9/11/2023

Factor Analysis Sectionl

Professor P. H. Lodhi
Former Head, Department of Psyhology
SP. Pune University, Pyne
Emal: phkoghiSmal.com
Preliminary Statistics
Mob 9422506443, 9049845968

Elementary Statistical Concepts Correlation: ConceptualComments

" Variance
" Correlation as the
" Standard deviation covariance of the
two standard scores
" Covariance
" Correlation as the product of the two
" Standard/Standardized scores standard scores
" Squared correlation as the
proportion of common variance

3
 9/11

Two Correlation Families Two Correlation Farnilies nt

2. Correlations Outside Product Moment Family

1. Product Moment Family Biserial correlation
" Usual product moment
correlatio i Polyserial correlation
" Phi coefficient Tetrachoric correlation

" Point biserial correlation . Polychoric correlation

Contingency coefficient
" Rank difference correlation
Cramer's V etc

Section I|
Factor Analysis
Two Types
" Exploratory Factor Analysis
Factor Analysis (Discussed Presently)
Anaiysis
" Confirmatory Factor
 9/11/2025

Correlations Among Six Psychological What should you Factor-analyze?

Variables

"Correlation matrix
fornboha00
Verbal abity.67 1.00 OR
Ego strength11 12031.00 " Covariance matrix?
Tension 17 141048 40 100

9 10

Major Steps in Exploratory Factor Factor Pattern Matrix

Analysis (Unrotated; PCA)

Factor 1 Factor 2 hso
Pre-Requisite: Data on Large Number of Variables
nfonaOT74
2 Choice of Model -Cormmon factor model OR Vetbal abilty 77
Component model
3Eraton ofUtotated Factots5 Ego strength51
4 Deciding the Number of Factors for Rotation Eponeess
Rotation:3
orlipgpna coblcue ortiogoga 0Blique
6 Think of Higher Order Factor Analysis ORSTOR

11 12

3
 Factor Analysis of Test Items (cont.) Confirmatory Factor Analysis
5. Choice of appropriate factor extraction When do you carry out factor analysis?
method. Principal Axes (PA)and Unweighted 1. When you have hypothesis stating
Least Square (ULS)methods are better suited " how many factors would be obtained and
6. Sometimes, especially with too many items, " Which variables would be loading on
the results are not clearly/meaningfully what factor
interpretable
7. Problems in factor analyzing ability items (Hypothesized binary target matrix)
8 Consider item
parceling. 2. Hypotheses of factorial invariance

21
22

Methods of Confirmatory Factor

Analysis -1 Confirmatory Factor Anaiysis (ML-CFA)
1. Methods based on structural equation Results
Variable Verbal
modeling or toss functions Arxiety
"Maximum likelihood CFA method Comprehension
Informationk
" Unweighted east square CFA method
General1zed least square CFA method
Verbal ability 86
.00
-Both Obique ldefauit) and Orthogonai Verbat analogies57
models. Rotation unnecessary Ego strength
-Only common factor model Guilt pro 70

fension
eness 00

23 70
 Factor Analysis
Factor analys1s iS used to uncover the latent structure of a set of
space trom a large no of variables to a smaller no, of variables It reduces attribute
procedure factors and as such is a non dependent

Factor analysis could be used for any of the

following purpose
1. To reduce a large no. of variable to a smaller no. of factors for
the large number of variables precludes modeling all the modeling purposes, where
measures, individually. As such ractor
analys1s IS integrated in structural equation modeling, helping create the latent variables modeied
by SEM (structurc cquation model).
2. To select a subset of variables from a large set based on which original variable have the
highest correlations with the principal component factors.
3. To create a set of factors be treated as uncorrelated variable as one approach to
handling multicollinearity regression.
Assumptions:
Factor analysis is a part of the Multiple General Linear Hypothesis (MGLH), family of
procedures and makes many of the same assumptions as multiple regressions. Linear relationship
interval or near-interval data, untruncated data, proper specification (relevant variable included
extraneous are excluded), lack of high multicollinearity and multivariable normality for purpose
of significant testing.
Factor analysis generates a table on which the rows and the observed row indicator variables end
the columns are the factor or latent variables which explain as much of the variable in those
variables as possible. The cells in this table are factor loadings and the meaning of the factors
must be induced from seeing which variables are most heavily loaded on which factors this
inferential process can be fraught with difficulty as diverse researchers impute different tables.
Methods
There are scveral different types of factor analysis
I. Principal component mcthod
2. Principal axes method
3. Summation method
4. Centroid method

