15

Factor analysis

Factor analysis is different from many of the other techniques presented in this book.
It is not designed to test hypotheses or to tell you whether one group is significantly
different from another. It is included in SPSS as a ‘data reduction’ technique. It takes
a large set of variables and looks for a way the data may be ‘reduced’ or summarised
using a smaller set of factors or components. It does this by looking for ‘clumps’ or
groups among the intercorrelations of a set of variables. This is an almost impossible
task to do ‘by eye’ with anything more than a small number of variables.
This family of factor analytic techniques has a number of different uses. It is used
extensively by researchers involved in the development and evaluation of tests and
scales. The scale developer starts with a large number of individual scale items and
questions and, by using factor analytic techniques, they can refine and reduce these
items to form a smaller number of coherent subscales. Factor analysis can also be used
to reduce a large number of related variables to a more manageable number, prior to
using them in other analyses such as multiple regression or multivariate analysis of
variance.
There are two main approaches to factor analysis that you will see described in the
literature—exploratory and confirmatory. Exploratory factor analysis is often used in
the early stages of research to gather information about (explore) the interrelation-
ships among a set of variables. Confirmatory factor analysis, on the other hand, is a
more complex and sophisticated set of techniques used later in the research process
to test (confirm) specific hypotheses or theories concerning the structure underlying
a set of variables.
The term ‘factor analysis’ encompasses a variety of different, although related,
techniques. One of the main distinctions is between what is termed principal com-
ponents analysis (PCA) and factor analysis (FA). These two sets of techniques are
similar in many ways and are often used interchangeably by researchers. Both attempt
to produce a smaller number of linear combinations of the original variables in a way

181
182 Statistical techniques to explore relationships among variables

that captures (or accounts for) most of the variability in the pattern of correlations.
They do differ in a number of ways, however. In principal components analysis the
original variables are transformed into a smaller set of linear combinations, with all
of the variance in the variables being used. In factor analysis, however, factors are esti-
mated using a mathematical model, whereby only the shared variance is analysed (see
Tabachnick & Fidell 2007, Chapter 13, for more information on this).
Although both approaches (PCA and FA) often produce similar results, books
on the topic often differ in terms of which approach they recommend. Stevens (1996,
pp. 362–3) admits a preference for principal components analysis and gives a number of
reasons for this. He suggests that it is psychometrically sound and simpler mathemati-
cally, and it avoids some of the potential problems with ‘factor indeterminacy’ associated
with factor analysis (Stevens 1996, p. 363). Tabachnick and Fidell (2007), in their review
of PCA and FA, conclude: ‘If you are interested in a theoretical solution uncontaminated
by unique and error variability … FA is your choice. If, on the other hand, you simply
want an empirical summary of the data set, PCA is the better choice’ (p. 635).
I have chosen to demonstrate principal components analysis in this chapter. If you
would like to explore the other approaches further, see Tabachnick and Fidell (2007).
Note: although PCA technically yields components, many authors use the term
‘factor’ to refer to the output of both PCA and FA. So don’t assume, if you see the
term ‘factor’ when you are reading journal articles, that the author has used FA. Factor
analysis is used as a general term to refer to the entire family of techniques.
Another potential area of confusion involves the use of the word ‘factor’, which
has different meanings and uses in different types of statistical analyses. In factor
analysis, it refers to the group or clump of related variables; in analysis of variance
techniques, it refers to the independent variable. These are very different things,
despite having the same name, so keep the distinction clear in your mind when you
are performing the different analyses.

STEPS INVOLVED IN FACTOR ANALYSIS

There are three main steps in conducting factor analysis (I am using the term
in a general sense to indicate any of this family of techniques, including principal
components analysis).

