See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/230600798

The measurement of trait emotional intelligence with TEIQue-SF: An analysis

based on unfolding item response theory models

Chapter in Research on Emotion in Organizations · January 2011

DOI: 10.1108/S1746-9791(2011)0000007016

CITATIONS READS

23 26,561

1 author:

Leonidas A Zampetakis
University of Crete
68 PUBLICATIONS 3,131 CITATIONS

SEE PROFILE

All content following this page was uploaded by Leonidas A Zampetakis on 07 October 2017.

The user has requested enhancement of the downloaded file.

CHAPTER 11

THE MEASUREMENT OF TRAIT

EMOTIONAL INTELLIGENCE WITH
TEIQUE-SF: AN ANALYSIS BASED
ON UNFOLDING ITEM RESPONSE
THEORY MODELS

Leonidas A. Zampetakis

ABSTRACT
The present chapter addresses a topic that is of growing interest – namely,
the exploration of alternative item response theory (IRT) models for
noncognitive assessment. Previous research in the assessment of trait
emotional intelligence (or ‘‘trait emotional self-efficacy’’) has been
limited to traditional psychometric techniques (e.g., classical test theory)
under the notion of a dominance response processes describing the
relationship between individuals’ latent characteristics and individuals’
response selection. The present study, presents the first unfolding IRT
modeling effort in the general field of emotional intelligence (EI). We
applied the Generalized Graded Unfolding Model (GGUM) in order to
evaluate the response process and the item properties on the short form
of the trait emotional intelligence questionnaire (TEIQue-SF). A sample
of 866 participants completed the English version of the TEIQue-SF.
Results suggests that the GGUM has an adequate fit to the data.

What Have We Learned? Ten Years On

Research on Emotion in Organizations, Volume 7, 289–315
Copyright r 2011 by Emerald Group Publishing Limited
All rights of reproduction in any form reserved
ISSN: 1746-9791/doi:10.1108/S1746-9791(2011)0000007016
289
290 LEONIDAS A. ZAMPETAKIS

Furthermore, inspection of the test information and standard error

functions revealed that the TEIQue-SF is accurate for low and middle
scores on the construct; however several items had low discrimination
parameters. Implications for the benefits of unfolding models in the
assessment of trait EI are discussed.

INTRODUCTION

Trait emotional intelligence (trait EI or trait emotional self-efficacy) is defined

as a constellation of emotion-related self-perceptions located at the lower
levels of personality hierarchies (Petrides, Pita, & Kokkinaki, 2007).
Consequently, it is not distinct from personality constructs, but part of
them. The conceptualization of EI as a personality trait is in accordance with
the subjective nature of emotional experience and leads to a construct that
lies wholly outside the taxonomy of human cognitive ability (Carroll, 1993).
Trait EI is operationalized through self-report questionnaires. In the present
study we focus on trait EI assessment in general and the Trait Emotional
Intelligence Questionnaire – Short Form (TEIQue-SF, version 1.50; Petrides
et al., 2010) in particular using modern psychometric methods.
TEIQue-SF is a 30 item scale designed to measure global trait EI and it is
based on the long form of the TEIQue (Petrides, 2009 – technical manual).
The scale seems to be a promising research tool for trait EI assessment, due to
its brevity, evidence of its predictive validity, and good basic psychometric
properties spanning student and nonstudent samples in numerous countries
(Austin, 2009; Freudenthaler, Neubauer, Gabler, Scherl, & Rindermann,
2008; Mikolajczak, Menil, & Luminet, 2007; Petrides & Furnham, 2006;
Sevdalis, Petrides, & Harvey, 2007; Smith, Heaven, & Ciarrochi, 2008;
Zampetakis, Kafetsios, Bouranta, Dewett, & Moustakis, 2009). However,
although researchers have devoted substantial efforts to demonstrating the
psychometric quality of the TEIQue-SF, the techniques used in these efforts
have not kept pace with the advances in psychometric theory and methods.
This is regrettable given that new modern test theory methods – notably Item
Response Theory (IRT) and its variants, such as unfolding (ideal point)
models – may provide additional insights over and above traditional
psychometric techniques (Classical Test Theory, CTT) (e.g., Cronbach,
Gieser, Nanda, & Rajaratnam, 1972).
Despite the lengthy history of the conceptual underpinnings (e.g.,
Coombs, 1964; Davison, 1977), unfolding models are relatively new.
Measurement of Trait Emotional Intelligence with TEIQue-SF 291

Recently, after a surge of publications aimed at developing the

models (i.e., The Hyperbolic Cosine Model – Andrich, & Luo, 1993; the
Graded Unfolding Model – Roberts, & Laughlin, 1996; the Generalized Graded
Unfolding Model – Roberts, Donoghue, & Laughlin, 2000; and the General
Hyperbolic Cosine Model – Andrich, 1996) and estimation procedures (e.g.,
Johnson & Junker, 2003; Luo, 2000; Luo, Andrich, & Styles, 1998; Roberts,
Donoghue, & Laughlin, 2002) applications of these models have begun to
appear in the literature (Chernyshenko, Stark, Drasgow, & Roberts, 2007;
Drasgow, Chernyshenko, & Stark, 2009; Scherbaum, Finlinson, Barden, &
Tamanini, 2006; Stark, Chernyshenko, Drasgow, & Williams, 2006;
Weekers, & Meijer, 2008; Zampetakis, 2010). However, more research is
needed to enhance confidence in the usefulness of unfolding models in
describing the response process on self-report inventories in general and EI
measures in particular.
The present study heeds this call by utilizing the Generalized Graded
Unfolding Model (GGUM) (Roberts et al., 2000; Roberts, Fang, Cui, &
Wang, 2006) with a sample of 866 British participants, to test TEIQue-SF
under an unfolding perspective. The purpose of the study is to extended
recent research studies (Chernyshenko et al., 2007; Scherbaum et al., 2006;
Weekers, & Meijer, 2008; Zampetakis, 2010) by demonstrating the potential
usefulness of unfolding models in EI assessment.

