A peer-reviewed electronic journal.

Copyright is retained by the first or sole author, who grants right of first publication to Practical Assessment, Research & Evaluation. Permission
is granted to distribute this article for nonprofit, educational purposes if it is copied in its entirety and the journal is credited. PARE has the
right to authorize third party reproduction of this article in print, electronic and database forms.
Volume 19, Number 5, June 2014

ISSN 1531-7714

Improving Your Exploratory Factor Analysis for Ordinal Data: A

Demonstration Using FACTOR
James Baglin, RMIT University, Melbourne, Australia
Exploratory factor analysis (EFA) methods are used extensively in the field of assessment and evaluation.
Due to EFAs widespread use, common methods and practices have come under close scrutiny. A substantial
body of literature has been compiled highlighting problems with many of the methods and practices used in
EFA, and, in response, many guidelines have been proposed with the aim to improve application.
Unfortunately, implementing recommended EFA practices has been restricted by the range of options
available in commercial statistical packages and, perhaps, due to an absence of clear, practical how-to
demonstrations. Consequently, this article describes the application of methods recommended to get the
most out of your EFA. The article focuses on dealing with the common situation of analysing ordinal data as
derived from Likert-type scales. These methods are demonstrated using the free, stand-alone, easy-to-use and
powerful EFA package FACTOR (http://psico.fcep.urv.es/utilitats/factor/, Lorenzo-Seva & Ferrando,
2006). The demonstration applies the recommended techniques using an accompanying dataset, based on the
Big 5 personality test. The outcomes obtained by the EFA using the recommended procedures through
FACTOR are compared to the default techniques currently available in SPSS.

Exploratory factor analysis (EFA) is a cluster of

common methods used to explore the underlying
pattern of relationships among multiple observed
variables. EFA is useful for assessing the
dimensionality of questionnaire scales that measure
underlying latent variables. Researchers use EFA to
hypothesise and, later, confirm, through replication or
confirmatory factor analysis (CFA), the model that gave
rise to the interrelationships among the scales
variables. From a practical standpoint, a researcher
might want to know if a new scale can be considered
uni- or multi-dimensional. Responding to this can help
researchers reduce the total number of variables into a
smaller number of factors, which are composed of
highly related variables. EFA remains a very popular
data analysis technique. However, criticisms of
conventional EFA practices have frequently been cited
(see Fabrigar, Wegener, MacCallum, & Strahan, 1999).
EFA requires subjective input from the researcher and
making poor decisions at each EFA stage can have a
substantial impact on results. The main areas of
concern include deciding on the following: an

appropriate extraction method, the number of factors

to retain, and the factor rotation technique (Costello &
Osborne, 2005)
Many researchers have attempted to improve the
practice
of
EFA
by
publishing
useful
recommendations, many of which appear in PARE
(Beavers et al., 2013; Costello & Osborne, 2005;
Courtney, 2013; DiStefano, Zhu, & Mindrilla, 2009;
Osborne & Fitzpatrick, 2012). The authors of these
articles should be commended for presenting clear and
practical
guidelines
for
EFA
practitioners.
Unfortunately, many of the recommendations are
challenging to implement using mainstream statistical
software such as IBM SPSS Statistics (IBM Corp,
2012). Commercial statistical packages, textbooks and
instructors appear to be lagging behind recommended
practices. Clear practical demonstrations for improving
EFA are needed. Basto and Pereira (2012) and
Courtney (2013) discuss the use of the R-Menu plugin
for IMB SPSS Statistics (IBM Corp, 2012) for
implementing judicious factor retention methods. The
R plugins for SPSS substantially improve EFA

Page 2

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

functionality, but the installation process can be
burdensome and the very idea of using free EFA tools
through a commercial package will seem nonsensical to
some. Using the EFA packages directly through the
free statistical package R (R Core Team, 2013) requires
familiarity of the open source software and a moderate
level of programming knowledge required may be offputting to some. As such, the aim of this paper is to
demonstrate a viable alternative for implementing
recommended EFA methods.
Specifically, this article will demonstrate how to
apply recommended EFA methods using the freely
available
PC
software
package
FACTOR
(http://psico.fcep.urv.es/utilitats/factor/,
LorenzoSeva & Ferrando, 2006). FACTOR is an easy to use,
standalone EFA package with many powerful and
cutting-edge features. It runs directly from an
executable file and does not require installation. The
demonstration, using an accompanying dataset, will
focus on the analysis of ordinal variables created
through the use of Likert-type rating scales. Ordinal
variables are the norm for most EFAs. Research shows
that the common use of Pearson correlations for EFA
models can substantially underestimate the strength of
the relationship between ordinal variables. This can
lead to spurious factors in EFA outcomes (Garrido,
Abad, & Ponsoda, 2013; HolgadoTello, Chacn
Moscoso, BarberoGarca, & VilaAbad, 2008; Olsson,
1979b).
Improving Your EFA
Prior to demonstrating the use of FACTOR,
methods for improving your EFA practices will be
briefly summarised. The reader is directed to excellent
articles in PARE by Costello and Osborne (2005) and
Beavers et al. (2013) for more comprehensive
treatments. At the outset, the decision faced by the
practitioner during EFA is choosing between EFA and
principal components analysis (PCA). Both EFA and
PCA are used to reduce the number of dimensions
inherent in the data. However, both make very
different assumptions. Factor analysis is concerned
with identifying the underlying factor structure that
explains the relationships between the observed
variables. On the other hand, PCA is used to reduce a
large number of interrelated variables into a smaller set
of "components" with minimal loss of information.
For example, a researcher with multicollinearity issues
in a multiple regression model might use PCA to
cluster highly related variables into a single predictor to

avoid biased parameter estimates. PCA does not

attempt to explain the underlying population factor
structure of the data and makes the often, unrealistic,
assumption that each variable is measured without
error. EFA, on the other hand, is based on the
common or shared variance between variables, which is
partitioned from the left-over variance unique to each
variable and any error introduced by measurement.
Hence, EFA is more theoretically aligned to the goals
of exploring the dimensionality of a scale proposing to
measure a latent variable. The researcher must also
decide between EFA vs. confirmatory factor analysis
(CFA). Generally speaking, CFA is selected when a
researcher has a hypothesised structure explaining the
relationship between variables and wishes to validate
the fit of a model using data taken from a sample
(Note. CFA can still be used in an exploratory manner,
see Schmitt, 2011). On the other hand, EFA is chosen
when the researcher cannot assume, a priori, a structure
to the variables relationships and must rely on the
sample to estimate it (Matsunaga, 2010). This
commonly occurs when new assessment instruments or
scales are developed and tested for the first time.
Another related issue is the type of correlation
matrix to be analysed. Conventional EFA is based on
the Pearson correlation matrix. Pearson correlations
assume data have been measured on, at least, an equal
interval scale and a linear relationship exists between
the variables. These assumptions are typically violated
in the case of variables measured using ordinal rating
scales (Timmerman & Lorenzo-Seva, 2011). Pearson
correlations have been found to underestimate the
strength of relationships between ordinal items
(Olsson, 1979a). In relation to EFA procedures, studies
have observed Pearson correlation matrices used on
ordinal data giving rise to spurious multidimensionality
and biased factor loadings (Bernstein & Teng, 1989).
Acknowledging the limitations of the Pearson
correlation, researchers have investigated the use of an
alternate measure of correlation known as the
polychoric correlation. The polychoric correlation,
which is an extension of the tetrachoric correlation, is a
technique for estimating the correlation between two
bivariate normally distributed continuous variables
measured using an ordinal scale (Olsson, 1979a).
To demonstrate the performance of the Polychoric
correlation, a simple simulation was performed using R
(R Core Team, 2013, contact author for the simulation
script). First, a bivariate normal population distribution
was generated with a population Pearson correlation of