AsSumptions of factor analvsis model

I) Measurement error has constant variance and is on average zero, i.e.

varle )=a
E[e] =0
2) No association between the factor and measurement error, cov(F, e. ) =0
3) No association betwecn errors, cov(e,.e, ) = 0
 41 Local i.e.. conditional independence): Given factor, observed variables are indenendans..
one another. covíX,|F)=0
Steps in Exploratory Factor Analvsis:
Collect data choose relevant variables
1.
2 Extract initial factors (via
3. Choose number of factors principal
to retain
component).
4. Choose estinmation method,
5. Rotatc and
estimate model.
interpret.
6.(a) Dec1de on changcs need to be made ( e.g.
(b) Repcat (4). (5). drop items include items)
7. Construct scales and use on
further analysis.
Principal component method:
In this method. tfactors are selected one at a
traction is created such that it represents the time
most
such that each factor best fits the data. The
tirst
Each highly correlated set of variables.
subsequent selected factor
This procedure is continued till explains less variance than its
all the factors are predecessor.
largest amount of residual selected.
variance in the entire set of
Ali the factors selected explain the
standardized response scores.
Centroid Method:
1) Obtain the
2)
correlation matrix.
Obtain grand matrix sum. row sum, column sum.
3) Calculate N=
Vgrandtotal
4) Multiply each column sum with N, which gives the first
5) To find the Second factor loading. find the cross factor loading.
factor loading horizontally and vertically and thenproduct matrix of factor I by testing tirst
columns multiplying corresponding rows arnd
6) Find the first factor residual matriX given as.
Residuals = total ariations- explained variations
7) Retlection.
Retlection means that each test vector
The major purpose of reflection is to getretains
its
length but extends in opposite directions.
a reflected cost matrix having the
grand tota! This step is taken due to the highest
that some of the tactors load1ng is Withpossible
total. The method of reflection is by trial reason
and error. This can be done by changing the signs grand
the varables trom positive to negative and negative to of
The outcome we get
positive column wise and row wise
is a retlective residual correlation matrix
$) Repeat stens trom (3) to (7).
 Methods for finding number of factors to be extracted
I) Thunb Rule: All the interrelated factors nust explain at least as rrnuch as variances as ani
average variable. Check, if avariable is under a factor then the percentage of variable erplaininz
variance should be less than the percentage of factor cxplain1ng
2) Eigen Value Index: When the cigen value of a factor is less than 1, it cxplains less variare
then the variables included in the factor itself such a factor should not be considered.

3) Fruckter Formula:
(2n -1)-V&n +1
Number of factors:
2
Where n is the number of variables included in the study.
correlation is observed end
4) Residual correlation matrix method: In this method, the residual
are zero, and then the eztraction
if it is soon that most of the correlation coefficient in this matrix
of the factor can be determinate.
to be retained from the
5) Scree Plot test: This method is to decide about the number of factors contributing
extracted factors. The test determines which of the extracted factors are actually against the
variance and does not measure random errors. The number of factors is plotted
factors.
proportion of variance. It extracts in the order of the extracted
Standardization of Responses:

where Y, is a value corresponding to aresponse and is the variance.

Factor Loading:
It is the correlation between a factor and a variable. It helps interpret the meaning of a factor by
indicating how well the factor fits the standardized responses to a variable. The greater the
value of factor loading the better is the fit of the factor to the data from the concerned
Statement. All variables load on all factors but they load highly on some specific factors. Range
is ±I
There may be some variables which may by loading highly to more than rwo factors, decide in
which factors which variables are to be kept.

