Ch9: Multivariate Correlational Research

Multivariate design

A design that measures more than two variables and looks at the relationships among them.

Longitudinal design

Measures the same variables in the same people at several points in time. Provides evidence for temporal precedence.

Cross-sectional correlation

A correlation between two variables measured at the same time. Can't show which came first.

Autocorrelation

The correlation of one variable with itself across time. Same as test-retest reliability.

Cross-lag correlation

A correlation between the earlier measure of one variable and the later measure of another. Usually the main interest, since it addresses directionality.

Cross-lag outcomes

If only one path is significant, it suggests that direction. If both are significant, the variables reinforce each other in a cycle.

Statistically significant

The 95% CI does not include zero. Also shown as p < .05 or an asterisk, and "n.s." means not significant.

Why not an experiment

Experiments are the only certain way to confirm causation, but some variables can't be manipulated or it would be unethical. Longitudinal correlation is a good alternative.

Multiple regression (multivariate regression)

Tests whether a relationship still holds when other variables are statistically controlled for. Helps rule out some third variables.

Control for

Checking whether a relationship between two key variables still holds when a third variable is accounted for statistically.

Criterion variable (dependent variable)

The outcome researchers most want to predict. Usually in the top row or title of a regression table.

Predictor variable (independent variable)

Any other variable in the regression used to predict the criterion.

Beta (β)

Shows the relationship between a predictor and the criterion with other predictors controlled. Positive, negative, or near zero, and larger means stronger. Can be compared within a table but not across tables.

b (unstandardized coefficient)

Shows direction of the association but is based on original units, so b values can't be compared within a table.

Beta near zero after controlling

The relationship is most likely explained by a third variable.

Regression signal phrases

"Controlled for," "adjusted for," and "considering" in media articles signal multiple regression.

Limits of regression

Can't establish causation, since researchers can't control for variables they didn't measure. Only an experiment rules out predisposition.

Pattern and parsimony

Building a case for causation from many different studies that all point to one causal theory. Used when an experiment isn't possible.

Parsimony

The degree to which a theory gives the simplest explanation. For causal claims, the theory requiring the fewest exceptions.

Smoking and lung cancer example

One theory (smoke toxicity) predicted five diverse findings, which makes third variables like coffee hard to sustain.

Pattern and parsimony in media

Journalists often report one study at a time, leaving out context and prior evidence. This makes science look unconnected and easy to criticize.

Mediator (mediating variable, mechanism)

A variable between two others that explains why they're related (ex., social ties explaining why deep talk leads to well-being).

Mediation and temporal precedence

The causal variable is measured first, the mediator second, and the outcome last.

Mediator vs. third variable

A mediator sits between the two variables and is of direct interest (A leads to C leads to B). A third variable is external, a nuisance, and explains the link only because it correlates with both.

Mediator vs. moderator

A mediator explains why two variables are linked. A moderator asks for whom or in what situation the link is strongest.

Multivariate designs and the four validities

Longitudinal designs give temporal precedence and regression helps internal validity. Construct, external, and statistical validity still need to be checked.