Classical Dominance Analyses and the Development of the TEIQue-SF

CTT was the dominant statistical approach to testing data until Lord and
Novick (1968) placed it in context with several other statistical theories of
test scores, notably item response theory (IRT). However, both CTT and
IRT are based on the notion of a dominance response processes
(monotonic) describing the relationship between individuals’ latent char-
acteristics and individuals’ response selection; i.e. the probability of item
endorsement increases as the trait level increases (Drasgow et al., 2009).
Techniques such as exploratory factor analysis (EFA), confirmatory
factor analysis (CFA), and IRT analyses with the one, two, or three-
parameter logistic model are all based on dominance assumptions. For
example, an item–total correlation or a factor loading implicitly assumes
that the relationship between trait level (e.g., total score) and item score can
be modeled by a monotonically increasing regression line (see Chernyshenko
et al., 2007).
292 LEONIDAS A. ZAMPETAKIS

The TEIQue-SF was designed according to procedures used in classical

psychometric scale development (Petrides, Pérez-González, & Furnham,
2007), in order to measure global trait emotional intelligence (trait EI).
The scale is based on the long form of the TEIQue (Petrides, 2009).
Specifically Petrides and Furnham (2001) after content analyzing salient
models of EI and cognate constructs in the literature, derived the sampling
domain of trait EI which consists of 15 facets, organized under four-
factors: well-being, self-control, emotionality, and sociability. The ratio-
nale for the scale construction was to include core elements common to
more than a single model, but exclude peripheral elements appearing in
only one specific conceptualization; thus the commonalities (shared core)
of the various items comprising the scale were carried over into a total
(internally consistent) score, with their random or unique components
(noise) being canceled out in the process. In turn, two items from each of
the 15 subscales of the TEIQue were selected for inclusion in the TEIQue-
SF, based primarily on their correlations with the corresponding total
subscale scores.
This procedure of scale development implies that a large number of items
was discarded during scale construction; requiring all items to have high
factor loadings and item–total correlations during scale construction
effectively eliminated items from the final scale. Examining the items in
TEIQue-SF, it becomes immediately apparent that they are either positive
(15 items; e.g., ‘‘Expressing my emotions with words is not a problem for
me’’) or negative (15 items, e.g., ‘‘I often find it difficult to show my affection
to those close to me’’). These items are adequate when examined by
traditional evaluative criteria (such as coefficient alpha which is currently
used as the major index of the scale’s quality). No items describing neutral
behavior exist (i.e., items with a moderate standing on the latent trait).
These items were excluded from the process of scale construction since they
were judged as poor by traditional criteria. This exclusion however, may
have unintended consequences in terms of content or measurement precision
in particular regions of the trait EI continuum.
Responses to the TEIQue-SF items are made on a Likert-type scale (e.g.
1 ¼ strongly disagree and 7 ¼ strongly agree); the total scale scoring is
derived by summing the score on each item in the scale (after reverse scoring
for negative items) and is used to locate respondents on the latent trait
continuum; the higher the score, the higher the trait EI of the individual.
The aforementioned procedure implies: (1) a dominance response process;
that is the probability of observing a high-item score increases mono-
tonically as the distance between person and item locations increases and
(2) the content of each item is equally informative at all levels of trait EI.
Measurement of Trait Emotional Intelligence with TEIQue-SF 293

Unfolding Models

Unfolding (or ideal point) models (Coombs, 1964; Davison, 1977) is a

relative new category of psychometric methods and they belong to the
general family of the IRT models. They are model-based approaches in
understanding the nonlinear relationships between individual characteristics
(e.g., ability and traits), item characteristics (e.g., discrimination and
location), and individuals’ response patterns. Specifically, an unfolding
model represents a proximity relation (Coombs, 1964) in which the
individual endorses an item to the extent that the individual and the item
are located relatively close to each other on a latent attitude continuum.
However, unfolding models are more general models than the commonly
used dominance IRT models, in the sense that a dominance model may be
considered a special case of an ideal point model, where the ideal point
is allowed to be located at infinity (Stark et al., 2006). Contrary, to well-
known IRT models (e.g., Rasch models, two and three parameter logistic
models) which are based on the notion of a dominance response processes
(monotonic), unfolding models are based on non-monotonic ideal point
response processes. Specifically, in unfolding models, the probability of
disagreeing with an item increases as the individual’s trait becomes more
distant in either direction from the item’s location on a trait continuum. In
contrast, in dominance models disagreement increases as the underlying
trait becomes displaced from the item in the negative direction and decreases
as the underlying trait becomes displaced from the item in the positive
direction (Stark et al., 2006) (see Fig. 1).
Three basic assumptions underlie unfolding theory (Coombs, 1964). First,
it is assumed that the trait under investigation is unidimensional and that
persons and stimuli (e.g., statements and items) can be located along the
same single dimension. Second, it is assumed that there is universal
agreement regarding the order of the stimuli along this dimension. Note that
this assumption holds even though individuals are expected to differ in
their locations along the continuum. Finally, it is assumed that a person’s
location is determined by an ideal-point process. That is, a person will be
located closest to his or her most preferred stimulus. An implication of this
response process, as noted in the previous paragraph, is that a person has
one of the two reasons for not endorsing a given item and that reason
depends on the person’s location. If the person is located higher on the
continuum than the item, then the person is said to disagree from above. If
the person is located at a lower level, then the person is said to disagree from
below (Andrich & Styles, 1998; Chernyshenko et al., 2007). In other words
under the unfolding perspective it is assumed that the probability of
294 LEONIDAS A. ZAMPETAKIS

Fig. 1. Unfolding Models versus Cumulative Models. (a) Item Response Function
for an Ideal Point Response Process. (b) Item Response Function for a Dominance
Response Process. Each Curve Represents the Likelihood of Endorsing Least
Positive to most Positive Response Options Based on Theta.