Step 1: Assessment of the suitability of the data for factor

analysis
There are two main issues to consider in determining whether a particular data set is
suitable for factor analysis: sample size, and the strength of the relationship among
the variables (or items). While there is little agreement among authors concerning
how large a sample should be, the recommendation generally is: the larger, the better.
 Factor analysis 183

In small samples, the correlation coefficients among the variables are less reliable,
tending to vary from sample to sample. Factors obtained from small data sets do not
generalise as well as those derived from larger samples. Tabachnick and Fidell (2007)
review this issue and suggest that ‘it is comforting to have at least 300 cases for factor
analysis’ (p. 613). However, they do concede that a smaller sample size (e.g. 150 cases)
should be sufficient if solutions have several high loading marker variables (above
.80). Stevens (1996, p. 372) suggests that the sample size requirements advocated by
researchers have been reducing over the years as more research has been done on the
topic. He makes a number of recommendations concerning the reliability of factor
structures and the sample size requirements (see Stevens 1996, Chapter 11).
Some authors suggest that it is not the overall sample size that is of concern—
rather, the ratio of participants to items. Nunnally (1978) recommends a 10 to 1 ratio;
that is, ten cases for each item to be factor analysed. Others suggest that five cases for
each item are adequate in most cases (see discussion in Tabachnick & Fidell 2007).
I would recommend that you do more reading on the topic, particularly if you have a
small sample (smaller than 150) or lots of variables.
The second issue to be addressed concerns the strength of the intercorrelations
among the items. Tabachnick and Fidell recommend an inspection of the correlation
matrix for evidence of coefficients greater than .3. If few correlations above this level
are found, factor analysis may not be appropriate. Two statistical measures are also
generated by SPSS to help assess the factorability of the data: Bartlett’s test of spheric-
ity (Bartlett 1954), and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy
(Kaiser 1970, 1974). Bartlett’s test of sphericity should be significant (p < .05) for the
factor analysis to be considered appropriate. The KMO index ranges from 0 to 1, with .6
suggested as the minimum value for a good factor analysis (Tabachnick & Fidell 2007).

Step 2: Factor extraction

Factor extraction involves determining the smallest number of factors that can be
used to best represent the interrelationships among the set of variables. There are a
variety of approaches that can be used to identify (extract) the number of underlying
factors or dimensions. Some of the most commonly available extraction techniques
(this always conjures up the image for me of a dentist pulling teeth!) are: principal
components; principal factors; image factoring; maximum likelihood factoring; alpha
factoring; unweighted least squares; and generalised least squares.
The most commonly used approach is principal components analysis. This will
be demonstrated in the example given later in this chapter. It is up to the researcher to
determine the number of factors that he/she considers best describes the underlying
relationship among the variables. This involves balancing two conflicting needs: the
need to find a simple solution with as few factors as possible; and the need to explain
as much of the variance in the original data set as possible. Tabachnick and Fidell
184 Statistical techniques to explore relationships among variables

(2007) recommend that researchers adopt an exploratory approach, experimenting

with different numbers of factors until a satisfactory solution is found.
There are a number of techniques that can be used to assist in the decision concern-
ing the number of factors to retain: Kaiser’s criterion; scree test; and parallel analysis.

Kaiser’s criterion
One of the most commonly used techniques is known as Kaiser’s criterion, or the
eigenvalue rule. Using this rule, only factors with an eigenvalue of 1.0 or more are
retained for further investigation (this will become clearer when you see the example
presented in this chapter). The eigenvalue of a factor represents the amount of the
total variance explained by that factor. Kaiser’s criterion has been criticised, however,
as resulting in the retention of too many factors in some situations.

Scree test
Another approach that can be used is Catell’s scree test (Catell 1966). This involves
plotting each of the eigenvalues of the factors (SPSS does this for you) and inspecting
the plot to find a point at which the shape of the curve changes direction and becomes
horizontal. Catell recommends retaining all factors above the elbow, or break in the plot,
as these factors contribute the most to the explanation of the variance in the data set.