Eigen Values: It is the measurement of the amount of variants explained by a factor. A factor
eigen value is the sum of the square of its factor loading.
Communality: It indicates the proportion of variance in the responses to the statement which is
explained by the identified factors.
 Percentage of Variance:
eiven value of the tactor.. t00
sum of all eigen values

If any variable is not combined in any of the groups, then it can be

another factor. To remove variable which is totally left or can be
changing its direction.
d1fferent use rotation i.e.
considered
we are basicallyas
Rotation of Factors: For the purpose of simpl1tying the interpretation of
increase the number of high and low positive loadings in the columns of aobtained factore
is used. There are two metihods for this: factor. factor t
I) Orthogonal Rotation/ Variance
Rotation: Here factors are rotated such that the original faetor
as well as rotated factors are
orthogonal. The line between the tfactors axis remains 90*
2) Promax Rotation: The factors are
rotated such that the line between original and
tactors IS more than or less than 90* rotated

Advantages of Factor Analysis:

I. Both objective and subjective attributes can be
2. It can be used to used.
identify the hidden dimensions or
apparent trom direct constraints which may or may not be
3. It is not extreinely analysis.
difficult to do and at the same time its
results. inexpensive and gives accurate
4. There is
flexibility naming and using dimensions.
in
Disadvantages of Factor Analvsis:
1. The
usefulness depends on the researcher's ability to
product attributes.
attributes are misseddevelop
If important a complete and
accordingly. accurate set of
the value of procedure
2, Naming of thc is reduced
factors can be dit+icult multiple
apparent reasons.
3. If the
attributes can be highly correlated with no
observed variatbles are completely
meaningtul unrelated the factor analysis is unable to
4 I: is not no to know produce
researcher's on this factors actually represents, only theory can help inform the
 17
Joumal of Psychological Researches, 2004, Vol 48, No 1, P17- P24
EVALUATING THE FIVE-FACTOR MODEL OF PERSONALITY:
CONFIRMATORY FACTOR ANALYSIS OF THE REVISED NEO
PERSONALITY INVENTORY
PHLodhi Savita Deo and Vivei M Belhekar