agreeing (or endorsing) with a statement is greatest when there is little

distance between an individual’s level of latent trait and the level of the trait
reflected in the item (Drasgow et al., 2009; Stark et al., 2006).
Measurement of Trait Emotional Intelligence with TEIQue-SF 295

Generalized Graded Unfolding Model

Roberts and colleagues (Roberts & Laughlin, 1996; Roberts et al., 2000,
2006) have developed several unfolding IRT models to describe ideal point
response processes (i.e., the generalized grading scale model, the graded
unfolding model, etc.). The most general of these models is the GGUM of
Roberts et al. (2000).
The GGUM estimates a single person parameter and three item
parameters. The person parameter estimated by the GGUM is theta (y) that
is, the individuals’ level of the latent trait. Theta is expressed as a standardized
score; an individual with y ¼ 1.0 has a value on the latent trait that is one
standard deviation above the mean. The first parameter estimated by GGUM
is the item location parameter (delta, d). This parameter identifies the location
of the item on the continuum of the latent trait. The location parameter is on
the same metric as theta (i.e., z-scores). Thus, it is used to determine if an
individual’s level of theta is above or below an item and the size of the
difference between the location of the item and the person.
The second parameter is the subjective response thresholds (t). These
parameters represent the location of the subjective boundaries between the
response options relative to the item location parameter; they are also on a
z-score metric. For each item, the number of subjective response thresholds
equals the number of objective response options. The value of the subjective
response threshold that is associated with the most positive objective
response is set to 0.0. The last parameter that is estimated is the item
discrimination parameter (a); it reflects the degree to which the subjective
response choices vary between items as the level of the latent trait changes.
The GGUM estimates only one discrimination parameter per item.
Mathematically, the GGUM is expressed as,

P½Z i ¼ zyj 
   
Pz Pz
exp ai ½zðyj  di Þ  tik  þ exp ai ½ðM  zÞðyj  di Þ  tik 
k¼0 k¼0
¼ C  
P Pw P
w
expðai ½wðyj  di Þ  tik Þ þ exp ai ½ðM  wÞðyj  di Þ  tik 
w¼0 k¼0 k¼0

where Zi represents the objective response to item i, z the level of agreement

with the item (z ¼ 0 is the strongest level of disagreement and z ¼ C is the
strongest level of agreement), C represents one minus the number of objective
response options, M equals 2C þ 1, yj represents the location of the individual j
on the latent trait continuum, di represents the location of item i on the latent
296 LEONIDAS A. ZAMPETAKIS

trait continuum, tk represents the kth subjective response threshold for item i,
and ai represents the discrimination parameter for item i.
The logic underlying the GGUM is in line with unfolding theory (see
Roberts et al., 2000): the GGUM assumes joint placement of persons and
items on a bipolar continuum spanning from negative to neutral to positive
positions. Persons are located on the trait continuum according to their
responses on a set of items; items are located according to their content.
The GGUM is an example of an ideal point IRT model, which seems to be
well suited for personality and attitude data; furthermore it does not require
reverse scoring and it can accommodate a wide variety of single statement
items (Chernyshenko et al., 2007; Roberts et al., 2000; Stark et al., 2006;
Weekers, & Meijer, 2008). When responses fit the GGUM, accurate item
parameter estimates can be obtained with approximately 750 examinees
using a marginal maximum likelihood procedure (Roberts et al., 2000).
Roberts et al. (2000) also showed that an expected a posteriori (EAP)
procedure can produce accurate person estimates when responses to 15–20
informative 6-category items are available.

Rationale of the Study

The present study heeds calls for increasing confidence in the benefits and
utility of unfolding theory (Stark et al., 2006; Weekers & Meijer, 2008;
Zampetakis, 2010), by applying the GGUM to the English version of the
TEIQue-SF.
As already stated, unfolding models are more general than dominance IRT
models. This implies greater flexibility regarding the assumptions of the
underlying response process, which in turn, is important for the prediction of
behavior. Using a more general model may prevent misspecification of the
response process. Recent studies indicate that unfolding models can provide
as good or better fit to personality items than dominance models (Drasgow
et al., 2009; Liao & Mead, 2009; Scherbaum et al., 2006; Stark et al., 2006).
Thus, one objective of the present study was to examine the fit of the GGUM
using data obtained with the TEIQue-SF, which was clearly constructed
under the notion of a dominance response process. An adequate fit of the
GGUM to the data would provide preliminary evidence that indeed
unfolding models are more general models than dominance IRT models.
In the case of the TEIQue-SF, since there are no neutral items included,
we expect that folding, which provides visible support for ideal point
models, will not become evident on the typical range of theta (i.e. 3 – þ 3
standard deviations from the mean). This implies that dominance models
Measurement of Trait Emotional Intelligence with TEIQue-SF 297

will approximate the response process equally well (see e.g., Stark et al.,
2006). As such, we expect that the correlations of the scores obtained for the
GGUM latent trait estimates and total summative scores (obtained from
classical test theory) will be highly correlated on the typical range of theta.
The TEIQue-SF consists of a combination of negatively and positively
worded items. Very often researchers, implicitly assume that positively worded
items measure the same latent trait as negatively worded items, which is not
necessarily. Furthermore, method effects associated with item wording have
been explored in a variety of instruments and have included systematic
measurement errors that disrupt analyses and interpretations of the results
(DiStefano & Motl, 2006; Meganck, Vanheule, & Desmet, 2008). For unfolding
models it is not necessary to use reverse-scored items (Stark et al., 2006).
According to Petrides (2009) although the TEIQue-SF is designed to
measure global trait EI, four subscales can be derived through a priori scoring
based on the scoring key of the full form of the inventory. The four subscales
are: well-being, self-control, emotionality, and sociability. We were interested
to investigate how different items of the TEIQue perform when they are
examined under their respective subscales, compared to the whole 30 item scale.
Finally, belonging to the family of IRT, unfolding models share the same
advantages of IRT over classical test theory methods: (1) item parameters are
not dependent on the sample used to generate the parameters, and are assumed
to be invariant (within a linear transformation) across divergent groups within
a research population and across populations; (2) the person parameter is not
specific to the set of items forming the test; and (3) measurement precision is
not assumed to be constant; instead IRT methods allow researchers to
calculate conditional standard errors of measurement. This can assist greatly
in the identification of items that may contribute little to measurement
precision (Embretson, 1996; Embretson & Reise, 2000).

