Exploratory Factor Analysis

1
Exploratory Factor Analysis

• COMS 712

2
Objectives for Tonight

• Learn about Exploratory Factor Analysis (EFA)
• Discuss the following articles
• Thoman, Palmer, Coker-Juneau Williams
• Levine McCroskey
• Go to the Lab

3
Exploratory Factor Analysis

• What is it?
• When do you use it?
• Key Questions you are trying to answer
• What are the procedures for doing an EFA?
• How many factors should I retain?
• How do I know which items to include on which
  factors?
• What are some limitations and data considerations
  relevant to EFA?

4
When should you use EFA

• Determine how to reduce the number of variables
  (data reduction).
• Increase the parsimony of analyses.
• Reduce the number of analyses performed (hence, ?
  the Type I error).
• Reduce the potential of multicolinearity in
  regression equations.

5
Procedures for EFA

• Select variables
• Perform Sphericity test
• Identify criteria for factor extraction
• Identify rotation method
• Perform analysis and interpret results
• Compute factors, calculate reliability, and
  perform primary analyses

6
How Many Factors to Retain

• The most widely used criterion is Kaisers (1960)
  Eigenvalue test
• Retain all factors with Eigenvalues gt 1
• Most robust when the number of variables is small
  (10-15) or moderate (20-30).
• The Scree Test is a visual method of identifying
  factors
• Retain all factors before the line flattens out
• Retain all factors until a minimum threshold of
  variance accounted for is achieved.

7
Scree Plot for Data
8
How Many Factors to Retain

• SPSS uses the Kaiser Eigenvalue criterion by
  default, but also provides a scree plot for
  inspection.
• If your data meets the following characteristics
  there is no reason to avoid the Kaiser criterion
• Number of variables lt 30 and Communalities are gt
  .70
• N gt 250 and Mean Communality is gt .60.
• Your factors should account for a meaningful
  amount of variance.

9
Which Items to Include?

• The factor analysis components are artificial
  derivates and are impossible to use when
  interpreting factor structure.
• Rotation methods are used to make factor
  structures more clear.
• Orthogonal Rotation Factors uncorrelated
• Oblique Rotation Factors correlated

10
Which Items to Include

• The Orthogonal rotation method is easier to
  interpret. By default SPSS uses the cleanest
  of the methods Varimax.
• Oblique rotation is more difficult to interpret
  and you have more options Promax, oblimin,
  oblimax and others.
• Try both and see what happens.

11
Which Items to Include

• You use rotated factor scores to place items
• 70/30 Criterion
• 60/40 criterion
• Items should not load on more than one factor.
• Unless your sample is gt 500 all factors should
  have at least 1/5 of the items.

12
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
13
Limitations and Data Considerations

• Factor structures are unstable in small samples.
• The assumption of uncorrelated factors is rarely
  present.
• Reliability coefficients and correlations must be
  analyzed. Retain only one score if both
  conditions are true
• Reliability for the factors is weaker than the
  overall scale.
• Correlations between factors are large