Parallel analysis
An additional technique gaining popularity, particularly in the social science literature
(e.g. Choi, Fuqua & Griffin 2001; Stober 1998), is Horn’s parallel analysis (Horn 1965).
Parallel analysis involves comparing the size of the eigenvalues with those obtained from
a randomly generated data set of the same size. Only those eigenvalues that exceed the
corresponding values from the random data set are retained. This approach to iden-
tifying the correct number of components to retain has been shown to be the most
accurate, with both Kaiser’s criterion and Catell’s scree test tending to overestimate the
number of components (Hubbard & Allen 1987; Zwick & Velicer 1986). If you intend
to publish your results in a journal article in the psychology or education fields you will
need to use, and report, the results of parallel analysis. Many journals (e.g. Educational
and Psychological Measurement, Journal of Personality Assessment) are now making
it a requirement before they will consider a manuscript for publication. These three
techniques are demonstrated in the worked example presented later in this chapter.

Step 3: Factor rotation and interpretation

Once the number of factors has been determined, the next step is to try to interpret them.
To assist in this process, the factors are ‘rotated’. This does not change the underlying
solution—rather, it presents the pattern of loadings in a manner that is easier to inter-
pret. SPSS does not label or interpret each of the factors for you. It just shows you which
 Factor analysis 185

variables ‘clump together’. From your understanding of the content of the variables (and
underlying theory and past research), it is up to you to propose possible interpretations.
There are two main approaches to rotation, resulting in either orthogonal (uncor-
related) or oblique (correlated) factor solutions. According to Tabachnick and Fidell
(2007), orthogonal rotation results in solutions that are easier to interpret and to
report; however, they do require the researcher to assume (usually incorrectly) that
the underlying constructs are independent (not correlated). Oblique approaches
allow for the factors to be correlated, but they are more difficult to interpret, describe
and report (Tabachnick & Fidell 2007, p. 638). In practice, the two approaches (or-
thogonal and oblique) often result in very similar solutions, particularly when the
pattern of correlations among the items is clear (Tabachnick & Fidell 2007). Many
researchers conduct both orthogonal and oblique rotations and then report the
clearest and easiest to interpret. I always recommend starting with an oblique rotation
to check the degree of correlation between your factors.
Within the two broad categories of rotational approaches there are a number of
different techniques provided by SPSS (orthogonal: Varimax, Quartimax, Equamax;
oblique: Direct Oblimin, Promax). The most commonly used orthogonal approach
is the Varimax method, which attempts to minimise the number of variables that
have high loadings on each factor. The most commonly used oblique technique is
Direct Oblimin. For a comparison of the characteristics of each of these approaches,
see Tabachnick and Fidell (2007, p. 639). In the example presented in this chapter,
Oblimin rotation will be demonstrated.
Following rotation you are hoping for what Thurstone (1947) refers to as ‘simple
structure’. This involves each of the variables loading strongly on only one component,
and each component being represented by a number of strongly loading variables.
This will help you interpret the nature of your factors by checking the variables that
load strongly on each of them.

Additional resources
In this chapter, only a very brief overview of factor analysis is provided. Although
I have attempted to simplify it here, factor analysis is actually a sophisticated and
complex family of techniques. If you are intending to use factor analysis with your
own data, I suggest that you read up on the technique in more depth. For a thorough,
but easy-to-follow, book on the topic I recommend Pett, Lackey and Sullivan (2003).
For a more complex coverage, see Tabachnick and Fidell (2007).

DETAILS OF EXAMPLE
To demonstrate the use of factor analysis, I will explore the underlying structure of
one of the scales included in the survey4ED.sav data file provided on the website
186 Statistical techniques to explore relationships among variables