inter-factor correiations
ABSTRACT for example. about the
invariance across different
the factorial shall use the
The present paper briefly reviews the populations, etc In this paper. we
difon confirmatory factor anaytu methods and term confirmatory factor analysis in a broad
the factor analytic evaluation of the Revsed NEO.
Sense.
Personaliry Iventory (NEO-PI-R) based on the be
Five-Factor Model (FFM). The utility of Broadly speaking. CFA methods can
orthogonal and oblique procrustes rotation of classified into three types.
var imaxed principal-components to hypothesized on loss functions:
1. Iterative methods based
binary target matriz, and the multiple group loss functiorns,
Mulaik (I988) listed three importantlikelibood,
factor aralysis (MGFA) has also been tested on and
namely, least squares, rmaximum
the empirical data of 214 subjects with equal in LISREL
generalized least squares. Accordingly,
have three CFA
gender representation (Lodhi, Deo, & Belhekar, (Joreskog & Sorbom, 1986), we
(MLCFA),
2002). These methods clearly recovered the five methods: (a) Maximun likelihood CFA and
factor structure from the NEO-PI-R data, thus CFA (ULS-CFA),
(b) Unweighted least square (GLS-CFA). As
indicating the robustness of the FFM. (c) Generalized least square CFA
is also
Bentler (1980) pointed out, LISREL notation
model because of
NTRODUCTION sometimes referred to as JKW
Factor analytic methods can be broadly of Joreskog (1973), Keesling
the pioneering role(1973) in developing a relatively
classified into two major heads exploratory (1972), ,and Wiley
methods and confirmatory methods. As
Mulaik standard notational framework The CFA, carried
as
(1988, p. 259) pointed out "Exploratory and out by these methods, is sometimes called 1996),
structural factor analysis (McArdle,
confimatory factor analysis reflect respectively
measurement model (Joreskog & Sorbom, 1986:
two difTerent approaches to the philosophy of
science."The term, confirmatory factor analysis Hoyle, 2000), etc. These methods have strong
(CFA), is used in diverse sense in
current base in covariance structure analysis (Bentler &
psychological and methodological literature. Bonnet, 1980) and structural equations modeling
According to one school of thought, confimatory (SEM) (Maruyama, 1988). CuTently large numbers
These
iterative
factor analysis methods refer to the functions. of software are available for SEM and CFA.
confirmatory methods based on loss sofware include AMOS, CALIS, COSAN, EQS.
developed by LISCOMP, LISREL, Mplus, Mx, ROMONA,
These mcthods have primarily been RAMpath, and SEPATH (Hoyle, 2000, p.477). It
Joreskog(1967, 1969, 1973,etc.) and have been
incorporated in now a classic and well-known may be pointed out that arnong the three CFA
another methods based on loss functioas, maximum
software known as LISREL According to
school of thought, confirmatory factor analysis likelihood CFA method is the most popular one.
refers to any method of factor analysis employed 2. Multiple group faction analysis (MGFA):
purpose of hypothesis testing (e.g-,
Tor t e o The researcher bas to hypothestze A briefhistory of dhe development of MGFA has
Nunnally, I701 which been presented by Harman (1970). Seveeral factor
as to how many factors he wotld get and analysts like Guttman (1952), Harman (1970),
variable would load on what factor. A more detailed Gorsuch (1974) etc. have emphasized the
bypothesis specification can of course be done,
"Department of Prychology University of Pune. Pune: 41| 007 098
" Department of pplied Pychology, Universiry of Mumbai, Mumbai: 400
 PH Lodhi, Savita Deo and
Vivek M
hypothes1s testing role of the MGFA Nunnally Till now, the NE0-P|-R
has been
Belhekgr
(1981, p 399) even claimed that he was "sold 29 used
countries, In languages and 36 cuitures in 33
on the general usefulness of multiple grouP 2002) few of the (McCe
methods for performing confirmatory factor employing NEO-PI-R the cross-cultural
varimax
studies
an aly s1s Mulaik (1988,p.273) also discussed
MGFA in his chapater on confirmatory factor components showed minor departures rotated from
prncipthaal
analysis under the caption 'Revival of an old
FFM Such findngs are not
Method' and pointed out that there is a "renewed
completety
unexpected in theory testing cross-cultural studue
interest in the multiple group method" examining factorial invariance In this contex.
McCrae, Zonderman, Costa, Bond, and
3J.Procrustes rotations : Procrustes rotations (1996, p 558) remarked Paunonen
can be employed for two purposes ure
not neeessarY optimal
that"exploratory analyses
obtatned tactor pattern can be rotated First. the
to some hypothesized
tor testing
In exploratory factor
other factor pattern, analy ses. such as PCA with
earlier study or on somesometimes obtained in an
other group, for maximum theory, maximally, can Suggest varimax rotation, the
the nunber of
congruence This practice is often followed in factors to be rotated, neither
the studies on
factorial invariance (Mulaik. is guided by Cxtraction nor
theory. It can be a crucialrotation
Second, the obiained factor pattern 1972) can be especially when the hypothesized structure issue
rotated to hypothesized (binary) target simple In this i5 not
practice is particularly suitable for matrix. This context McCrac et al (1996, p
$38) have rightly
evaluation may be
It modei or theory
pointed out that
testing thhough procrustesnoted that thein hypothsis differences in the observed "smali
of factors carried out by
is rotation, extraction vield large differences in the correlations can then
exploratory and the soiutions may position of the axes.
by theorymethods,
only the rotation is guided appear to be
procrustes rotations can be either The different." McCrae. Costa, del Pilar. dramatically
oblique orthogonal or Parker (1998, p. 179) have Rolland, and
rotation is designed to opined that "varimax
Factor Analytic and is not optimize simple structure