Page 3

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

r = .557 (Figure 1a). Next, 1000 random samples of
size n = 30 from the bivariate data, x,y, were taken
from the population. For each sample, the normally
distributed bivariate data were converted to ordinal
data measured on a five-point Likert-type scale using

generalised least squares, maximum likelihood,

principal axis factoring, alpha factoring and image
factor to name a few (Costello & Osborne, 2005). The

Figure 1a. Bivariate normal distribution, Pearson r =

.557.

arbitrary and unequal splits in the distribution (Figure

1b). These splits were determined subjectively and
should only be considered one hypothetical example of
an infinite number of ways to split a continuous scale.
Even though an underlying bivariate normal
distribution was used, the resulting ordinal conversion
resulted in highly skewed distributions. The simulation
involved calculating a Pearson correlation and a
Polychoric correlation for each sample of the ordinal
data. The mean Pearson correlation for the 1000
simulated samples resulted in Pearsons r = 0.462,
which was a substantial underestimation. On the other
hand, the Polychoric correlation was able to recover the
populations true correlation closely by returning a
sampling mean of r = 0.560. This very simple
simulation demonstrates that the polychoric correlation
is an unbiased estimate of the population Pearson
correlation for bivariate normally distributed variables
which have been measured using an ordinal scale.
Therefore, its not surprising that researchers
recommend the use of polychoric correlations for EFA
performed on data from ordinal variables (Garrido et
al., 2013; HolgadoTello et al., 2008; Timmerman &
Lorenzo-Seva, 2011).
The next major decision faced by a researcher
during EFA is deciding on the extraction method.
Common examples include unweighted least squares,

Figure 1b. A normally distributed random variable is

arbitrarily converted to an ordinal random variable
measured on a five-point Likert scale (ranging from
strongly disagree to strongly agree). Even though the
variable is normally distributed in the population, the
ordinal conversion results in a highly skewed
distribution.

method selected should be based on the nature of the

underlying distribution of the data. For example,
maximum likelihood is recommended when data are
multivariate normally distributed, while principal axis
factoring makes no distributional assumptions
(Fabrigar et al., 1999). For other methods, clear
guidelines are lacking. A common problem with EFA
extraction methods concerns the reporting of indices of
model fit including the amount of variance explained
by an EFA model. PCA allows a researcher to report
the total variance explained as a measure of goodness
of fit. PCA is able to do this because it only has to
distinguish between the explained variability of the
components and the total variability inherent in all the
observed variables. EFA can produce a measure of
goodness of fit based only on the common variance,
after error and unique variance is partitioned out, but
there are challenges with this computation for regular
EFA methods including principal axis factoring,

Page 4

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

unweighted least squares and maximum likelihood EFA
(Lorenzo-Seva, 2013). Minimum rank factor analysis
(MRFA, Shapiro & ten Berge, 2002; ten Berge & Kiers,
1991), as currently implemented only in FACTOR, is
able to estimate the percentage of common variance
explained by an EFA model. This is an attractive
property for an EFA extraction method.
Deciding how many factors to retain following
extraction has received substantial attention in the
literature. While researchers can use their own
judgement to decide on the number of factors to retain,
more often than not, researchers revert to conventional
guidelines, including the Kaiser criteria (eigenvalue > 1)
and scree plot. Unfortunately, the Kaiser criteria and
scree plot have been shown to overestimate the
number of dimensions in the data (Hubbard & Allen,
1987; Ruscio & Roche, 2012; Zwick & Velicer, 1986).
While the scree plot is substantially more accurate than
the Kaiser criteria (Zwick & Veliver, 1986), ambiguous
scree plots suffer from issues with subjectivity as
evidence by poor inter-rater reliability (Crawford &
Koopman, 1979). Parallel analysis (PA, Horn, 1965) is a
commonly recommended method for factor retention
(Fabrigar et al., 1999). Horns original PA works by
generating many (e.g. 500) random parallel datasets,
with the same number of variables and cases as the
sample dataset. Each of these parallel datasets, which
are filled with independent randomly generated data, is
analysed using PCA. The mean eigenvalues of the
factors extracted from the random parallel datasets are
compared to the samples eigenvalues. All factors
where the samples eigenvalues are greater than the
means of the random parallel datasets are retained. PA
has been found to be superior to conventional methods
for correctly identifying the true number of dimensions
(Hubbard & Allen, 1987; Ruscio & Roche, 2012; Zwick
& Velicer, 1986). PA analysis was second only to the
comparison data (CD) method in simulation studies
performed by Ruscio and Roche (2012). In FACTOR,
a powerful variation of PA using MRFA and polychoric
correlations is available. Timmerman and Lorenzo-Seva
(2011) found PA-MRFA, which is based on random
permutation of the sample data and comparing the
percentage of common variance extracted by MRFA,
outperformed Horns PA and PA based on principal
axis factoring (Humphreys & Ilgen, 1969).
Assuming a scale is multidimensional, factor
rotation will be necessary to aid the interpretation of
the model. There are two main classes for rotation,
orthogonal and oblique. Orthogonal rotation seeks to

find a solution that minimises the relationship between

factors. This method has been criticised as most factors
that make up a latent variable are expected to share
some degree of relationship (Gaskin & Happell, 2013).
Costello and Osborne (2005) also demonstrated that an
oblique rotation could be used to estimate an
orthogonal model, but not vice versa. Therefore,
oblique rotation, which allows relationships between
factors, should be preferred in most situations, unless a
strong argument can be made as to why the factors
should not be correlated (Beavers et al., 2013; Costello
& Osborne, 2005; Gaskin & Happell, 2013; Matsunaga,
2010). In FACTOR, many common and advanced
methods for oblique and orthogonal rotation are
available. For example, FACTOR provides the option
to use the oblique Promin rotation method which has
been found to perform comparably to direct oblimin
for recovering a true rotated solution and to slightly
outperforms it when fewer complex variables (variables
that load highly onto more than one factor) are present
(Lorenzo-Seva, 1999).
The following recommendations can be made
based on this brief summary on how to improve your
EFA for ordinal data:

EFA is theoretically based on exploring the

underlying relationship among variables that
comprise a scale used to measure a latent
variable.
Polychoric correlations are preferred in place of
Pearson correlations when variables are
measured using an ordinal scale. However,
polychoric correlations still make the
assumption that the ordinal variables are
derived from a bivariate normal population
distribution.
Decisions about the extraction method to be
used for EFA should be based on the scale and
shape of the datas distribution.
The Kaiser criteria and scree plot should be
avoided for factor retention decisions. Scree
plots are suitable for getting a sense of the
number of dimensions, but ambiguities can
arise. Parallel analysis-based methods are
recommended as they outperform both the
Kaiser criteria and scree plot.
By default, oblique rotation methods should be
chosen as most factors in a multidimensional
scale will share some degree of relationship.
Orthogonal rotation should only be used when

Page 5

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

there is a strong justification to assume no
relationship between factors.
Each of these recommendations will be met in the
following sections by using FACTOR. The
demonstration will use a motivating example. The
options available and output generated by FACTOR
are substantial. The aim of this article is to demonstrate
and discuss the most pertinent for getting the most out
of your EFA for ordinal data.
Demonstrating FACTOR
The Example
The example data was downloaded from the
Online Personality Tests website (http://personalitytesting.info/). This website stores the responses from
over half a million online anonymously completed
personality tests. The tests and data are made freely
available to the public for educational purposes. The
test chosen was the well-known Big 5 Personality Test
based on the Big-Five markers reported by Goldberg
(1992). According to Big 5 theory, personality is
comprised of five main traits including extraversion
(E), agreeableness (A), conscientiousness (C),
emotional stability (N) and intellect/imagination (O).
While the Big 5 has generated substantial evidence of
construct validity through previous factor analytic
studies, it was interesting to investigate if EFA
performed in FACTOR using the recommend methods
could extract the theoretically proposed factor structure
and compare these results to what would have been
obtained using a default approach in SPSS. The Big 5
dataset contained the results of more than 11,000 tests.
Computationally, this was very large for the PA
procedure that would be implemented. Therefore, a
random sample of 500 tests was analysed, which was
considered a more representative sample size used in
EFA studies. The Big 5 test used by the Online
Personality Tests website included 50 self-reported
personality items, for example, I get stressed out
easily, which were rated on an ordinal five-point scale
ranging from (1) disagree to (5) agree. The dataset used
in the following demonstration can be obtained by
contacting the author.
Preparation
The first step is to download and open FACTOR
(http://psico.fcep.urv.es/utilitats/factor/).

Unfortunately, FACTOR is only available for PCs.

FACTOR runs directly from an executable file and
does not require installation. FACTOR requires the
data to be in a simple format using the .dat file
extension. The data file must have no labels and no
extra variables, just the variables to be included in the
EFA. No missing values are permitted. If participants
have missing values, the participant must either be
removed entirely (case-wise deletion) or an appropriate
missing value imputation implemented (see Shrive,
Stuart, Quan, & Ghali, 2006). A file with descriptive
variable labels can be attached separately. The easiest
way to create a .dat file is to export your data from the
statistical package in the .dat format. Packages such as
SPSS will have built-in options for this purpose. You
can also convert common dataset formats using Excel,
but not directly. To do so, use Excel to save the dataset
as a plain text file, .txt, in a tab delimited format. Find
the dataset saved on your computer and change its
extension from .txt to .dat. A tab delimited .txt file and
a .dat file are equivalent. The file is now ready to be
used in FACTOR.
A label file can also be created to accompany the
dataset. Open a plain file in a text editor and type the
labels for the variables in column order. Each row
corresponds to a column label. Save this file as a simple
.txt file. Figure 2Figure displays an example. The Big 5
items were labelled using an item number, 0 49 and a
letter corresponding to the theoretically aligned trait.

Figure 2. FACTOR label file example.

Page 6

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Figure 3. FACTORs Main window and Read Data window.

Extraction and Retention
The EFA demonstration will be broken into two
major steps: 1) Extraction and Retention and 2)
Rotation. To begin, open FACTOR and click on the
read data button (Figure 3). Set the number of
participants and number of items. You will also need to
locate the file with the participants scores (the .dat
dataset) and the variables labels file if you have one.
Click OK and select the Configure Analysis button.

methods including the option to analyse polychoric

correlations, perform parallel analysis and execute
advanced extraction and rotation methods such as
MRFA and Promin. In the configuration menu, select
the option to analyse polychoric correlations, due to
the ordinal nature of the data. PA should also be
selected and based on the minimum rank factor
analysis (MRFA, Timmerman & Lorenzo-Seva, 2011).
Note that Horns method is available as a configurable
option (Figure 4b). The PA-MRFA is advised based on
simulation studies which have found it to outperform
Horns PA (Timmerman & Lorenzo-Seva, 2011). Other
conventional extraction methods, including PCA are
also available in addition to the useful option to save
factor scores (see DiStefano et al., 2009). In the first
step, its not important to specify the correct rotation

Figure 4b. Main FACTOR PA Configuration window

Figure 4a. Main FACTOR Configuration window
The configuration window is where the many
features of FACTOR become apparent (Figure 4a).
The practitioner has access to many powerful EFA

method or the number of factors to be retained. The

number of factors to be extracted and rotation method
can be specified in the second step once the PA has
advised the number of dimensions to retain. You can
also specify the name of the output file from the

Page 7

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

analysis to be saved in a .txt file. Click OK and hit the
compute button to run the PA. FACTOR will give you
a warning before proceeding if you specify zero
number of factors. Ignore this message and proceed.
The author advises that separate files be saved for each
step of the EFA. This enables the practitioner to
separate the results of the PA from the rotation.
PA is a simulation method and, as such, can take
considerable computational time depending on the
sample size and number of variables. For some
perspective, the PA for the Big 5 data took 16 minutes
on a Core I7 Intel processor with 8 GB RAM. Its also
common for Windows to report FACTOR to be nonresponsive even though the simulation is correctly
running in the computers memory (Figure 5). Once
the PA is complete you will be presented with a
comprehensive output file. The most pertinent output
will now be presented and discussed.

Figure 5. Parallel analysis will take some time. Be

patient. FACTOR is still responding despite the
warning.
The first part of the output lists the details of the
analysis. The log includes the data file location, number
of participants, variables, factors extracted, factor
retention method, type of correlation matrix analysed,
and the factor extraction and rotation method. The
second important table presents the items univariate
statistics (Table 1). FACTOR recommends the use of
polychoric correlations when either the skewness or
kurtosis statistics are higher than one in absolute value.
Items 6 and 8 in Table 1 exhibit kurtosis in excess of
this threshold. While such rules of thumb may be
convenient, the author advises that the decision to use
either polychoric or Pearson correlations be based on
the level of the items measurement. Also, recall that
polychoric correlation assume variables represent
bivariate normal distributions measured using an
ordinal scale. Multivariate tests for skewness and
kurtosis proposed by Mardia (1970) are also reported.
For the Big 5 data, the test for skewness was not
statistically significant, p = 1.00. However, there was
evidence of excessive kurtosis, p < .001. Mardias test
result can be used for supporting the decision to us
polychoric correlations.

