Lecture 7

Question 1

Multiple regression analysis:

> why is R² smaller than the sum of R² of all predictors?

Answer 1

predictors are mutually correlated

> mutual correlation high = colinearity

Question 2

2 criteria to find “best” regression model

Answer 2

best regression model

1. only has predictors that contribute significantly to the model
2. the highest R² of all models that meet criterion 1

Question 3

how to find the best regression model

> what does not work? why?

> what are alternatives?

> what is the general problem?

Answer 3

how to find best regression model

> adding predictors until no variance is added, does not work

>>> predictors can suppress errors of other predictors, work as a team

> you could test all models and choose the best, but a lot of work

> shortcut techniques (forward selection, backwards elimination, stepwise procedure)

> general problem: the best model in the sample is not necessarily the best model in the population

Question 4

why is the best model in the sample not necessarily the best model in the population?

Answer 4

because of chance capitalization on idiosyncrasies in the data set

Question 5

what is hierachical multiple regression?

advantages?

disadvantages?

Answer 5

hierarchical multiple regression - technique to find a model that fits the data but also theory

> step wise entry of predictors on the basis of an a priori theory

advantages:

1. no chance capitalization due to idiosyncrasies in data
2. fit mode provides evidence for quality of the theory

disadvantage:

1. model is perhaps not the best for the dataset at hand

Question 6

how to implement categorical predictor into MRA?

how to implement interaction term?

Answer 6

use of dummy variable

> 1 / 0 plus regression coefficient

interaction term (with dummy):

> S x A interaction: product of dummy variable and predictor variable created interaction term

Question 7

ANCOVA

> what is a covariate?

Answer 7

Covariate: variable that is not controlled (or cannot be controlled) but is measured

Question 8

what is the key characteristic of a quasi experimental study?

> problems?

Answer 8

quasi experimental design:

> independent variable is not manipulated, but its levels are selected (for example clinical vs control group)

problem:

> the groups may differ

Question 9

quasi experimental design:

> what are ways to increase validity?

Answer 9

quasi experimental design:

1. add a real independent variable to the design
2. match groups on potentially confounding variable
3. measure potentially confounding variables

> use as covariate