3 (or <-3) and influential observations are often identified using Cook’s D
3. Provide a summary of the regression results.
   • provide a FIGURE of the model, in this case a scatterplot with the fitted line overlaid and 95% confidence intervals if you can
   • Make a TABLE presenting the fitted regression model (coefficients and tests of significance for those coefficients)
   • describe the variance explained by the model (based on r2)
   • describe the model itself based on the y-intercept and slope terms
   • note any limitations or issues with the model fit or interpretation of the model
4. Perform a One-way ANOVA for:
   • OUTCOME variable indtot: “Inventory of Drug Use Consequences (InDue) total score - Baseline”
   • GROUP variable racegrp: “Racial Group of Respondent”
   • I would suggest merging “other” and “hispanic” together and create a 3-group variable for race, since the “other” category is only about 6% of the sample.
   • options - you can use either an ANOVA or GLM modeling approach
   • if the GROUP variable is significant, also perform post hoc tests - use some kind of pairwise error rate adjustment (i.e. bonferroni, sidak, Tukey’s HSD, etc) - be sure to report which one you used and why
5. Perform model diagnostics:
   • homoscedasticity - look at a test for equal variance (Levene’s test or Bartlett’s test).
   • if this test of equal variances fails, you may want to report a modified F-test (e.g. Welch’s test)
6. Present a summary of the ANOVA results.
   • Make a FIGURE of the group mean differences - either an error-bar plot or a series of boxplots one for each group to show the group differences in the outcome
   • Make a TABLE presenting the ANOVA results
   • describe the model results - was the GROUP (racegrp) significant?
   • If GROUP is significant, what did the post hoc tests reveal?

Variables in HELP dataset to be used for Homework 04