Common methods for determining the suitability

of the data for factor analysis are also reported by
FACTOR. Check the polychoric correlation matrix for
evidence of non-zero correlations, a statistically
significant Bartletts test and a Kaiser-Meyer-Olkin
(KMO) statistic above .80 (Beavers et al., 2013). The
Big 5 example satisfied all these checks with evidence
of substantial correlations between items, Bartletts
statistic = 11497.3 (df = 1125), p < .001, and KMO =
0.88. On a practical note, estimating the polychoric
Table 1. FACTORs Univariate Descriptive Statistics Output
for the First 10 Big 5 Items
Confidence
Var Mean
Variance Skewness Kurtosis
Interval
V1 2.494 (2.36 2.63)
1.406
0.350
-0.742
V2 2.118 (1.98 2.26)
1.480
0.915
-0.214
V3 3.276 (3.15 3.40)
1.228
-0.243
-0.769
V4 3.166 (3.02 3.31)
1.578
-0.164
-0.973
V5 3.918 (3.80 4.04)
1.107
-0.785
-0.017
V6 2.916 (2.76 3.07)
1.841
0.061
-1.177
V7 3.850 (3.73 3.97)
1.135
-0.851
0.136
V8 3.010 (2.85 3.17)
1.874
-0.027
-1.259
V9 3.268 (3.13 3.41)
1.488
-0.278
-0.942
V10 1.944 (1.83 2.06)
1.013
0.891
0.047

correlation requires the convergence of an iterative

maximum likelihood algorithm. Researchers have noted
that the polychoric correlation estimation commonly
fails to converge (Babakus, Ferguson, & Jreskog,
1987; Timmerman & Lorenzo-Seva, 2011). When this
occurs in FACTOR, the pair of variables responsible
for non-convergence will be reported and the Pearson
correlation inserted into the correlation matrix instead.
Therefore, the correlation matrix may be a polychoric
and Pearson correlation matrix. For the Big 5 example,
all polychoric correlations converged. Once the
suitability of the data for EFA has been satisfied, its
time to interpret the results of PA-MRFA.
FACTOR will report the advised number of
dimensions based on the PA-MRFA. You will notice
that PA-MRFA is based on the random explained
common variance, and not eigenvalues as proposed by
Horn (1965). The method also uses random
permutation of the data instead of purely randomly
generated data. There are two criteria that FACTOR
will apply to make the decision for how many factors to
retain. One method is based on the mean of random
variance extracted and the other on the 95th percentile
of random percentage of variance. Simulation studies

Page 8

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

suggest that the 95th percentile criteria is more accurate
(Timmerman & Lorenzo-Seva, 2011). FACTOR
compares the mean or the 95th percentile of the factors
percentage of common variance explained from the
randomly permutated data to the observed explained
common variance from the sample. If a factors
observed percentage exceeds the random percentage,
the factor is retained. This occurred five times in the
case of the Big 5 example (Table2). The real data
percentage of variance for the first five factors
exceeded the 95th percentile of the random common
variance extracted. For factor 6, the common variance
from the 95th percentile of random variance, 3.6
exceeded the real-data percentage, 2.7, hence five
factors were retained.

the number of advised dimensions to five, select the

MRFA option for extraction, and select a rotation
method. The oblique rotation method is recommended
for EFA. Promin (Lorenzo-Seva, 1999) rotation is the
advised option by the developers of FACTOR.
However, other more familiar oblique rotation
methods are available including the commonly used
direct oblimin method. Save the output file using a
different file name to the PA output. These options are
summarised in Figure 6.

Table 2. Parallel Analysis Based On

Minimum Rank Factor Analysis of
Polychoric Correlations (Timmerman &
Lorenzo-Seva, 2011) for the Big-5 Example
Var

1
2
3
4
5
6
7
8
9
10

Real-data %
of variance

20.0*
11.8*
9.6*
8.6*
5.8*
2.7
2.6
2.2
2
1.9

Mean of
random
variance

4.2
4
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1

95 percentile
of random %
of variance

4.5
4.2
4
3.9
3.8
3.6
3.5
3.4
3.3
3.2

Note. Only the first 10 factors are shown.

Percentage of variance relates to common variance.
The factors with stars are retained as the real-data
percentage of common variance is higher than the
95th percentile of the PA-MRFAs random datasets

Factor Rotation
Once the PA has advised the number of
dimension supported by the data, the next step is to
extract the number of advised dimensions and select a
rotation option to aid in factor interpretation. To save
time, ensure you dont select the PA option in the
second step or else the PA will be run again, wasting
time. Instead, select the Minimum Average Partial
(MAP) option as this is a non-simulation method. Set

Figure 6. FACTOR Rotation Configuration for the

second step of the Big 5 example.

The overall fit of the five factor model can be

expressed by reporting the overall percentage of
common variance explained. This estimate was made
possible with the use of MRFA. For the five factor
model of the Big 5 data, the overall percentage of
common variance explained was 63.03%. The output
also reports the unrotated loading matrix, structure
matrix, communalities and rotated (pattern) matrix.
The rotated (pattern) loading matrix, structure matrix
and communalities are reported in Appendix A. All
communalities were above the minimum of .4 as

Page 9

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Table 3. A Comparison of the Big 5 Rotated Factor Solutions for FACTOR and SPSS
Item
16_N
41_N
26_N
36_N
6_N
46_N
11_N
1_N
31_N
21_N
20_E
0_E
30_E
40_E
10_E
25_E
35_E
45_E
5_E
15_E
38_C
28_C
33_C
43_C
3_C
48_C
13_C
23_C
18_C
8_C
2_A
22_A
42_A
32_A
47_A
12_A
37_A
7_A
17_A
27_A
4_O
49_O
39_O
24_O
34_O
14_O
44_O
19_O
29_O
9_O

F1
0.796
0.695
0.662
0.661
0.611
0.399
-0.437
-0.596
-0.641
-0.726

FACTORa
F2
F3

F4

F5

F1

F2

-0.316

0.791
0.769
0.758
0.684
0.653
-0.541
-0.651
-0.720
-0.802
-0.808

SPSSb
F3
0.755
0.649
0.621
0.609
0.592
0.377
-0.380
-0.555
-0.594
-0.678

F4

F5

-0.766
-0.742
-0.728
-0.662
-0.642
0.538
0.609
0.689
0.761
0.766
0.843
0.795
0.787
0.784
0.690
0.657
0.610
0.597
-0.395
-0.628

0.812
0.779
0.757
0.736
0.677
0.636
0.580
0.551
-0.373
-0.609
0.726
0.702
0.661
0.617
0.575
0.468
-0.414
-0.588
-0.593
-0.719

-0.700
-0.686
-0.617
-0.569
-0.542
-0.437
0.396
0.549
0.577
0.697
0.704
0.683
0.666
0.642
0.555
0.548
0.457
-0.479
-0.513
-0.666

0.654
0.659
0.619
0.619
0.496
0.491
0.424
-0.465
-0.481
-0.616

Reliabilityc
0.896
0.931
0.923
0.879
0.873
.816
.849
.911
.888
.839
% Varianced
23.1
13.5
10.7
8.9
6.7
Note. Rotated loading with values < .3 suppressed. a Matrix: Poloychoric correlations , Extraction: MRFA , Retention: PA,
Rotation: Promin, b Matrix: Pearson correlations, Extraction: Principal Axis Factoring, Retention: Scree plot, Rotation: Direct
oblimin. c Factor: Mislevy and Bock (1990), SPSS: Cronbachs . d Percentage of common variance explained based on
MRFA.