accompanying this book. One of the scales used was the Positive and Negative Affect
Scale (PANAS: Watson, Clark & Tellegen 1988) (see Figure 15.1). This scale consists
of twenty adjectives describing different mood states, ten positive (e.g. proud, active,
determined) and ten negative (e.g. nervous, irritable, upset). The authors of the
scale suggest that the PANAS consists of two underlying dimensions (or factors):
positive affect and negative affect. To explore this structure with the current commu-
nity sample the items of the scale will be subjected to principal components analysis
(PCA), a form of factor analysis that is commonly used by researchers interested in
scale development and evaluation.
If you wish to follow along with the steps described in this chapter, you should
start SPSS and open the file labelled survey4ED.sav on the website that accompanies
this book. The variables that are used in this analysis are labelled pn1 to pn20. The
scale used in the survey is presented in Figure 15.1. You will need to refer to these
individual items when attempting to interpret the factors obtained. For full details
and references for the scale, see the Appendix.
Example of research question: What is the underlying factor structure of the Positive
and Negative Affect Scale? Past research suggests a two-factor structure (positive
affect/negative affect). Is the structure of the scale in this study, using a community
sample, consistent with this previous research?
What you need: A set of correlated continuous variables.
What it does: Factor analysis attempts to identify a small set of factors that represents
the underlying relationships among a group of related variables.

Figure 15.1 This scale consists of a number of words that describe different feelings and emotions. For
Positive and each item indicate to what extent you have felt this way during the past few weeks. Write
Negative Affect a number from 1 to 5 on the line next to each item.
Scale (PANAS)
very slightly or not a little moderately quite a extremely
at all bit
1 2 3 4 5
1. interested________ 8. distressed________ 15. excited________
2. upset________ 9. strong________ 16. guilty________
3. scared________ 10. hostile________ 17. enthusiastic_______
4. proud________ 11. irritable________ 18. alert________
5. ashamed________ 12. inspired________ 19. nervous________
6. determined_______ 13. attentive________ 20. jittery________
7. active________ 14. afraid________
 Factor analysis 187

Assumptions:
1. Sample size. Ideally, the overall sample size should be 150+ and there should be a
ratio of at least five cases for each of the variables (see discussion in Step 1 earlier
in this chapter).
2. Factorability of the correlation matrix. To be considered suitable for factor analysis,
the correlation matrix should show at least some correlations of r = .3 or greater.
Bartlett’s test of sphericity should be statistically significant at p < .05 and the
Kaiser-Meyer-Olkin value should be .6 or above. These values are presented as
part of the output from factor analysis.
3. Linearity. Because factor analysis is based on correlation, it is assumed that
the relationship between the variables is linear. It is certainly not practical to
check scatterplots of all variables with all other variables. Tabachnick and Fidell
(2007) suggest a ‘spot check’ of some combination of variables. Unless there is
clear evidence of a curvilinear relationship, you are probably safe to proceed
provided you have an adequate sample size and ratio of cases to variables (see
Assumption 1).
4. Outliers among cases. Factor analysis can be sensitive to outliers, so as part of your
initial data screening process (see Chapter 6) you should check for these and either
remove or recode to a less extreme value.

PROCEDURE FOR FACTOR ANALYSIS

Before you start the following procedure, choose Edit from the menu, select Options,
and make sure there is a tick in the box No scientific notation for small numbers in
tables.

Procedure (Part 1)
1. From the menu at the top of the screen, click on Analyze, then select
Dimension Reduction, and then Factor.
2. Select all the required variables (or items on the scale). In this case, I
would select the items that make up the PANAS Scale (pn1 to pn20).
Move them into the Variables box.
3. Click on the Descriptives button.
In the Statistics section, make sure that Initial Solution is ticked.
In the section marked Correlation Matrix, select the options Coefficients
and KMO and Bartlett’s test of sphericity. Click on Continue.
4. Click on the Extraction button.
In the Method section, make sure Principal components is shown, or
choose one of the other factor extraction techniques (e.g. Maximum
likelihood).
188 Statistical techniques to explore relationships among variables

In the Analyze section, make sure the Correlation matrix option is

selected.
In the Display section, select Screeplot and make sure the Unrotated
factor solution option is also selected.
In the Extract section, select Based on Eigenvalue or, if you want to force
a specific number of factors, click on Fixed number of factors and type in
the number. Click on Continue.
5. Click on the Rotation button. Choose Direct Oblimin and press Continue.
6. Click on the Options button.
In the Missing Values section, click on Exclude cases pairwise.
In the Coefficient Display Format section, click on Sorted by size and
Suppress small coefficients. Type the value of .3 in the box next to
Absolute value below:. This means that only loadings above .3 will be
displayed, making the output easier to interpret.
7. Click on Continue and then OK (or on Paste to save to Syntax Editor).