METHODS

Participants and Procedure

The participants comprised 866 individuals, of whom 432 were male and 416
were female (18 participants did not record their gender). Participants were
recruited both from university campuses and from the general community.
A variety of recruitment methods and incentives were used, including word
of mouth, advertising through social network sites, course credit, and course
data collection. Most of the questionnaires were completed in participants’
own time, although some were collected during supervised class sessions.
298 LEONIDAS A. ZAMPETAKIS

Participants’ age ranged from 17 to 80 years, with a mean age of 26.97

years (SD ¼ 10.29). The sample was highly educated, with 20% holding
high-school diplomas, 41% undergraduate diplomas, 26% postgraduate
diplomas, and 3% PhDs (10% ‘‘other’’).
All participants having provided their informed consent were asked to
read the instructions at the top of the form and to attempt to answer all
questions. After completing the measure, participants were debriefed and
thanked for their time.
For an analysis of this dataset using the related technique of graded response
model (GRM; Samejima, 1969), see ‘‘Study 2’’ in Cooper and Petrides (2010).

Classical Item Analysis of the TEIQue-SF

The TEIQue-SF consists of 30 items designed to measure global trait

emotional intelligence (e.g., I usually find it difficult to regulate my emotions
and I’m usually able to influence the way other people feel). The TEIQue-SF is
derived from the full form of the TEIQue, which covers 15 distinct facets.
Based primarily on correlations with total facet scores, two items from each
of the 15 facets were selected for inclusion in the short form, which uses a
Likert-style response option format, ranging from 1 (Completely Disagree)
to 7 (Completely Agree). A global trait EI score is calculated by summing up
the item scores and dividing by the total number of items. The TEIQue-SF
does not yield scores on the 15 trait EI facets.
The four subscales of the TEIQue-SF are derived through a priori scoring
based on the scoring key of the full form of the inventory (Petrides, 2009).
Items 3, 18, 14, and 29 contribute only to the global trait EI score without
belonging to a specific subscale (Petrides, 2009). The latest version of the
TEIQue-SF (v. 1.50) is available, free of charge, for research purposes from
www.psychometriclab.com.
Initially, we performed a traditional item analysis of the TEIQue-SF data.
First, all negatively worded items in the pool were reverse scored so that
only positive item–total correlations were observed. The internal consistency
of the responses was then examined using all of the items on the TEIQue-SF
and then with each subscale. The classical item analyses also estimated the
item discrimination as the corrected item–total correlations and the item
difficulty as the item mean. The classical indices of item discrimination and
difficulty are presented in Table 1. The indices indicate that most of the
items are highly and equally discriminating. The mean item scores indicate
that most of the items are ‘‘easy’’ in that the mean is above the midpoint on
Measurement of Trait Emotional Intelligence with TEIQue-SF 299

Table 1. Item parameters from the Classical Item Analysis and

Generalized Graded Unfolding Model (GGUM) for the TEIQue-SFa.
Item TEIQue-SF CITC M Standard a (SE) d (SE) Max
Dimension Deviation IIF

1. Expressing my Emotionality 0.34 4.91 1.66 0.26 (0.04) 6.06 (0.08) 0.20
emotions with words
is not a problem for
me
2. I often find it Emotionality 0.33 5.30 1.47 0.25 (0.04) 5.59 (0.07) 0.22
difficult to see things
from another
person’s viewpointR
3. On the whole, I’m a 0.47 5.16 1.32 0.61 (0.05) 4.64 (0.08) 0.69
highly motivated
person
4. I usually find it Self-control 0.39 4.59 1.57 0.31 (0.04) 5.29 (0.06) 0.25
difficult to regulate
my emotionsR
5. I generally don’t find Well-being 0.45 5.62 1.54 0.50 (0.04) 5.78 (0.05) 0.91
life enjoyableR
6. I can deal effectively Sociability 0.49 5.25 1.32 0.64 (0.04) 3.94 (0.06) 0.92
with people
7. I tend to change my Self-control 0.32 4.12 1.62 0.24 (0.03) 5.87 (0.07) 0.15
mind frequentlyR
8. Many times, I can’t Emotionality 0.45 4.91 1.64 0.37 (0.03) 4.97 (0.05) 0.41
figure out what
emotion I’m feelingR
9. I feel that I have a Well-being 0.45 5.64 1.19 0.68 (0.06) 4.04 (0.06) 1.31
number of good
qualities
10. I often find it Sociability 0.39 4.58 1.67 0.33 (0.04) 6.08 (0.11) 0.28
difficult to stand up
for my rightsR
11. I’m usually able to Sociability 0.34 4.67 1.26 0.38 (0.04) 5.20 (0.06) 0.23
influence the way
other people feel
12. On the whole, I have Well-being 0.46 5.23 1.59 0.45 (0.04) 4.63 (0.07) 0.53
a gloomy perspective
on most thingsR
13. Those close to me Emotionality 0.41 5.64 1.47 0.42 (0.04) 5.14 (0.08) 0.60
often complain that
I don’t treat them
rightR
300 LEONIDAS A. ZAMPETAKIS

Table 1. (Continued )
Item TEIQue-SF CITC M Standard a (SE) d (SE) Max
Dimension Deviation IIF