Page 10

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

suggested by Costello and Osbourne (2005). The
rotated (pattern) matrix, with loading below .3
suppressed is reported in Table 3. All items load onto
their theoretical personality dimension. All loadings are
checked to be above the minimum of .32 as advised by
Tabachnick and Fidell (2001). No complex or crossloading items are present.
FACTOR also reports reliability estimates based
on Mislevy and Bock (1990) and the McDonalds
Omega statistic (Table 3). The Mislevy and Bock
reliability estimate reflects the proportion of variance in
a group of items factor score accounted for by the
underlying common latent variable that drives the item
scores. In other words, imagine for each person in the
population, there was a factor score (summed or
average scores across items that compose a factor) and
an individuals true score on a latent variable. A
correlation is calculated between the factor scores and
the true latent variable. If this correlation is squared
you obtain the reliability estimate proposed by Mislevy
and Bock, which reflects the proportion of variability in
the factors scores explained by the latent variable. The
Miselvy and Bock estimate is different to the
McDonald estimate as it is based on the standardised
factor scores, whereas the McDonald estimate is based
on the unstandardised factor loadings (Zinbarg, 2006).
FACTOR does not report conventional Cronbachs
alpha statistics as estimates of internal consistency for
each factor. These estimates would need to be obtained
through another statistical package such as SPSS.

Table 4. Eigenvalues from EFA Extracted using Principle

Axis Factoring for the Big 5 Data
Rotation
Initial Eigenvalues
Loadings
Factor
% of
CumuTotal
Total
Variance
lative %
1
8.748
17.497
17.497
6.382
2
5.222
10.444
27.941
4.181
3
4.226
8.452
36.393
4.520
4
3.553
7.106
43.499
2.573
5
2.752
5.503
49.002
5.781
6
1.320
2.641
51.643
3.293
7
1.305
2.611
54.254
1.116
8
1.126
2.252
56.506
.920
9
1.004
2.007
58.513
2.227
10
.946
1.893
60.406
11
.921
1.841
62.247
12
.915
1.830
64.076

SPSS
The EFA for the Big 5 data was performed in
SPSS using principal axis factoring. This allowed a
comparison using the standard methods available in
SPSS. Had the conventional Kaiser Criteria been
followed, the practitioner would have extracted 9
factors (Table 4). The scree plot suggested the correct
five factors (Figure 7), however, such an unambiguous
plot shouldnt be expected in practice. Ambiguities are
common. Scree plots may provide an initial idea of the
number of dimensions, but PA analysis should be
preferred. PA has been shown to be more accurate and
avoids issues with subjectivity. The five factor solution
is presented alongside the five factor model discovered
using FACTOR in Table 3 (Appendix B contains the
complete results including the rotated loading matrix,
structure matrix and communalities). The major
difference between the two solutions concerns the
factor loadings. The EFA in SPSS was based on

Figure 7. Scree plot of factor eigenvalues for EFA using

principal axis factor for factor extraction. The eigenvalues
drop-off substantially after the fifth factor, therefore, five
factors are advised. This was an example of an
unambiguous scree plot, however, in practice, ambiguities
can arise. Scree plots may be suitable for getting an idea
of the number of dimensions, but PA should be preferred

Pearson correlations and as a result the resulting factor

loadings are systematically lower than the factor
loadings based on the polychoric correlations matrix
analysed in FACTOR. A cross-loading also appears for
item 6_N. In conclusion, it was interesting to note that
both FACTOR and SPSS EFAs arrived at the correct
theoretical model. Therefore, a practitioner might
consider what was gained by performing the
recommended EFA practices through FACTOR. The
EFA performed in FACTOR provided objective
evidence to retain five factors, demonstrated

Page 11

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

systematically higher factor loadings, contained no
complex factors and provided an overall indicator of
model fit.
Conclusions
A major barrier to the implementation of
recommended EFA practices relates to a lack of
features in major software, particularly SPSS, and,
perhaps, a lack of clear how-to demonstrations. While
SPSSs EFA features can be greatly enhanced with the
R-menu plugin, alternatives are available. This paper
demonstrated recommended methods for improving
your EFA for ordinal data using FACTOR, a free,
powerful and standalone EFA package for the PC.
FACTOR provides the practitioner with a powerful
arsenal of EFA options including the use of polychoric
correlations, parallel analysis and minimum rank factor
analysis. This demonstration concentrated on the
salient features of FACTOR that are most likely to
assist the EFA practitioner. FACTOR includes other
advanced options including semi-specified target
matrices of the factor loadings, second order factors
and advanced output for evaluating EFA models.
Despite these features there are some minor limitations
of the current version. FACTOR will run only on PC,
does not accept many standard data formats, the PA
procedures take a long time to simulate and the very
promising comparison dataset (CD) method (Ruscio &
Roche, 2012) for factor retention is yet to be
implemented. Yet, FACTORs benefits far outweigh
its caveats. FACTOR is free, packaged with many
powerful and cutting edge EFA procedures and is
relatively easy to use. It is hoped that this
demonstration helps other researchers to improve their
EFA for ordinal data by implementing recommended
procedures that are commonly inaccessible.
References
Babakus, E., Ferguson, J. C. E., & Jreskog, K. G. (1987).
The sensitivity of confirmatory maximum likelihood
factor analysis to violations of measurement scale and
distributional assumptions. Journal of Marketing Research,
24(2), 222228.
Basto, M., & Pereira, J. M. (2012). An SPSS R-Menu for
ordinal factor analysis. Journal of Statistical Software, 46(4),
129.
Beavers, A. S., Lounsbury, J. W., Richards, J. K., Huck, S.
W., Skolits, G. J., & Esquivel, S. L. (2013). Practical
considerations for using exploratory factor analysis in
educational research. Practical Assessment, Research &
Evaluation, 18(6), 113. Retrieved from
http://pareonline.net/pdf/v18n6.pdf

