Interpretations
Types of Regression
Assumptions
Wildcard
100
You predict adolescents' well-being from age, gender, and number of friends. Interpret R2 = 0.14.
14% of variance in adolescents' well-being is explained by age, gender, and number of friends.
100
Your select predictors and enter them all at the same time into the model. What type of regression is this?
Standard multiple regression
100
Which, if any, regression assumptions are violated in this residuals vs. predicted values of Y plot?
No assumptions are violated
100
You find that shorter sleep duration predicts worse memory performance in older adults. Can you conclude that short sleep duration causes memory problems, and that people need to sleep more to remember better?
No, this relationship could be due to third variables (medications, health problems, stress...). Memory could also affect sleep duration.
200
You predict neural BOLD signal in the amygdala from demographics (Step 1) and emotional vs. neutral stimuli (Step 2). The incremental R2 for Step 2 is 0.05. Interpret the incremental R2.
Additional 5% of variance in the amygdala BOLD signal is explained by emotional vs. neutral stimuli, after accounting for demographics.
200
You select predictors and enter them in multiple blocks that you choose into the model. What type of regression is this?
Hierarchical multiple regression
200
Which, if any, regression assumptions are violated in this residuals vs. predicted values plot?
Linearity Assumption is violated
200
What is detected by Mahalanobis Distance?
Multivariate outliers (unusual combinations on multiple variables)
300
You predict social well-being in older adults. The unstandardized regression coefficient for "Do you volunteer (Yes/No) is 8. Interpret the coefficient.
Volunteering is associated with an 8 point increase in social well-being in older adults
300
You select a pool of predictors, from which the statistical software selects the best ones to add to the model first. What type of regression is this?
Stepwise multiple regression (Forward Selection)
300
Which, if any, regression assumptions are violated in this plot?
Homoscedasticity is violated
300
What is an advantage of reporting standardized regression coefficients over unstandardized regression coefficients in a results section or results table?
Standardized coefficients are interpretable in terms of effect size and comparable.
Unstandardized coefficients are highly impacted by the measurement scale of the variables involved.
The significance test results will of course be identical for both :)
400
You predict number of lifetime illnesses in infants from duration of breastfeeding. The standardized regression coefficient for is -0.15. Interpret the coefficient.
1 SD increase in breastfeeding duration is associated with 0.15 SD lower number of lifetime illnesses in infants.
400
You select a pool of predictors, which are all entered into the model. The statistical program then removes non-significant predictors. What type of regression is this?
Stepwise multiple regression (Backward Deletion)
400
Which if any regression assumptions are violated in this residual vs. predicted values plot?
Normality is violated
400
What question is answered by this test in multiple regression?
Does the regression model predict the outcome better than chance?
If the F-test is significant, the model (set of predictors) predict variance in the DV better than chance.
500
Label the following effect sizes as small, medium, or large:
R2=0.14
R2=0.05
Beta=-0.15
R2=0.14 medium
R2=0.05 small
Beta=-0.15 small
500
You select a pool of predictors, which are entered and/or removed from the model by the program based on statistical criteria. What type of regression is this?
Stepwise multiple regression (bidirectional elimination)
500
Name one option each for dealing with violations of the following assumptions:
1) Normality/Homoscedasticity
2) Linearity
3) Independence of Error
4) Absence of Multicollinearity
1) Maximum Likelihood Estimation with Robust standard errors (MLR) in R Lavan
2) Consider adding quadratic/cubic terms of predictors that cause non-linearity
3) If clustering issue => Multilevel Modeling or Complex sampling, if time-sequence dependence => Consider time-sequence as additional predictor
4) Combining highly related predictors or excluding one
500
You predict children's academic achievement from different types of ADHD symptoms. Inattention has negative correlation with achievement, but in your model it has a positive significant coefficient. What is the name of this phenomenon?
Suppressor effect (these should be interpreted with caution).