Model Evaluation of the Five-Factor of variablesnecessarily
that show appropriate for the analysis
Currently. the
the facets of E and A. circumplex ordering. as do
Five-Factor-Modei
personality is probably the most (FFM) of matrices for the two groupsWhen are
the
correlatior
model in personality psy chology The popuiar trait analyzed and rotated. separatel: facto
this modei on impact of
personal1ty research in difterent
vary ing
Structure maysometimes the
merely be dueapparent
rotational shift in the data due to
countries. cultures and to "a
volumelanguages
error"
documented Chae, sample-spec1ti
is well
overcome(Piedmont
in a recent &
and Allik (2002). In the
eidted by
context FFM, Costa McCrae such problems., 1997, etp. 140). To
and
well McCrae (1992) have
of
Suggested MeCrae a! (19961
orthogonal procrustes
known inventory, presented.
the rreyisednoy a very
sed NEO Personality
of the
altermatives. As such, rotation as one
Inventory (NEO.P1-R). The and McCrae
Piedmont Chae (1997), MeCrae et al (996). et al
five major NEO-PI-R measures
dimensions of personality: Neuroticism and
N),
Extraversion (E), Openness to (2002),Angleitner and
Lima (2002), Ostendorf
(1998).
(2000), Leininger
Agreeableness (A). ard Experience (O), Costa (2002) Piedmont, Bain, McCrae, and
Each of these
d1mension Conscientiousness
(domain)
C
rotatior to employed orthogonal
the context American factor procrustes
has been
conceptualized in terns of six
nav ing thirty facet
turther
facet scales. of the normative
stud', of factorial partern. I
scales in the NEO-P[-R. thus procrustes
Costa and by Mulaik rotation
(1972). has also been 1nvarance.
factor patern of the McCrae|992) have presented the recommended
componerts NEO-PI-R using the principa Church
of CFA appr
and Burke
(19941
Toach using ioss examined
araly
sis
(PCAifollowed by the utii t
rotatiorn The results verv weli varimax PI-R data and functions tc the EC
support the FFM
highliznted
technicue for testing the
personal1tylmitations c! the
structure moce s
Five-Factor Model
in view of the limited sinple struture of ferality
measurcs. McCrae et al (199%, p 153), wh
cxarnined MICEA in the contet of NEO P data,
rernarked, "In typical applications, therefore,
variables are hypothesized to load n a single
specified factor, and their loadings on ther factts
are fixed at zero. Mazitnum likelihnd estirnates
are then used to deterrnine the optirmal values for
the hypothesized factor loadings, and the fit of
the solution (...) is then evaluated ln real datz results of la) bth orthozonel 0bliue
procrustet rotations to 2 bypothesized iary
the secondary loadings of variables are rarely
exactly zero, but in simple structure rnodels they art ratr, and (b) ooliQU 2s l l 2s orogoal
are assurmed to fluctuate randomly about Zero. If. MGFA Even tht visual eeminaticn of these
however, small loadings are in factt meaninzful, resultsWould demonstrate the utility of these
CFA with a simple structure model may not fit methods in recovering the five-factor stucture
well" McCrae et al further rernarked, The factors from the NEO-PI-R. Lncidetzlly t may be ponted
of the FFM were initially identified with rotation out that in rone of these etods the ton-t22eed
procedures (like varimax) designed to approximate loadings in the actual rotated factor pattern
matrices are fised to zErO
simple structure. Yet the FFM does not postulate
perfect simple structure that is, it does not METHOD AND ANALYSES
assume that all personality traits define one and
only one factor." Guided by these argurments, Data
McCrae et al empirically explored the utility of ML
CFA for the evaluation of FFM using NEO-PI-R. The present analyses are based on the data
data. The traditional style application of ML-CFA coliected on 2 sample of 214 post-graduate
to the NEO-PI-R data did not indicate a very g0d students (Mean age = 21.71 yTS, SD = I 29). Of
fit to the hypothesized model, though when the these subjects, 107 were males and the remzining
non-targeted loadings were fixed to small values 107 were females. This sample was drawn from
instead of zero, the fit of the model considerably ten University departments end, 2s such, in can
improved. McCrae et al concluded that ML-CFA be considered to be fairly heterogeneous. These
has serious problems when used to examine subjects received a Marathi adaptarioz of the
personality structure. In the same paper McCrae NEO-PI-R (Lodhi et al, 2002). As pointed out
et al also studied the utility of procrustes rotation above, the NEO-PI-R measures five major
to normative American factor
Paterm matrix in dimensions of personality: Neuroticism (N),
Vaiuating the FFM and
eval the positive Extraversion (E), Openness to Experience (0),
findings. Agreeableness (A), and Conscientiousness (C).
Each of these dimension (omain) has been firther
OBJECTTVES conceptualized in tems of six facet scaies, thus
The above review indicates that the ML-CFA, having thirty facet scales in the NEO-PI-R. The
the most widely used iterative method based on statistica! analyses of these data including (a)
loss function, has reasonably severe limitations descriptive statistics and internal consistency
in evaluating the FFM. On the contrary, orthogonal relizbilities for 30 facets and ive domains, (b)
procrustes rotation to normative American factor zender differençeson 30 facets and ive domains,
pattern has consistently yielded quite satisfactory (c) domain corfelarons, (d) Corelations of the
results. However, as we have pointed out in our domain scales with the NEO Eive-Factor Inventory
carlier work (Lodhi, Deo, & Belhekar, 2002, p.238), (NEO-FFI) have been presented in Lochi et al
tations to
"Orthogonal and oblique Procrustes rotations (2002). Lodhi et al also presented two factor
a hypothesized binary target matrix are partem matrices - frs, the variumax-rotated principal
underutilized for evaluating the FFM." We have components, and second, the orthogonal
 PH Lo ihi Sita Deo an ve Benes
20