Bernstein, I. H., & Teng, G. (1989). Factoring items and

factoring scales are different: Spurious evidence for
multidimensionality due to item categorization.
Psychological Bulletin, 105(3), 467477. doi:10.1037/00332909.105.3.467
Costello, A. B., & Osborne, J. W. (2005). Best practices in
exploratory factor analysis: Four recommendations for
getting the most from your analysis. Practical Assessment,
Research & Evaluation, 10(7), 19. Retrieved from
http://pareonline.net/pdf/v10n7.pdf
Courtney, M. G. R. (2013). Determining the number of
factors to retain in EFA: Using the SPSS R-Menu v2.0
to make more judicious estimations. Practical Assessment,
Research & Evaluation, 18(8), 114. Retrieved from
http://pareonline.net/pdf/v18n8.pdf
Crawford, C. B., & Koopman, P. (1979). Note: Inter-rater
reliabiliy of scree test and mean square ratio test of
number of factors. Perceptual and Motor Skills, 49(1),
223226. doi:10.2466/pms.1979.49.1.223
DiStefano, C., Zhu, M., & Mindrilla, D. (2009).
Understanding and using factor scores: Considerations
for the applied researcher. Practical Assessment, Research
& Evaluation, 14(20), 111. Retrieved from
http://pareonline.net/pdf/v14n20.pdf
Fabrigar, L. R., Wegener, D. T., MacCallum, R. C., &
Strahan, E. J. (1999). Evaluating the use of exploratory
factor analysis in psychological research. Psychological
Methods, 4(3), 272299. doi:10.1037/1082-989X.4.3.272
Garrido, L. E., Abad, F. J., & Ponsoda, V. (2013). A new
look at Horns parallel analysis with ordinal variables.
Psychological Methods, 18(4), 45474.
doi:10.1037/a0030005
Gaskin, C. J., & Happell, B. (2013). On exploratory factor
analysis: A review of recent evidence, an assessment of
current practice, and recommendations for future use.
International Journal of Nursing Studies.
doi:10.1016/j.ijnurstu.2013.10.005
Goldberg, L. R. (1992). The development of the markers for
the Big-Five factor structure. Psychological Assessment, 4,
2642.
HolgadoTello, F. P., ChacnMoscoso, S., Barbero
Garca, I., & VilaAbad, E. (2008). Polychoric versus
Pearson correlations in exploratory and confirmatory
factor analysis of ordinal variables. Quality & Quantity,
44(1), 153166. doi:10.1007/s11135-008-9190-y
Horn, J. L. (1965). A rationale and test for the number of
factors in factor analysis. Psychometrika, 30(2), 179185.
doi:10.1007/BF02289447
Hubbard, R., & Allen, S. J. (1987). An empirical comparison
of alternative methods for principal component
extraction. Journal of Business Research, 15(2), 173190.
doi:10.1016/0148-2963(84)90047-X

Page 12

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Humphreys, L. G., & Ilgen, D. R. (1969). Note on a
criterion for the number of common factors.
Educational and Psychological Measurement, 29(3), 571578.
doi:10.1177/001316446902900303
IBM Corp. (2012). IBM SPSS Statistics 21 for Windows.
Armonk, NY: IBM Corp.
Lorenzo-Seva, U. (1999). Promin: A Method for Oblique
Factor Rotation. Multivariate Behavioral Research, 34(3),
347365. doi:10.1207/S15327906MBR3403_3
Lorenzo-Seva, U. (2013). How to report the percentage of
explained common variance in exploratory factor analysis.
Technical report. Tarragona. Retrieved from
http://psico.fcep.urv.cat/utilitats/factor/
Lorenzo-Seva, U., & Ferrando, P. J. (2006). FACTOR: A
computer program to fit the exploratory factor analysis
model. Behavior Research Methods, 38(1), 8891.
Mardia, K. V. (1970). Measures of multivariate skewness and
kurtosis with applications. Biometrika, 57(3), 519530.
doi:10.1093/biomet/57.3.519
Matsunaga, M. (2010). How to factor-analyse your data
right: Dos dont's, and how-to's. International Journal of
Psychological Research, 3(1), 97110.
Mislevy, R. J., & Bock, R. D. (1990). BILOG 3: Item analysis
and test scoring with binary logistic regression models.
Mooresville, IN: Scientific Software.
Olsson, U. (1979a). Maximum likelihood estimation of the
polychoric correlation coefficient. Psychometrika, 44(4),
443460.
Olsson, U. (1979b). On the robustness of factor analysis
against crude classification of the observations.
Multivariate Behavioral Research, 14(4), 485500.
doi:10.1207/s15327906mbr1404_7
Osborne, J. W., & Fitzpatrick, D. C. (2012). Replication
analysis in exploratory factor analysis: What it is and
why it makes your analysis better. Practical Assessment,
Research & Evaluation, 17(15), 18. Retrieved from
http://pareonline.net/pdf/v17n15.pdf

R Core Team. (2013). R: A language and environment for

statistical computing. Vienna, Austria. Retrieved from
http://www.r-project.org
Ruscio, J., & Roche, B. (2012). Determining the number of
factors to retain in an exploratory factor analysis using
comparison data of known factorial structure.
Psychological Assessment, 24(2), 28292.
doi:10.1037/a0025697
Schmitt, T. A. (2011). Current methodological
considerations in exploratory and confirmatory factor
analysis. Journal of Psychoeducational Assessment, 29(4),
304321. doi:10.1177/0734282911406653
Shapiro, A., & ten Berge, J. M. F. (2002). Statistical inference
of minimum rank factor analysis. Psychometrika, 67(1),
7994. doi:10.1007/BF02294710
Shrive, F. M., Stuart, H., Quan, H., & Ghali, W. A. (2006).
Dealing with missing data in a multi-question
depression scale: A comparison of imputation
methods. BMC Medical Research Methodology, 57.
doi:doi:10.1186/1471-2288-6-57
Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate
statistics (4th ed.). Boston, MA: Allyn & Bacon.
ten Berge, J. M. F., & Kiers, A. L. (1991). A numerical
approach to the approximate and the exact minimum
rank of a covariance matrix. Psychometrika, 56(2), 309
315.
Timmerman, M. E., & Lorenzo-Seva, U. (2011).
Dimensionality assessment of ordered polytomous
items with parallel analysis. Psychological Methods, 16(2),
20920. doi:10.1037/a0023353
Zinbarg, R. E. (2006). Estimating generalizability to a latent
variable common to all of a scales indicators: A
comparison of estimators for h. Applied Psychological
Measurement, 30(2), 121144.
doi:10.1177/0146621605278814
Zwick, W. R., & Velicer, W. F. (1986). Comparison of five
rules for determining the number of components to
retain. Psychological Bulletin, 99(3), 432442.
doi:10.1037/0033-2909.99.3.432.

