PP 17

What are factor loadings?

How strongly a variable is related to a factor/component.

How to interpret loadings?

• High loading = the item strongly belongs to that factor

• Low loading = the item has little relationship with that factor.

What is Kaiser's Criterion method and limitations?

• Keep factors with eigenvalue > 1

• Keeps too many factors

• Least accurate method.

What is scree plot method?

• A graph of eigenvalues

• Keep the factors before the graph starts to bend.

What is parallel analysis method?

• Compares real data with randomly generated data

• Keep the factors that explain more variance than would be expected by chance

• Most recommended method.

What does EFA separate?

Common variance which reflects the latent construct from unique variance.

In EFA, what do factor loadings depend on?

The number of extracted factors.

What kind of stuff is PCA used for?

Data reduction.

What kind of stuff is EFA used for?

Questionnaire development and validation.

What does common variance represent?

The latent construct.

What is unique variance treated as?

Error/item specific variance.

What does total variance equal?

Common variance + unique variance.

What does PCA say about variance and communality?

Assumes all variance is common and starts with communality = 1 for every variable.

What does EFA say about communality?

It estimates the communality of each variable.

What is an eigenvector?

Represents the direction (axis) of a factor or component.

What is an eigenvalue?

How much variance a factor/component explains, the larger the eigenvalue the more important.