14. I often find it 0.48 5.07 1.50 0.53 (0.04) 4.29 (0.05) 0.65
difficult to adjust my
life according to the
circumstancesR
15. On the whole, I’m Self-control 0.49 4.92 1.45 0.60 (0.05) 4.52 (0.08) 0.81
able to deal with
stress
16. I often find it Emotionality 0.41 4.72 1.84 0.31 (0.03) 5.67 (0.08) 0.32
difficult to show my
affection to those
close to meR
17. I’m normally able to Emotionality 0.34 4.91 1.46 0.30 (0.04) 5.99 (0.07) 0.22
‘‘get into someone’s
shoes’’ and
experience their
emotions
18. I normally find it 0.48 4.80 1.55 0.56 (0.04) 3.97 (0.08) 0.68
difficult to keep
myself motivatedR
19. I’m usually able to Self-control 0.37 5.04 1.41 0.38 (0.03) 4.89 (0.06) 0.34
find ways to control
my emotions when I
want to
20. On the whole, I’m Well-being 0.56 5.50 1.27 0.96 (0.08) 4.11 (0.05) 1.58
pleased with my life
21. I would describe Sociability 0.45 5.02 1.26 0.55 (0.06) 4.95 (0.06) 0.45
myself as a good
negotiator
22. I tend to get involved Self-control 0.36 4.18 1.56 0.29 (0.03) 6.78 (0.07) 0.18
in things I later wish
I could get out ofR
23. I often pause and Emotionality 0.15 4.63 1.52 0.07 (0.02) 13.85 (0.05) 0.01
think about my
feelings
24. I believe I’m full of Well-being 0.52 5.17 1.23 0.78 (0.06) 4.12 (0.06) 0.90
personal strengths
25. I tend to ‘‘back Sociability 0.30 4.60 1.60 0.24 (0.03) 6.08 (0.11) 0.16
down’’ even if I
know I’m rightR
26. I don’t seem to have Sociability 0.43 4.80 1.41 0.42 (0.03) 4.41 (0.06) 0.40
any power at all over
other people’s
feelingsR
Measurement of Trait Emotional Intelligence with TEIQue-SF 301

Table 1. (Continued )
Item TEIQue-SF CITC M Standard a (SE) d (SE) Max
Dimension Deviation IIF

27. I generally believe Well-being 0.48 5.34 1.33 0.70 (0.04) 4.20 (0.07) 0.94
that things will work
out fine in my life
28. I find it difficult to Emotionality 0.49 5.37 1.59 0.51 (0.08) 5.09 (0.08) 0.79
bond well even with
those close to meR
29. Generally, I’m able 0.50 5.52 1.22 0.75 (0.09) 3.68 (0.05) 1.16
to adapt to new
environments
30. Others admire me Self-control 0.30 4.58 1.50 0.28 (0.07) 6.56 (0.08) 0.17
for being relaxed

Note: N ¼ 886. R ¼ item was reverse scored for classical test theory (CTT) analyses;
CITC ¼ corrected item total correlation; a ¼ discrimination parameter; d ¼ location parameter;
and Max IIF ¼ maximum value of the item information function. Items 3, 18, 14, and 29
contribute only to the global trait EI score, under CTT analyses.
a
TEIQue-SF items reproduced with permission by K. V. Petrides (copyright holder). The
instrument is available at www.psychometriclab.com

the scale. Cronbach alpha reliabilities for the well-being, self-control,

emotionality, and sociability were 0.75 (6 items), 0.66 (6 items), 0.66
(8 items), and 0.70 (6 items), respectively. Cronbach’s reliability coefficient
for all 30 items was 0.88. The mean and standard deviation for the total trait
EI scale and the four subscales was: 4.99 (SD ¼ 0.68), 5.41 (SD ¼ 0.91),
4.57(SD ¼ 0.92), 5.05(SD ¼ 0.86), and 4.82 (SD ¼ 0.89), respectively.
For all participants, the skewness and kurtosis values for the items ranged
from 1.35 to 0.06 and 0.98 and 2.42, respectively. For the global score
and subscale scores the skewness and kurtosis values ranged from 0.56 to
0.04 and 0.09 to 0.34. According to Curran, West, and Finch (1996), for
univariate normality, skewness, and kurtosis values of 0–2, and 0–7,
respectively, can be taken as demonstrating sufficient normality. On this
basis, all of the TEIQue-SF items as well as the global score and subscale
scores showed sufficient normality.

Assessing the Dimensionality of the TEIQue-SF

IRT models assume that the latent trait construct space is either strictly
unidimensional, or as a practical matter, dominated by a general underlying
302 LEONIDAS A. ZAMPETAKIS

factor. According to Drasgow and Hulin (1990), the unidimensionality

assumption is reasonably met if there is a dominant factor in the data. As a
rule of thumb, the first factor needs to account for at least 20% of the total
variance for the item parameters to be stable (Reckase, 1979). Furthermore,
Morizot, Ainsworth, and Reise, (2007) proposed as a rule of thumb that
the ratio of the first to the second eigenvalue should be above 3, for
unidimensionality to be considered appropriate.
We checked the assumption of the unidimensionality of the whole scale
(and of the four subscales) by examining the amount of variance explained
by the first factor on exploratory factor analysis (EFA). We extracted
factors from the sample correlation matrix using the exploratory maximum
likelihood (ML) method as implemented in the FACTOR software
(Lorenzo-Seva & Ferrando, 2006). The number of factors extracted was
based on the results of the minimum average partial test (MAP; O’Connor,
2000) and a visual inspection of the scree plot.
The Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy was
0.89, indicating that EFA was appropriate for the 30 item scale and for this
sample. The results of the MAP test suggested a two-factor solution,
although a visual inspection of the scree plot pointed to the presence of one
dominant factor. The first six eigenvalues were 6.91, 2.11, 1.63, 1.52, 1.40,
and 1.22. The first eigenvalue accounted for 23.05% of the total variance,
whereas the second factor accounted for an additional 7%. Using Morizot
et al.’s (2007) rule of thumb, the ratio of the first to second eigenvalue was
3.27, indicating the presence of a sufficiently strong single factor for an IRT
analysis.
Results for the first eigenvalue of four subscales, that is, well-being, self-
control, emotionality, and sociability were 2.67 (44.5% of the total
variance), 2.22 (37% of the total variance), 2.45 (30% of the total variance),
and 2.37 (39% of the total variance), respectively. The percentage of
variance accounted for exceeds the rule of thumb of 20% and EFA results
support the existence of a sufficiently strong unidimensional factor for the
30 item TEIQue scale and for each one of the 4 subscales.