Page 13

Practical Assessment, Research & Evaluation, Vol 19, No 5

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Appendix A: Complete EFA Results for Big 5 Example Analysed Using FACTOR
Rotated (Pattern) Matrixa
Structure Matrixa
Variable
F1
F2
F3
F4
F5
F1
F2
F3
F4
0_E
-0.033
0.769
0.031 -0.018
0.052
0.215
0.761 -0.200 -0.027
1_N
-0.596
0.078 -0.036
0.037 -0.086
-0.574 -0.121
0.001 -0.040
2_A
-0.026 -0.034
0.022
0.726
0.018
0.075 -0.038 -0.158
0.715
3_C
0.122 -0.095
0.690 -0.119 -0.098
-0.010 -0.288
0.749 -0.278
4_O
-0.039 -0.047 -0.049 -0.047
0.704
0.046
0.120 -0.102 -0.077
5_E
-0.048 -0.802 -0.100
0.043
0.050
-0.281 -0.776
0.129
0.052
6_N
0.611
0.268
0.007 -0.022
0.069
0.702
0.477 -0.137
0.069
7_A
0.067
0.034
0.055 -0.588
0.095
-0.004
0.056
0.182 -0.597
8_C
-0.027
0.065 -0.628 -0.096
0.018
0.044
0.249 -0.622
0.065
9_O
0.010
0.092
0.181
0.066 -0.666
-0.066 -0.114
0.214
0.055
10_E
0.110
0.653 -0.229
0.018 -0.045
0.338
0.747 -0.437
0.103
11_N
-0.437
0.265
0.208 -0.115
0.144
-0.370
0.095
0.185 -0.242
12_A
0.007 -0.067
0.063
0.468
0.253
0.088 -0.020 -0.071
0.438
13_C
0.201 -0.129
0.610
0.057
0.025
0.111 -0.242
0.611 -0.075
14_O
-0.013
0.067
0.159 -0.072
0.548
0.061
0.143
0.094 -0.144
15_E
0.074 -0.808
0.008 -0.005
0.095
-0.172 -0.765
0.235 -0.007
16_N
0.796 -0.112
0.025
0.046 -0.020
0.762
0.130 -0.030
0.161
17_A
-0.036 -0.002
0.264 -0.593
0.088
-0.141 -0.078
0.415 -0.673
18_C
-0.097
0.208 -0.395
0.055 -0.114
0.001
0.270 -0.450
0.152
19_O
-0.048
0.076
0.108
0.171 -0.479
-0.077 -0.082
0.102
0.161
20_E
0.115
0.791
0.109
0.079 -0.020
0.366
0.791 -0.160
0.076
21_N
-0.726
0.019
0.025
0.121 -0.003
-0.705 -0.220
0.061
0.003
22_A
0.018
0.077
0.024
0.702 -0.028
0.143
0.075 -0.183
0.701
23_C
-0.067
0.093
0.597
0.005 -0.143
-0.117 -0.142
0.591 -0.154
24_O
-0.076
0.161 -0.042
0.181
0.642
0.101
0.301 -0.207
0.149
25_E
-0.183 -0.541
0.011
0.011 -0.191
-0.384 -0.648
0.212 -0.014
26_N
0.662 -0.083
0.145 -0.014 -0.127
0.601
0.055
0.122
0.055
27_A
0.078
0.029
0.072 -0.719
0.058
-0.022
0.039
0.238 -0.729
28_C
0.057
0.032
0.795
0.011 -0.090
-0.024 -0.211
0.787 -0.185
29_O
-0.148 -0.013 -0.018 -0.011 -0.513
-0.227 -0.175
0.064 -0.003
30_E
0.043
0.758
0.005 -0.044
0.034
0.283
0.778 -0.221 -0.034
31_N
-0.641 -0.237
0.006
0.063
0.024
-0.705 -0.438
0.123 -0.040
32_A
-0.008 -0.025
0.165
0.617
0.017
0.065 -0.067
0.008
0.571
33_C
-0.115
0.150
0.787 -0.006 -0.004
-0.147 -0.125
0.756 -0.230
34_O
-0.047 -0.019 -0.069
0.155
0.555
0.058
0.118 -0.165
0.137
35_E
0.062 -0.651 -0.040
0.038
0.022
-0.133 -0.614
0.138
0.051
36_N
0.661 -0.031 -0.050
0.045 -0.056
0.655
0.183 -0.112
0.162
37_A
-0.117 -0.026
0.057 -0.414 -0.047
-0.201 -0.095
0.191 -0.445
38_C
-0.083
0.112
0.843
0.040
0.011
-0.124 -0.166
0.805 -0.194
39_O
-0.075 -0.070 -0.055 -0.128
0.666
-0.015
0.077 -0.071 -0.160
40_E
-0.050
0.684 -0.029
0.005
0.058
0.180
0.691 -0.239
0.008
41_N
0.695
0.025
0.091
0.060
0.071
0.713
0.236 -0.010
0.139
42_A
0.082
0.005
0.050
0.661 -0.094
0.166
0.000 -0.123
0.665
43_C
-0.178
0.123
0.784
0.005
0.027
-0.213 -0.164
0.760 -0.229
44_O
0.117 -0.201
0.199
0.044
0.457
0.106 -0.116
0.182 -0.016
45_E
0.075 -0.720
0.014 -0.018
0.011
-0.158 -0.698
0.227 -0.017
46_N
0.399
0.264 -0.189
0.031
0.065
0.517
0.464 -0.325
0.141
47_A
-0.004
0.009
0.122
0.575
0.158
0.098
0.013 -0.050
0.534
48_C
0.026 -0.185
0.657 -0.059
0.113
-0.092 -0.349
0.713 -0.236
49_O
-0.045
0.162
0.069
0.082
0.683
0.112
0.288 -0.078
0.023
Note. a Matrix: Poloychoric correlations, Extraction: MRFA, Retention: PA, Rotation: Promin

F5
0.225
-0.152
-0.034
-0.178
0.696
-0.135
0.221
0.137
0.109
-0.668
0.150
0.124
0.207
-0.051
0.547
-0.084
0.064
0.082
-0.035
-0.489
0.164
-0.114
-0.046
-0.202
0.664
-0.346
-0.067
0.105
-0.169
-0.535
0.219
-0.129
-0.042
-0.079
0.544
-0.118
0.036
-0.055
-0.077
0.652
0.214
0.164
-0.121
-0.063
0.401
-0.147
0.206
0.116
-0.002
0.702

Communality
0.765
0.608
0.715
0.805
0.838
0.860
0.791
0.680
0.609
0.853
0.825
0.645
0.550
0.678
0.809
0.802
0.802
0.744
0.668
0.561
0.860
0.764
0.673
0.751
0.748
0.777
0.663
0.739
0.794
0.674
0.848
0.895
0.649
0.814
0.765
0.653
0.747
0.503
0.867
0.810
0.664
0.799
0.745
0.894
0.574
0.684
0.766
0.574
0.797
0.771