The syntax from this procedure is:

FACTOR
/VARIABLES pn1 pn2 pn3 pn4 pn5 pn6 pn7 pn8 pn9 pn10 pn11 pn12 pn13
pn14 pn15 pn16 pn17 pn18 pn19 pn20
/MISSING PAIRWISE
/ANALYSIS pn1 pn2 pn3 pn4 pn5 pn6 pn7 pn8 pn9 pn10 pn11 pn12 pn13
pn14 pn15 pn16 pn17 pn18 pn19 pn20
/PRINT INITIAL CORRELATION KMO EXTRACTION ROTATION
/FORMAT SORT BLANK(.3)
/PLOT EIGEN
/CRITERIA MINEIGEN(1) ITERATE(25)
/EXTRACTION PC
/CRITERIA ITERATE(25) DELTA(0)
/ROTATION OBLIMIN
/METHOD=CORRELATION .
 Factor analysis 189

Selected output generated from this procedure is shown below:

190 Statistical techniques to explore relationships among variables
Factor analysis 191
192 Statistical techniques to explore relationships among variables

INTERPRETATION OF OUTPUT
As with most SPSS procedures, there is a lot of output generated. In this section,
I will take you through the key pieces of information that you need.

Interpretation of output—Part 1
Step 1
To verify that your data set is suitable for factor analysis, check that the Kaiser-Meyer-
Olkin Measure of Sampling Adequacy (KMO) value is .6 or above and that the Bartlett’s
Test of Sphericity value is significant (i.e. the Sig. value should be .05 or smaller). In this
example the KMO value is .874 and Bartlett’s test is significant (p = .000), therefore
factor analysis is appropriate. In the Correlation Matrix table (not shown here for space
reasons), look for correlation coefficients of .3 and above (see Assumption 2). If you
don’t find many in your matrix, you should reconsider the use of factor analysis.

Step 2
To determine how many components (factors) to ‘extract’, we need to consider a few
pieces of information provided in the output. Using Kaiser’s criterion, we are interested
only in components that have an eigenvalue of 1 or more. To determine how many
components meet this criterion, we need to look in the Total Variance Explained table.
Scan down the values provided in the first set of columns, labelled Initial Eigenval-
ues. The eigenvalues for each component are listed. In this example, only the first four
components recorded eigenvalues above 1 (6.25, 3.396, 1.223, 1.158). These four compo-
nents explain a total of 60.13 per cent of the variance (see Cumulative % column).

Step 3
Often, using the Kaiser criterion, you will find that too many components are extracted,
so it is important to also look at the Screeplot. What you look for is a change (or
elbow) in the shape of the plot. Only components above this point are retained. In
this example, there is quite a clear break between the second and third components.
Components 1 and 2 explain or capture much more of the variance than the remain-
ing components. From this plot, I would recommend retaining (extracting) only two
components. There is also another little break after the fourth component. Depend-
ing on the research context, this might also be worth exploring. Remember, factor
analysis is used as a data exploration technique, so the interpretation and the use you
put it to is up to your judgment rather than any hard and fast statistical rules.

