Chapter 6: Multivariate correlational research

Question 1

Multivariate correlational research

Answer 1

Studies that try to measure more than two variables and look at the correlation between them
Brings us closer to making claims about causality because we cal already rule out some confounders

Question 2

Longitudinal design
Prospective design
Cross-sectional design

Answer 2

Longitudinal: measuring the same variable at multiple points in time in a sample (measuring each variable) → can’t rule out alternative explanations
Prospective: measuring the same variable at different points in time (independent at one point, dependent at other point)
Cross-sectional: measuring the variables one time

Question 3

Three types of correlations for interpreting results of longitudinal designs

Answer 3

1. Cross-sectional correlations
2. Auto-correlations
3. Cross-lag correlations

Question 4

Cross-sectional correlations

Answer 4

Correlation between two variables that were measured at the same time
Example: relation between TV violence and aggression at moment 1 and TV violence and aggression at moment 2 → positive relation between TV violence and aggression at 8-9 years old, no effect when they’re older

Question 5

Auto-correlations

Answer 5

Correlation between variables on a specific measuring moment and the same variable on another measuring moment
Example: relation between TV violence at moment 1-2 and aggression at moment 1-2 → significant correlation between aggression at moment 1-2, no relation between TV violence 1-2

Question 6

Cross-lag correlations

Answer 6

Relation between two different variables, measured on two different points in time
Example: relation between TV violence 1 and aggression 1 & TV violence 2 and aggression 2 → significant correlation between TV violence 1 and aggression 2 → covariance and temporal precedence
! What we are actually modeling, is a change over time: ‘… is related to an increase in …’

Question 7

Multiple regression analysis

Answer 7

We will try to predict the dependent variable while using the independent variables
Advantage: you can see if the effect still stands even if you control for other predictors
! Can not offer definitive evidence for causal effects
How to recognize it in research: ‘controlling for…’, ‘correcting for…’, ‘when we take … into account…’

Question 8

Two ways to write a beta

Answer 8

1. β = standardized beta: can be compared to each other
2. b = unstandardized beta: can not be compared to each other

Question 9

Parsimonious explanations

Answer 9

An explanation that is simple, straightforward and requires the fewest assumptions or steps to account for the observed data
Strong evidence for causal effect

Question 10

Mediation mechanism

Answer 10

Question: why is there a relationship between X and Y?
Mediator (M) explains the relationship between X and Y
Mediators describe a causal chain, they describe a process or mechanism
Idea: numerous multiple regression analyses
Example: ‘if you give children more recess time, they will have more physical activity and will be tired so they won’t cause trouble and have less behavior problems in class’ → mediator: physical activity

Question 11

Four steps to demonstrate mediation

Answer 11

1. Predict Y with X → demonstrates path C
2. Predict M (mediator) with X → demonstrates path A
3. Predict Y with M → demonstrates path B
4. Predict Y with X and M: if the correlation in this path is smaller than in step 1 (c-path → mediation

Question 12

Mediator vs confounder

Answer 12

Mediator: explains why there is an effect from X to Y → strength of correlation changes (does not disappear)
Confounder: alternative explanation → effect disappears if we take it into account

Question 13

Controlling for confounders

Answer 13

1. Control by design
2. Control by randomization
3. Statistical control