Analytic Strategy

Two alternative approaches to assess model fit have explicitly been

advocated to be applied for fitting IRT models: (1) w2 goodness of fit tests
for single items (singlets), pairs of items (doublets), and three items (triplets)
(Drasgow, Levine, Tsien, Williams, & Mead, 1995) and (2) graphical
Measurement of Trait Emotional Intelligence with TEIQue-SF 303

inspection of observed vs. expected item response curves (fit plots;

Chernyshenko, Stark, Chan, Drasgow, & Williams, 2001; Drasgow et al.,
1995).
Following Drasgow et al. (1995) and Chernyshenko et al. (2001) model-
data fit was examined with fit plots and w2 goodness of fit tests for single
items, pairs, and triplets. The idea behind fit plots is to plot both the
expected category response function and the observed category response
function with 95% confidence intervals (CI; estimated from the observed
frequency of category choices, see Drasgow et al., 1995). Deviations of
expected and observed category response functions, i.e., if the 95% CIs do
not include the expected response functions, indicate model misfit. Cherny-
shenko et al. (2001) demonstrated the usefulness of fit plots by showing that
some kinds of misfit can be detected by fit plots but not by summary
statistics.
For the w2 goodness of fit tests, indices are formed of the difference
between the expected frequency of responses for the options and the
observed frequency of responses for the options. However, w2 are very often
viewed as inconclusive evidence of adequate model-data fit because of their
sensitivity to sample size and their insensitivity to certain forms of model-
data misfit (Chernyshenko et al., 2001). To bypass the sensitivity to sample
size and to allow comparisons between different samples and tests, w2 was
adjusted to the magnitude that would be expected in a sample of 3,000
(Chernyshenko et al., 2001; Drasgow et al., 1995). Then, the ratio of w2 to
the degrees of freedom was computed. Drasgow et al., (1995) found that
best fitting models had small (below 3.0), adjusted w2 to degrees of freedom
ratios for item singlets as well as small ratios for pairs and triplets.
To summarize the results of the w2 analyses for the model, frequency
tables were constructed for the adjusted w2 statistics having values in seven
intervals. Means and standard deviations of adjusted w2/df ratios were also
computed for each subscale.
Item parameters for the GGUM were estimated with the GGUM2004
(version 1.1) computer program using a marginal maximum likelihood
(MML) approach (Roberts et al., 2002; Roberts et al., 2006). GGUM2004
was also used to compute the item and test information functions. Although
GGUM2004 also contains item and model fit statistics, such as infit and
outfit statistics, according to the manual these statistics are generalized from
cumulative IRT applications and are not mathematically deduced for
GGUM. Little is known about their distribution, their power, and their
Type I error rate. Therefore the MODFIT (version 1.1) computer program
(Stark, 2001) was used to compute w2 statistics and fit plots for the GGUM.
304 LEONIDAS A. ZAMPETAKIS

Tests of Model Fit

w2 to degrees-of-freedom ratio fit statistics, provided by MODFIT, showed

favorable results indicating the appropriateness of the GGUM for the 30
item scale (Table 2). In addition, all subscales showed very good fit for
singlets, doublets, and triplets.
Taken together, the results of the w2 to degrees-of-freedom ratio fit
statistics were in favor of retaining the GGUM. In the second step of model
fit evaluation, fit plots were considered for all category response functions.
No significant misfit occurred and the expected response functions were
almost always included in the 95% CIs for the observed response functions.
Since the w2 tests and the fit plots did not show large differences between
the observed responses and the expected response functions, the unidime-
sional GGUM was accepted for the TEIQue-SF scale; that is, a single latent
trait is sufficient to account for item responding in the TEIQue-SF. This
holds also for the four subscales.

Parameter Estimates and Item Properties

Table 2 presents the parameters from the GGUM for each item on the
TEIQue-SF. According to Baker (2001), discrimination values 0.01–0.24 are
very low, 0.25–0.64 are low, 0.65–1.34 are moderate, 1.35–1.69 are high, and
more than 1.7 are very high. In our study, the majority of items had low to
moderate discrimination parameters, with the values of the discrimination
parameter range from 0.08 to 0.95. In the classical item analysis, there were
several items that were equally discriminating; however the corresponding
discrimination parameter from GGUM analysis indicated that the dis-
crimination was not equivalent. For example, the 8th, 9th and 21st items

Table 2. Frequencies of the Values of the Adjusted (N ¼ 3,000) w2

Statistic to Degrees of Freedom from the Model Fit Analysis for the
GGUM Model for All 30 Items of the TEIQue-SF.
Model Mean Standard
o1 1–o2 2–o3 3–o4 4–o5 5–o7 W7 Deviation

GGUM Singlets 30 0 0 0 0 1 0 0.08 0.02

Doublets 3 5 4 5 5 6 2 3.83 2.26
Triplets 0 1 2 1 1 4 1 4.91 2.26
Measurement of Trait Emotional Intelligence with TEIQue-SF 305

on the TEIQue-SF scale have values of the discrimination parameters

that are equal in the classical item analysis, but the values of the GGUM
discrimination parameters demonstrate a difference. Moreover, the classical
item analysis suggests that there is much less variability in the discrimi-
nation parameter than the GGUM analysis. These parameters are important
because they indicate the score differences one could expect for individuals
with similar levels of the latent trait of TEIQue-SF. Greater differences are
expected when there is a higher level of discrimination. For example, when
estimating the total (summative) scale scores under CTT, there are
12 respondents with a score of 4.6, 17 respondents with a score of 4.8, and
21 respondents with a score of 4.9. In the case of GGUM estimated
person parameters, there are only 2 individuals (out of 866) with the same
score (3.6). For the global GGUM TEIQue-SF score and subscale scores, the
skewness and kurtosis values ranged from 0.16 to 0.75 and 0.44 to 1.22,
respectively.
The location parameter, d (Table 2) indicates that the items of the
TEIQue-SF are either on the extreme positive, or on the extreme negative
dimension of the trait EI continuum; note that under the unfolding
perspective, the reverse scoring of negatively worded items is not necessary.
These values also indicate the point on the trait EI continuum where
yd ¼ 0.00. For example, on the item 27, ‘‘I generally believe that things will
work out fine in my life,’’ yd will equal 0.00 at 4.20. Thus, perfect
proximity is achieved for individuals with high levels of trait EI (at least four
standard deviations above the mean). This also indicates that the point at
which the response curves become non-monotonic is at the upper positive
end of the distribution. This was the case for all of the items on the TEIQue-
SF scale. The extreme positive and extreme negative location parameters,
suggest that folding occurs at high trait EI levels where very few individuals
were located. Under the unfolding perspective when the items are extreme
relative to the sample of persons, as in our case, then the IRFs will be
approximately monotonic and S-shaped and the data could be characterized
well by either unfolding or dominance models (see Fig. 2, panel a). This is
clearly indicated by the Pearson’s correlation (0.97, po0.01) between the
GGUM latent trait scores and total (summative) scale scores under CTT for
the TEIQue scale (see Table 3).
However, when we examined the four subscales individually, several items
exhibited folding. Specifically 2 items of the well-being subscale, items 20
and 27 showed folding at the higher end of the trait continuum; 3 items of
the self-control subscale, items 4, 7, and 22 showed folding at the lower end;
4 items of the emotionality subscale, items 8, 13, 16, and 28 showed folding
306 LEONIDAS A. ZAMPETAKIS

Fig. 2. Item Response Functions for Item 4 When Examined Under (a) the
TEIQue-SF Scale and, (b) for the Self-Control Subscale.
Measurement of Trait Emotional Intelligence with TEIQue-SF 307

Table 3. Descriptive Statistics and Pearson’s Product Moment

Correlations of the GGUM Latent Trait Scores and Total (Summative)
Scale Scores under CTT for the TEIQue-SF Scale and Subscales.
M Standard 1 2 3 4 5 6 7 8 9 10
Deviation

1. Global trait EI_ctta 4.99 0.69 –

2. Well-Being_ctt 5.41 0.91 0.80 –
3. Self-Control_ctt 4.57 0.92 0.71 0.46 –
4. Emotionality_ctt 5.05 0.86 0.77 0.51 0.39 –
5. Sociability_ctt 4.82 0.89 0.72 0.50 0.37 0.42 –
6. Global trait E_ggumb 0.00 0.93 0.97 0.82 0.68 0.69 0.71 –
7. Well-Being_ggum 0.00 0.88 0.79 0.96 0.47 0.49 0.50 0.84 –
8. Self-Control_ggum 0.00 0.82 0.70 0.47 0.94 0.37 0.39 0.69 0.49 –
9. Emotionality_ggum 0.00 0.84 0.75 0.49 0.39 0.93 0.43 0.71 0.49 0.40 –
10. Sociability_ggum 0.00 0.84 0.72 0.51 0.36 0.43 0.97 0.73 0.52 0.39 0.45 –

Note: N ¼ 886; all correlations are significant at po0.01 (two tailed).

a
Scores obtained under the CTT (classical test theory) scoring protocol (summative).
b
Scores obtained under the GGUM (generalized graded unfolding model).

at the lower end of the trait continuum and finally an item of the sociability
subscale, (item 26) showed folding at the lower end of the trait continuum.
This finding was interesting given that all subscales were derived though
a priori scoring; they have not been derived through CTT methods.
Furthermore this explains the lower correlations among the subscales under
the two modes (Table 3).
In Fig. 2 we present the IRF for item 4, of the self-control scale: ‘‘I usually
find it difficult to regulate my emotions.’’ Panel a represents the IRF of the
item when examined under the whole TEIQue-SF scale; clearly a monotonic
relationship exists suggesting that individuals high in trait EI would have a
much lower probability of agreement than those at lower levels. In panel b,
we present the IRF for item 4 when this is examined under the self-control
subscale; clearly the item shows unfolding at the low levels of the self-
control continuum; this moderately negative item (d ¼ 1.39) is endorsed
mostly by respondents with moderately low levels of self-control, and
progressively less by those rat higher levels on this subscale.
Scatter plots of the theta values obtained under the two models, about
the diagonal line indicated that the theta values are differentially ordered
(for the global score and subscale scores) especially at the upper and lower
extremes (Fig. 3).
308 LEONIDAS A. ZAMPETAKIS

Global trait EI scores under GGUM 3

-1

-3

-5
2 3 4 5 6 7
Global trait EI scores under CTT
Self control scores under GGUM
Well Being scores under GGUM

2
2
1

0 1

-1 0

-2 -1
-3
-2
-4
-3
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Well Being scores under CTT Self control scores under CTT
Emotionality scores under GGUM
Sociability scores under GGUM

2
2

1
0
0

-1
-2
-2

-3
-4
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Sociability scores under CTT Emotionality scores under CTT

Fig. 3. Scatter Plot Comparisons of the Theta Value Estimates from GGUM
and Traditional Scoring System (CTT) for Global Trait EI and the Four
Subscales. Every Circle Represents a Person’s Trait Estimates under the Two
Perspectives.
Measurement of Trait Emotional Intelligence with TEIQue-SF 309

