Page 209 - Quantitative Data Analysis

Quantitative Data Analysis
Simply Explained Using SPSS

Assumptions of PCA/EFA

Following are the assumptions underlining in PCA/EFA.

1. Scale of Measurement: All analyzed variables should be
either an interval or ratio type.

2. Random Sampling: Each subject can contain one and only
one observed variable. These set of scores should represent
a random sample drawn from the population.

3. Linearity: There should be linear relationship between all
observed variables.

4. Normal distribution: The distribution of each observed
variable should have normal distribution with mean is equal
to zero and standard deviation is one. Variables with non-
normal skewness or kurtosis can be transformed for better
approximated normality

5. Bivariate normal distribution: Each pair of observed
variables should display a bivariate normal distribution.

Similarities in PCA and EFA

Principal Components Analysis (PCA) is similar to FA in that it is an
analysis of the correlations among items in order to determine the
minimum number of factors accounting for the variation in test
scores. However, PCA does not assume a causal structure. It is
more simply a variable reduction technique designed to display the
underlying mathematical structure of a set of items (variables).

The Theory and Applications of Statistical Inferences 193