to
Rotation

Rolaton
rustes
Procrustes
Matrix
0rthogonal
Target
Binary 8888558s28ARI8R 823
A:
- Hypothesizd
Solutions

Procrustes
Rotation
to
rrocrustes
lable
I Otaton
Two
lor
Scales

Facel
iNEOrR 8838539388S8233

the
than
onder.Mipdetness greater
Disciplne:
Selt
C5
Gompelernce
Compliance DutitulneSs
loadinas
Hwotheied Straighttc
A2
Feeling:

Met
Five-Factor Model
21
-Solutions

(MGFA)Solulton
al

Analysis
Factor
(orthogonal:
Group
Multiple
MGFA
Two
2 forB.
Table
and
SealesSolution)
Se
Facet (Oblique
(oblique
NEO-PI-R
MGFA
MGEA
the A:
forA.
Matrices
Pattern
Factor
 Belhekar
Deo and Vivek M
PH Lodh Savita
22
highest loading on factor
expected to have their
procnustes rotated factors to nornmative
American by all extraversion facet
This requirement is met
factor pattem. Lodh1 et al also presented
another as such, factor
scales except E4 (Activity) and,(E) The next SIY
study examining the validity of the Marathi 2can be iabeled as
Extraversion
adaptation of the NEO-P[-R in terms of its belonging to the Openness
facet scales (0l to 06),
correlates with the Eysenckian scales their highest loading
to Experience domain, have can
Analyses land Il : Orthogonal and Oblique As such factor 3
on factor 3 as cxpected
Experience (0)
Procrustes Rotations to the Hypothesized Binary clearly be labeled as Openness to belonging
Target Matrix The next six facet scales (Al to A6), to have
The 30 x 30 correlation matrix, representing to Agreeableness domain, were expected
has turrned
the corelations among the facet scales of the NEO their highest loading on factor 4 This
the facet
PI-R was subjected to principal components out to be the case though the loading for
cutting
analysis The first five principal components, scale Aa (Modesty) falls little short of our
indicated tby Cattells scree and also suggested point of 040. Thus factor 4 is an Agreeableness
by the five-factor theory were subjected to varimax factor. The last six facet scales (CI to C6),
rotation. For carrying out procrustes rotation (both belonging to Conscientiousness domain. loaded
orthogonal and oblique). the five-factor on factor $ as expected As such factor 5 can
hypothesis :n the context of the NEO-PI-R was clearly be labeled as Conscientiousness factor
represented in terms of 30 x 5 binary target matrix, The five factors together have condensed 532
wherein the targeted loadings were indicated by per cent of the total variance
unities and the non-targeted loadings were The Cormpari1son of table I-A and table [-B
indicated by zeros. Orthogonal procrustes rotation indicates that both orthogonal and oblique
was accomplished by following Schonemann,
(1966) and McCrae. et al (1996). The obtained procrustes transformations to hypothesized binary
results are presented in table [-A. The obligue target matrix have yielded highly similar results.
procrustes rotation was accomplished by following Even in table 1-B, only E4 (Activity) is having its
Mulaik (1972). The obtained results are presented highest loading on non-intended factor. As such
in table -B the results in table 1-B can be interpreted on the
above lines and it can be concluded that the results
Aaalysis M and IV: Oblique and Orthogonal very well support the FFM. Although the inter
Muitiple Group Factor Analysis factor correlations of the oblique solutior. are
Retaining unities in the diagonal spaces of reported here for space reasons, their absolute
values range from 0.04 to 0.22. Thus even the
the 30x30 correlation matrix, we obtained oblique oblique rotation has yielded practically orthogona!
muitiple group factor solution by following
Gorsuch(1974) and Harman (1970). The obtained factors. Incidentally. it may be stated that Horn
resul:s are presented in table 2-4. The oblique (1967) has strongly criticized oblique procrustes
MGFA soiution was transformed into an rotation because it can considerably capitalize on
orthogonal solution by following Harman (1970). chance by
by introducing highh obliqueness (i.e.,
high
inter-factor correlations) in the process of rotation.
The obtained results are presented in tabie 2-B. Since the inter-factor corelations obtained in the
RESULTS AND DISCUSSION present analysis are quite low, this criticism is not
applicable in the present context. Since even the
4 reference to table |A ndicates hat the
five-factor model is clearly supported All the first
application of the oblique procrustes rotation has
Six facet scales (NI to N6). belonging to
yielded quite low-correlated factors, it can be
concluded that the five factors can be regarded
Neuroticism domain, were expected to have their as for practical purposes,
highest loadings on factor . This has
ha turned out orthogonal
to be the çase Thus factorl can clearlv be labeled orthogonality of ive factors is not a rotational
artifact as in the case of varimax or
as Neuroticisn N), The next six facet scales El
procrustes
orthogonai
to E6). belonging to Extraversion doma1n. were
rotation We suggest researchers
using the NEO-P|-R and doing princ1pal
 Five-Factor Model
component
their s-var
varimaxed im ax factor analysus,
principal
procrustes totation components
to
to submit
oblique
Bentler, P. M &Boaett, D. G U940,
23