Step 4
The third way of determining the number of factors to retain is parallel analysis (see
discussion earlier in this chapter). For this procedure, you need to use the list of eigen-
 Factor analysis 193

values provided in the Total Variance Explained table and some additional information
that you must get from another little statistical program (developed by Marley Watkins,
2000) that is available from the website for this book. Follow the links to the Additional
Material site and download the zip file (parallel analysis.zip) onto your computer.
Unzip this onto your hard drive and click on the file MonteCarloPA.exe.
A program will start that is called Monte Carlo PCA for Parallel Analysis. You will
be asked for three pieces of information: the number of variables you are analysing
(in this case, 20); the number of participants in your sample (in this case, 435); and
the number of replications (specify 100). Click on Calculate. Behind the scenes, this
program will generate 100 sets of random data of the same size as your real data
file (20 variables × 435 cases). It will calculate the average eigenvalues for these 100
randomly generated samples and print these out for you. See Table 15.1.
Your job is to systematically compare the first eigenvalue you obtained in SPSS
with the corresponding first value from the random results generated by parallel
analysis. If your value is larger than the criterion value from parallel analysis, you

Table 15.1
Output from
parallel analysis
194 Statistical techniques to explore relationships among variables

retain this factor; if it is less, you reject it. The results for this example are summarised
in Table 15.2. The results of parallel analysis support our decision from the screeplot
to retain only two factors for further investigation.

Table 15.2
Comparison of
eigenvalues from
PCA and criterion
values from parallel
analysis

Step 5
Moving back to our SPSS output, the final table we need to look at is the Com-
ponent Matrix. This shows the unrotated loadings of each of the items on the four
components. SPSS uses the Kaiser criterion (retain all components with eigenvalues
above 1) as the default. You will see from this table that most of the items load quite
strongly (above .4) on the first two components. Very few items load on Components 3
and 4. This suggests that a two-factor solution is likely to be more appropriate.

Step 6
Before we make a final decision concerning the number of factors, we should have
a look at the rotated four-factor solution that is shown in the Pattern Matrix table.
This shows the items loadings on the four factors with ten items loading above .3 on
Component 1, five items loading on Component 2, four items on Component 3 and
only two items loading on Component 4. Ideally, we would like three or more items
loading on each component so this solution is not optimal, further supporting our
decision to retain only two factors.
Using the default options in SPSS, we obtained a four-factor solution. It is now
necessary to go back and ‘force’ a two-factor solution.

Procedure (Part 2)
1. Repeat all steps in Procedure (Part 1), but when you click on the Extraction
button click on Fixed number of factors. In the box next to Factors to
extract type in the number of factors you would like to extract (e.g. 2).
2. Click on Continue and then OK.

Some of the output generated is shown below.

Factor analysis 195
196 Statistical techniques to explore relationships among variables
 Factor analysis 197

Interpretation of output—Part 2: Oblimin rotation of

two-factor solution)
The first thing we need to check is the percentage of variance explained by this two-
factor solution shown in the Total Variance Explained table. For the two-factor
solution only 48.2 per cent of the variance is explained, compared with over 60 per
cent explained by the four-factor solution.
After rotating the two-factor solution, there are three new tables at the end of the
output you need to consider. First, have a look at the Component Correlation Matrix
(at the end of the output). This shows you the strength of the relationship between the
two factors (in this case the value is quite low, at –.277). This gives us information to
decide whether it was reasonable to assume that the two components were not related
(the assumption underlying the use of Varimax rotation) or whether it is necessary to
use, and report, the Oblimin rotation solution shown here.
198 Statistical techniques to explore relationships among variables