Practical Assessment, Research & Evaluation, Vol 19, No 5

Page 14

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Appendix B Complete EFA Results for Big 5 Example Analysed using SPSS
Rotated (Pattern) Matrixa
Structure Matrixa
Variable
F1
F2
F3
F4
F5
F1
F2
F3
F4
F5
0_E
-0.742
0.016
0.000
0.064
0.004
-0.752
0.035
0.148
0.187 -0.184
1_N
-0.013 -0.025 -0.555 -0.096 -0.030
0.104
0.050 -0.556 -0.146 -0.004
2_A
0.030 -0.700 -0.027 -0.011 -0.011
0.048 -0.699
0.074 -0.044 -0.148
3_C
0.102
0.132
0.112 -0.105
0.677
0.267
0.251
0.027 -0.145
0.731
4_O
0.006
0.039 -0.035
0.654 -0.060
-0.112
0.056
0.024
0.655 -0.086
5_E
0.761 -0.044 -0.082
0.033 -0.068
0.755 -0.060 -0.217 -0.098
0.118
6_N
-0.316
0.012
0.592
0.087 -0.013
-0.449 -0.071
0.661
0.198 -0.126
7_A
-0.039
0.549
0.062
0.114
0.079
-0.061
0.561 -0.007
0.142
0.175
8_C
-0.074
0.078 -0.022
0.036 -0.609
-0.232 -0.044
0.015
0.084 -0.612
9_O
-0.040 -0.051
0.011 -0.616
0.177
0.106 -0.037 -0.041 -0.620
0.191
10_E
-0.642 -0.025
0.132 -0.026 -0.246
-0.725 -0.085
0.271
0.107 -0.419
11_N
-0.197
0.123 -0.380
0.128
0.187
-0.100
0.229 -0.358
0.118
0.175
12_A
0.050 -0.437
0.007
0.221
0.030
0.028 -0.424
0.083
0.196 -0.062
13_C
0.120 -0.038
0.185
0.009
0.580
0.230
0.053
0.139 -0.028
0.593
14_O
-0.084
0.071 -0.004
0.491
0.127
-0.134
0.118
0.043
0.500
0.092
15_E
0.766
0.004
0.037
0.082
0.033
0.753 -0.007 -0.106 -0.044
0.221
16_N
0.036 -0.053
0.755 -0.006
0.014
-0.105 -0.166
0.755
0.058 -0.026
17_A
0.012
0.577 -0.033
0.101
0.281
0.061
0.644 -0.128
0.101
0.401
18_C
-0.187 -0.060 -0.075 -0.094 -0.373
-0.251 -0.127 -0.019 -0.051 -0.424
19_O
-0.038 -0.153 -0.040 -0.465
0.103
0.076 -0.142 -0.059 -0.474
0.090
20_E
-0.766 -0.076
0.143 -0.009
0.072
-0.773 -0.067
0.299
0.125 -0.144
21_N
0.053 -0.100 -0.678 -0.025
0.027
0.198
0.006 -0.677 -0.104
0.055
22_A
-0.081 -0.686
0.017 -0.051 -0.007
-0.064 -0.690
0.133 -0.061 -0.170
23_C
-0.048
0.021 -0.054 -0.148
0.551
0.127
0.140 -0.090 -0.176
0.555
24_O
-0.180 -0.178 -0.063
0.619 -0.071
-0.284 -0.156
0.062
0.640 -0.186
25_E
0.538 -0.007 -0.197 -0.197
0.040
0.620
0.013 -0.322 -0.308
0.196
26_N
0.027
0.009
0.621 -0.113
0.136
-0.041 -0.062
0.597 -0.065
0.120
27_A
-0.030
0.697
0.077
0.085
0.099
-0.048
0.710 -0.020
0.118
0.229
28_C
-0.014
0.008
0.056 -0.105
0.779
0.189
0.159
0.008 -0.142
0.780
29_O
0.055
0.015 -0.138 -0.481
0.001
0.162
0.018 -0.197 -0.503
0.053
30_E
-0.728
0.040
0.073
0.050 -0.023
-0.757
0.041
0.215
0.181 -0.205
31_N
0.285 -0.044 -0.594 -0.002
0.020
0.407
0.044 -0.644 -0.109
0.113
32_A
0.027 -0.569 -0.004 -0.013
0.122
0.072 -0.544
0.069 -0.046
0.010
33_C
-0.112
0.027 -0.102 -0.021
0.757
0.102
0.203 -0.125 -0.054
0.740
34_O
-0.013 -0.141 -0.037
0.496 -0.087
-0.108 -0.135
0.039
0.495 -0.147
35_E
0.609 -0.037
0.029
0.011 -0.017
0.598 -0.057 -0.082 -0.088
0.127
36_N
-0.035 -0.055
0.609 -0.037 -0.056
-0.160 -0.160
0.623
0.028 -0.105
37_A
0.041
0.396 -0.112 -0.034
0.079
0.081
0.427 -0.188 -0.042
0.181
38_C
-0.080 -0.018 -0.076 -0.008
0.812
0.142
0.165 -0.099 -0.050
0.792
39_O
0.034
0.113 -0.069
0.619 -0.058
-0.073
0.134 -0.031
0.615 -0.058
40_E
-0.662 -0.008 -0.022
0.068 -0.051
-0.682
0.001
0.117
0.178 -0.224
41_N
-0.092 -0.065
0.649
0.080
0.071
-0.213 -0.144
0.681
0.151 -0.004
42_A
-0.009 -0.617
0.079 -0.115
0.013
0.010 -0.630
0.163 -0.129 -0.116
43_C
-0.073
0.024 -0.157
0.005
0.736
0.142
0.204 -0.183 -0.039
0.730
44_O
0.159 -0.038
0.101
0.424
0.179
0.115 -0.004
0.108
0.396
0.181
45_E
0.689
0.018
0.042
0.002
0.037
0.689
0.005 -0.096 -0.110
0.213
46_N
-0.297 -0.043
0.377
0.081 -0.196
-0.432 -0.133
0.459
0.176 -0.305
47_A
-0.015 -0.542 -0.001
0.130
0.084
-0.004 -0.519
0.092
0.107 -0.041
48_C
0.186
0.073
0.015
0.093
0.636
0.327
0.204 -0.056
0.030
0.692
49_O
-0.182 -0.081 -0.035
0.659
0.039
-0.273 -0.040
0.074
0.681 -0.061
Note. a Matrix: Pearson correlations, Extraction: Principal Axis Factoring, Retention: Scree plot, Rotation: Oblique

Communality
0.570
0.320
0.491
0.573
0.434
0.584
0.551
0.340
0.389
0.413
0.604
0.232
0.229
0.395
0.276
0.577
0.575
0.500
0.219
0.251
0.628
0.473
0.486
0.334
0.484
0.465
0.393
0.527
0.621
0.278
0.582
0.499
0.313
0.568
0.277
0.361
0.399
0.209
0.637
0.399
0.472
0.483
0.415
0.560
0.230
0.478
0.381
0.292
0.517
0.497

Practical Assessment, Research & Evaluation, Vol 19, No 5

Page 15

Baglin, IMPROVING YOUR EXPLORATORY FACTOR ANALYSIS

Citation:
Baglin, James (2014). Improving Your Exploratory Factor Analysis for Ordinal Data: A Demonstration Using
FACTOR. Practical Assessment, Research & Evaluation, 19(5). Available online:
http://pareonline.net/getvn.asp?v=19&n=5.
Author:
James Baglin
School of Mathematical and Geospatial Sciences
RMIT University, Melbourne, Australia
james.baglin [at] rmit.edu.au