Purpose
To researchfor and define the fundamental
constructs or dimensions assumed to underlie
the original variables
Data summarization
Define a small numbers of factors that maximize the
explanation of the entire variable set
Data reduction
Identifying representative variables from a much larger set
of variables
Exploratory and confirmatory
3.
Kinds of researchquestions
Defining the underlying structure
Number of factor
Nature of factor
Importance of solutions and factors
Testing theory in factor analysis
Estimating scores on factors
4.
Limitations to factoranalysis
Theoretical issues
Factors
Selection of observed variable
Practical issues
Sample size and missing data
Normality
Linearity
Outlier
Multicollinearity and singularity
Factorability of R
5.
What is factor?
Linear combination of the original variables
Factor
Principle
component
Common factor
Total variance Shared variance
Fundamental equation forfactor analysis
General form
Matrix
X: observed variables
μ: means
Λ: factors loadings
F: factors
E: errors matrix
R: correlations
D: unique variance
(R − D) = AA′
X = 𝜇 + ΛF + E X = 𝜇 + ΛF
R = AA′
PCAE=0
D=0
11.
Fundamental equation forfactor analysis
Eigen formulation
λ: eigenvalue
K: eigenvector
R − λI K = 0
12.
Eigenvalue and communalities
Eigen formulation R − λI K = 0
m
5 5 3 1
3 2 5 5
4 3 5 4
3 2 5 3
1
1
P
F =
λ1 λ2 λ3 λ 𝑚
𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒 =
𝑖=1
𝑝
𝑓𝑖𝑗
2
= 𝜆𝑗
𝑐11
𝑐22
𝑐33
𝑐 𝑝𝑝
𝑐𝑜𝑚𝑚𝑢𝑛𝑎𝑙𝑖𝑡𝑦 =
𝑖=1
𝑚
𝑓𝑖𝑗
2
= 𝐶𝑖𝑖
𝑓𝑖𝑗
facto
rvariable
13.
Axis rotations
Orthogonalrotation
Non-correlations among factors are assumed
Method
Quartimax
Simplify the rows of a factor matrix (largest variance on factor)
Varimax
Simplify the columns of a factor matrix (largest variance on
variable)
Equimax
Compromise quartimax and varimax
Oblique rotation
Correlations among factors are assumed
Data reduction
Preservation
Variable with significant factor loading
Variable with communality > .50
Deletion
Cross loading
20.
Practice issue
Metricvariables
Sample size
200 (Comrey, 1973)
10 observations per variable (Kline, 2005)
Some degree of multicollinearity
Bartlett test of sphericity
Measure of sampling adequacy (MSA)
Kaiser-Meyer-Olkin measurement of sampling
adequacy (KMO test)
21.
Sampling adequacy test
Bartlett test of sphericity (p < .05)
Are correlations among observations big enough?
MSA (> .05) and KMO (> .07) test
Are partial correlations among observations small
enough?
V1
V3 V4
22.
Comparison
PCA FA
Purpose
Maximize
explanation oftotal
variance
Maximize explanation
of shared variance
Term Principle component Common factor
Meaning Linear combination Construct
Relationship
among factor
0 R
Diagonal
matrix
1 Communality
23.
Comparison
PCA andFA reveal almost identical results
High ratio: variable/factor
High common variance
Different results (Widaman, 1993)
Low ratio: variable/factor (3:1)
Low common variance (<.4)
Which one is preferred?
Psychological concept
Satisfaction on product/consumer behavior
24.
Latent variable
Psychologicalconcept
Behavioral pattern/sampling behavior/item
V1 V2 V3 V4 V5 V6 V7 V8 V9
F1 F2 F3
u u u u u u u u u
Procedure
1. Research question
2.Select the type of factor analysis
3. Check the assumptions
4. Derive factor analysis and assess overall fit
5. Interpret the factors
6. Validation of factor analysis
7. Additional use of factor analysis results
After EFA
Itemanalysis
Appropriateness of item
Item discrimination
Homogeneity of item
Indices
Critical ratio (CR) t > 3
Item total correlation >.3
Cronbach’s α ↑ if item deleted