In this case the correlation between the two components is quite low, so we would
expect very similar solutions from the Varimax and Oblimin rotation. If, however,
your components are more strongly correlated (e.g. above .3), you may find discrep-
ancies between the results of the two approaches to rotation. If that is the case, you
need to report the Oblimin rotation.
Oblimin rotation provides two tables of loadings. The Pattern Matrix shows the
factor loadings of each of the variables. Look for the highest loading items on each
component to identify and label the component. In this example, the main loadings
on Component 1 are items 17, 12, 18 and 13. If you refer back to the actual items
themselves (presented earlier in this chapter), you will see that these are all positive
affect items (enthusiastic, inspired, alert, attentive). The main items on Component 2
(19, 14, 3, 8) are negative affect items (nervous, afraid, scared, distressed). In this case,
identification and labelling of the two components is easy. This is not always the case,
however.
The Structure Matrix table, which is unique to the Oblimin output, provides
information about the correlation between variables and factors. If you need to
present the Oblimin rotated solution in your output, you must present both of these
tables.
Earlier in the output a table labelled Communalities is presented. This gives
information about how much of the variance in each item is explained. Low values
(e.g. less than .3) could indicate that the item does not fit well with the other items in
its component. For example, item pn5 has the lowest communality value (.258) for
this two-factor solution, and it also shows the lowest loading (.49) on Component 2
(see Pattern Matrix). If you are interested in improving or refining a scale, you could
use this information to remove items from the scale. Removing items with low com-
munality values tends to increase the total variance explained. Communality values
can change dramatically depending on how many factors are retained, so it is often
better to interpret the communality values after you have chosen how many factors
you should retain using the screeplot and parallel analysis.
Warning: the output in this example is a very ‘clean’ result. Each of the variables
loaded strongly on only one component, and each component was represented by a
number of strongly loading variables (an example of ‘simple structure’). For a discus-
sion of this topic, see Tabachnick and Fidell (2007, p. 647). Unfortunately, with your
own data you will not always have such a straightforward result. Often you will find
that variables load moderately on a number of different components, and some
components will have only one or two variables loading on them. In cases such as
this, you may need to consider rotating a different number of components (e.g. one
more and one less) to see whether a more optimal solution can be found. If you find
that some variables just do not load on the components obtained, you may also need
to consider removing them and repeating the analysis. You should read as much as
 Factor analysis 199

you can on the topic to help you make these decisions. An easy-to-follow book to get
you started is Pett, Lackey & Sullivan (2003).

PRESENTING THE RESULTS FROM FACTOR ANALYSIS

The information you provide in your results section is dependent on your discipline
area, the type of report you are preparing and where it will be presented. If you are
publishing in the areas of psychology and education particularly there are quite strict
requirements for what needs to be included in a journal article that involves the use
of factor analysis. You should include details of the method of factor extraction used,
the criteria used to determine the number of factors (this should include parallel
analysis), the type of rotation technique used (e.g. Varimax, Oblimin), the total
variance explained, the initial eigenvalues, and the eigenvalues after rotation.
A table of loadings should be included showing all values (not just those above .3).
For the Varimax rotated solution, the table should be labelled ‘pattern/structure coef-
ficients’. If Oblimin rotation was used, then both the Pattern Matrix and the Structure
Matrix coefficients should be presented in full (these can be combined into one table
as shown below), along with information on the correlations among the factors.
The results of the output obtained in the example above could be presented as
follows:

The 20 items of the Positive and Negative Affect Scale (PANAS) were subjected to
principal components analysis (PCA) using SPSS version 18. Prior to performing PCA,
the suitability of data for factor analysis was assessed. Inspection of the correlation
matrix revealed the presence of many coefficients of .3 and above. The Kaiser-
Meyer-Olkin value was .87, exceeding the recommended value of .6 (Kaiser 1970,
1974) and Bartlett’s Test of Sphericity (Bartlett 1954) reached statistical significance,
supporting the factorability of the correlation matrix.
Principal components analysis revealed the presence of four components with
eigenvalues exceeding 1, explaining 31.2%, 17%, 6.1% and 5.8% of the variance
respectively. An inspection of the screeplot revealed a clear break after the
second component. Using Catell’s (1966) scree test, it was decided to retain two
components for further investigation. This was further supported by the results of
Parallel Analysis, which showed only two components with eigenvalues exceeding
the corresponding criterion values for a randomly generated data matrix of the
same size (20 variables × 435 respondents).
The two-component solution explained a total of 48.2% of the variance, with
Component 1 contributing 31.25% and Component 2 contributing 17.0%. To aid in
the interpretation of these two components, oblimin rotation was performed. The
200 Statistical techniques to explore relationships among variables