obliqueness in the and examine Significance tests and goodness of fit in the
a very
welcome resulting carefully
solution the
It will be
analysis of covariance structures
in future studics.findingif our results are Psychological Bulletin 88, 588-606.

The
repiicated Church,T. A. &Burke, P. J. (1994). ExporseA
presentcd results,
in based on
table 2-A, very oblique
FFM. We, however, MGFA and
and confirmatory test of the big five and
Tellegen's three- nd four-dimensjonal
do not clearly support the models Journal of Porsonality and Social
MGFA very consider oblique
of the model suitable to evaluate the Psychology 66, 93-1|4
the
since when obliqueness
diagonal spaces of the unities are retained in Costa, P. T. Jr., & McCrae, R. R. (1992). Revised
expectedcorrelation
every facet scale is NEO Personality Inventory (NEO-PLR) and
in the to load on mnatrix and
one factor NEO Five-Factor Inyentory (NEO-FFl)
be shown hypothesized binary target matrix, it can
that the inter-factor
professional · manual. Odessa, FL:
Psychological Assessment Resources.
the inter-domain correlations and
The examination correlations
of the
would be equivalent. Gorsuch, R. L. (1974). Factor Analysis.
MGFA, presented in tableresults
of orthogonal Philadelphia: W. B. Saunders Company.
all the facet scales are 2-B, also indicates that
on the intended factors, having their highest loading Guttman, L. (1952). Multiple group methods for
thus supporting the FFM. common factor analysis: Their basis,
However, unlike oblique MGFA solution, in the computation, and interpretation.
orthogonal MGFA, the three facet scales (03: Psychometrika. 17, 209-222.
Feelings; C1: Competence; and CS: Self-discipline) Harman, H. H. (1970) Modem factor anaysis (2nd
do not have _imple structure.
ed., 3d impression). Chicago: The University
Thus it is clear that all the four confimatory of Chicago Press.
methods employed in the present work have
recovered the Five-Factor structure from the NEO Horn, J. L. (1967). On subjectivity in factor
PI-R data. The confirmation of the FFM by analysis. Educational and Psychological
diverse methods indicates its robusthess. Measurement. 21, 811-820.
Acknowledgement
Hoyle, R H. (2000). Confrmatory fåctor analysis.
In H. E. A. Tínsley & S. D.
We gratçfully acknowledge Professor R. R. Brown (Eds.),
McCrae, National Institute of Aging, Gerantology
Handbook of applied multjvariate statistics,
and mathem atical modeling (pp. 465-497).
Research Center, Baltimore, USA, for valuable San Digo: Academic Press.
suggestions during the course of our work on the
Five-Factor Model. Joreskog, K. G.( (1967). ^ome
contriþutions to
maximum likelihood factor analysis.
REFERENCES Psychometrika 32, 443-482.
Angleitner, A., & Ostendorf, F. (2000, July). The Joreskog, K. G. (1969), A general approach to
FFM: A comparison of German speaking confirmatory maximum likelihood factor
countries (Austria, Former East and West analysis. Psychometrika. 34, 183-202.
Germany, and Switzerland). Paper' prèsented
at the XXVIIth International Congress of Joreskog, K. G. (1973). Agenéral method for
Psychology, Stockhotm, Sweden. estimating a linear structural equation system.
In A. S. Goldberger &0.D.
Bentler, P. M.(1980). Multivariat analysis with Duncan (Eds.),
Structural equation models the social
latent variables: Causal modeling. Annual sciences (pp. 85-112). New York: 'Acadermi
Review of Psycholog 31, 419-456. Press: