Week 5 Flashcards
Hierarchical multiple regression (9 cards)
Why use hierarchical multiple regression where predictors are entered in a specific order?
Enables us to examine the contribution of additional variables to the model when added in a separate step.
We can only include continuous and binary (coded 0 and 1) data into regression analyses.
Therefore, if we have nominal/categorical data with three or more categories, what needs to happen?
We need to dummy code them.
This involves converting the one nominal variable that has three or more categories into multiple binary variables instead.
Residuals plots - in the statistics tab after you select analyse/regression/linear regression.
Any residual value greater than 2 will be listed in the output. These outliers can have a big impact on the effect of a variable estimated in your model.
These plots help us to understand whether we have met the assumptions of homoscedasticity and normally distributed errors. (*ZRESID (Y) & *ZPRED (X) shows us whether we have normally distributed errors, *SRESID (Y) & *ZPRED (X) will also show heteroscedasticity)
In a hierarchical regression, what variables are typically entered in step 2 of the analysis?
a. Variables well established in the literature as being associated with the outcome.
b. More exploratory or new variables where the relationship with the outcome variable is not established.
c. Categorical variables
d. Continuous variables
b. More exploratory or new variables where the relationship with the outcome variable is not established.
(Usually known predictors are entered in step 1, then exploratory predictors are entered in step 2+)
Which of the following types of variables would need to be dummy coded before being entered into a regression analysis?
a. Continuous variables
b. Non-normally distributed variables
c. Binary variables (i.e., with 2 categories)
d. Categorical or ordinal variables (with 3+ categories)
d. Categorical or ordinal variables (with 3+ categories)
To meet the assumption of homoscedasticity, the residuals at each level of the predictor variable should have…
a. The same variance
b. Different variance
c. The same mean
d. Different mean
a. The same variance
What does the F Change statistic tell us?
a. Whether the model is significant
b. Whether an individual predictor variable is significantly associated with the outcome variable
c. Whether the new additional variables significantly increase the proportion of variance accounted for
d. Whether there is multicollinearity across your predictor variables
c. Whether the new additional variables significantly increase the proportion of variance accounted for
If you want to compare the strength of associations between different predictor variables and the outcome variable, what statistic would you read?
a. Unstandardised beta
b. Standardised beta
c. t-test
d. VIF
b. Standardised beta
What is the difference between the unstandardised beta and the unstandardised beta coefficient?
They both represent the change in the dependent variable based on either a one unit change in the dependent variable (unstandardised beta, B) or for a one standard deviation change in the independent variable (standardised beta β )