rotated solution revealed the presence of simple structure (Thurstone 1947), with
both components showing a number of strong loadings and all variables loading
substantially on only one component. The interpretation of the two components
was consistent with previous research on the PANAS Scale, with positive affect
items loading strongly on Component 1 and negative affect items loading strongly
on Component 2. There was a weak negative correlation between the two factors
(r = –.28) The results of this analysis support the use of the positive affect items
and the negative affect items as separate scales, as suggested by the scale authors
(Watson, Clark & Tellegen 1988).

You will need to include both the Pattern Matrix and Structure Matrix in your report,
with all loadings showing. To get the full display of loadings, you will need to rerun
the analysis that you chose as your final solution (in this case, a two-factor Oblimin
Table 1
Pattern and Structure Matrix for PCA with Oblimin Rotation of Two Factor Solution of PANAS Items

Item Pattern coefficients Structure coefficients Communalities

Component 1 Component 2 Component 1 Component 2
17. enthusiastic .825 –.012 .828 –.241 .686
12. inspired .781 .067 .763 –.149 .586
18. alert .742 –.047 .755 –.253 .572
13. attentive .728 –.020 .733 –.221 .538
15. excited .703 .119 .710 –.236 .462
1. interested .698 –.043 .683 –.278 .505
9. strong .656 –.097 .670 –.076 .475
6. determined .635 .107 .646 –.338 .377
7. active .599 –.172 .605 –.069 .445
4. proud .540 –.045 .553 –.195 .308
19. nervous .079 .806 –.144 .784 .620
14. afraid –.003 .739 –.253 .742 .548
3. scared –.010 .734 –.207 .740 .543
8. distressed –.052 .728 –.213 .737 .553
20. jittery .024 .718 –.242 .717 .507
2. upset –.047 .704 –.175 .712 .516
11. irritable –.057 .645 –.236 .661 .440
10. hostile .080 .613 –.176 .593 .355
16. guilty –.013 .589 –.090 .590 .352
5. ashamed –.055 .490 –.191 .505 .258
Note: major loadings for each item are bolded.
 Factor analysis 201

rotation), but this time you will need to turn off the option to display only coefficients
above .3 (see procedures section). Click on Options, and in the Coefficient Display
Format section remove the tick from the box: Suppress small coefficients.
If you are presenting the results of this analysis in your thesis (rather than a
journal article), you may also need to provide the screeplot and the table of unrotated
loadings (from the Component Matrix) in the appendix. This would allow the reader
of your thesis to see if they agree with your decision to retain only two components.
Presentation of results in a journal article tends to be much briefer, given space
limitations. If you would like to see a published article using factor analysis, go to
http://www.hqlo.com/content/3/1/82 and select the pdf option on the right-hand side
of the screen that appears.

ADDITIONAL EXERCISES
Business
Data file: staffsurvey4ED.sav. See Appendix for details of the data file.

1. Follow the instructions throughout the chapter to conduct a principal compo-

nents analysis with Oblimin rotation on the ten agreement items that make up
the Staff Satisfaction Survey (Q1a to Q10a). You will see that, although two factors
record eigenvalues over 1, the screeplot indicates that only one component should
be retained. Run Parallel Analysis using 523 as the number of cases and 10 as
the number of items. The results indicate only one component has an eigenvalue
that exceeds the equivalent value obtained from a random data set. This suggests
that the items of the Staff Satisfaction Scale are assessing only one underlying
dimension (factor).

Health
Data file: sleep4ED.sav. See Appendix for details of the data file.

1. Use the procedures shown in Chapter 15 to explore the structure underlying the
set of questions designed to assess the impact of sleep problems on various aspects
of people’s lives. These items are labelled impact1 to impact7. Run Parallel Analysis
(using 121 as the number of cases and 7 as the number of items) to check how
many factors should be retained.