Chapter 9: Multivariate Correlation Techniques

Unlike Bivariate Correlations, Multivariate correlations can

build a case for cause

Multivariate design

a study design measuring the association between more than two measured variables

Multivariate Correlational techniques

1. pattern and parsimony

2. Longitudinal correlations

   • auto-correlations

   • Cross-lag

Pattern and parsimony

Find the most parsimonious explanation for a pattern of association across multiple studies

pattern: result patterns across numerous correlational studies point to a single, causal direction

(ex: smoking causes lung cancer)

Pattern and Parsimony with the Theory data cycle

step 1: specify a theory for a causal path

step 2: specify predictions from theory

step 3: test/collect data

step 4: find the most parsimonious explanation for the pattern of data

Longitudinal Research designs

a study in which the same variables are measured in the same people at different points in time

1. Autocorrelational

2. Cross-lag

Why are longitudinal research designs multivariate

because we are interested in multiple variables and their relationships across time

Autocorrelations

a type of longitudinal research design measuring the correlation of one variable with itself at 2 or more points in time

Investigates stability/instability of traits and behaviors across time

ex: puzzle time week 1——→ puzzle time week 2

Cross-lag correlations

a type of longitudinal research design measuring the correlation of one variable and a later different variable

Investigates TEMPORAL PRECEDENCE

ex: sadness time 1 ——— sleep time 2

Cross-sectional correlation

the correlation between 2 measured variables at the same point in time (Bivariate)

Investigates Covariance

Mediators

a variable that helps explain the relationship between two other variables (explanatory)

what is the cause? HOW and WHY effect occurs?

ex: variable x associated with variable y. variable x——→variable z———>variable y

Tests for mediation

• tests why 2 variables are linked

• ideally investigated overtime

tests for moderation

• tests when 2 variables are linked and for whom

• investigation 1 time (cross-sectional) or over time (longitudinal)

Moderator vs Mediator example

Conscientiousness linked to better physical health

Mediation: Following Dr’s orders

Conscientious people are more likely to follow the Doctors orders which is why they have better health (Conscientiousness—→ follow Dr’s orders——→better health)

Moderation: age

The link between conscientiousness and better physical health is stronger for adults and weaker in children

Mediator vs Third Variables

mediator:

relevant to the relationship being studied because it has explanatory power in a causal way

Third Variable:

Unstudied variable which effects the association studied giving the false impression that the two variables are related (confounds)

When to use pattern and parsimony

1. an experiment would be unethical

2. struggle to rule out third variables (internal validity issue)