Item and Test Information Functions

In Table 1 we present the maximum value of the item information functions

for the GGUM. As can be seen in the table, several of the items provide
substantial amounts of information (e.g., items 9 and20) while others
provide very little (e.g., item 23). The test information function and standard
error of measurement for the GGUM are presented in Fig. 4, for the
TEIQue-SF. The TEIQue-SF demonstrates the highest levels of

TIF Plot

70.0

60.0
Sum of IIF Values

50.0

40.0

30.0

20.0

10.0

0.0
–3.0 –2.0 –1.0 0.0 1.0 2.0 3.0

Theta

SE Plot

0.4

0.3
Standard Error

0.3

0.2

0.2

0.1

0.1

0.0
–3.0 –2.0 –1.0 0.0 1.0 2.0 3.0

Theta

Fig. 4. The Test Information Function (TIF) for All of the Items in the Analysis
along with the Standard Error (SE) Plot.
310 LEONIDAS A. ZAMPETAKIS

measurement precision at the lower levels of the trait EI continuum. From

the standard error plot it can be seen that the standard error is
sGGUMr0.20 for trait levels of about –3.0ry r1.0, which is considerably
lower than a standard error computed under the CTT approach. Under the
classical approach, equal standard errors of measurement would be assumed
for all trait levels and estimated from the reliability estimate of the test, e.g.,
by using Cronbach’s a ¼ 0.88 for the present sample, resulting in a standard
error estimate of sqrt(1a) ¼ 0.35.

DISCUSSION
In this chapter, a unidimensional unfolding IRT model, the GGUM
(Roberts et al., 2000), was successfully fitted to data obtained with the UK
version of the TEIQue-SF (Petrides & Furnham, 2006). Self-report scales
are amenable to testing using unfolding models given underlying assump-
tions about the spread of the latent trait on a continuum spanning from low
to high. Parameter estimates, test information function, and the standard
error plot, demonstrated that the 30-item TEIQue-SF scale can be used to
measure trait EI accurately at lower to middle scores of the trait level scores.
However, the results from the unfolding model suggested that at higher trait
EI scores the standard error increased, indicating less measurement
precision for those with very high global trait EI. Furthermore, our results
indicated that all the items of the TEIQue-SF had extreme item location
parameters and moderate to low discrimination parameters. This implies
that: (a) not all items span the trait EI continuum evenly, which may
adversely affect the accuracy of measurement and (b) that items with low
discrimination parameters provide little information and thus contribute
little to reducing the error in trait estimates.
At this point, one might sensibly ask why GGUM is a good idea to use in
the assessment of trait EI, as it is not clear that this model is the best one to
use to fit to these data. Furthermore, the form of support for the model is a
dominance model, which does not need the added mathematical complica-
tion of the GGUM formulation compared to standard monotonic (i.e., non-
ideal point) models that are typically used in personality research. Stated
differently if item-total correlations were used to select items for the
TEIQue-SF, then why apply an unfolding IRT model, such as the GGUM,
as opposed to a dominance model, such as the GRM?
As already stated unfolding models are more general models than the
commonly used dominance IRT models (Stark et al., 2006). This implies
Measurement of Trait Emotional Intelligence with TEIQue-SF 311

that unfolding models can fit monotonically increasing IRFs (the curves that
graphically indicate the probability of item endorsement or agreement
against trait level) but do not necessarily require this property. The fact that
folding was not evident in our data was due to the extreme positive and
extreme negative item location parameters; folding occurs at high trait EI
levels where very few individuals were located.
Recently Cooper and Petrides (2010) fitted a dominance model IRT
model, the GRM to the same dataset as we did. As far as the fit of the GRM
model is concerned it showed negligible (according to Stephen Stark,
personal communication) improvement over the GGUM (the adjusted w2
to degrees of freedom ratio for single items was 1.05, for doublets 3.71, and
for triplets 3.96; Cooper & Petrides, 2010, pp. 454). However, the added
mathematical complication of the GGUM formulation may further help the
development of more precise scales, since unfolding approaches can include
items located at any point on a trait continuum.
Taken as a whole, the findings from our analyses conducted using GGUM
suggest that the TEIQue-SF is less likely to be well suited for individuals with
higher levels of trait EI; for such individuals the development of new items
should be considered. Future psychometric development of TEIQue-SF,
research could concentrate on the inclusion of items with location parameters
that are spread evenly across the trait continuum. Specifically, under an
unfolding perspective, neutral items (those with d parameters close to zero)
could generally help to measure respondents who are above and below
average, whereas positive and negative items could provide high information
in the middle and the extremes (Chernyshenko et al., 2001). Furthermore,
future revisions of the scale should address the relatively poor psychometric
qualities of some items. For instance items 1, 2, 7, 23 and 25 (Table 1) have
very low discrimination parameters and low information values. Finally,
future studies could use GGUM to evaluate the effects that gender may have
on the item properties of the scale through differential item functioning
(DIF).
In conclusion, these analyses were conducted to extend recent research
studies (e.g., Chernyshenko et al., 2007; Scherbaum et al., 2006; Weekers, &
Meijer, 2008; Zampetakis, 2010) by demonstrating the potential usefulness
of unfolding models in the assessment of trait EI. We believe that unfolding
models although more complex than dominance models are worth the extra
effort and that researchers should put forth the time and effort to use these
models in their own research. The current study presents the first ideal point
IRT modeling effort in the general field of EI. Single-peaked response
functions have rarely been used in substantive research although modern
312 LEONIDAS A. ZAMPETAKIS

computing algorithms have overcome the problems of time-consuming

and laborious traditional analysis of single-peaked response data. As such,
it helps address the criticism that ‘‘a good deal of EI research has been
conducted without particularly advanced psychometrics’’ (Matthews,
Zeidner, & Roberts, 2007, p. 24).

ACKNOWLEDGMENTS
The author would like to thank Dr. K.V. Petrides and Dr. A. Cooper, for
providing the data analyzed herein and their helpful comments on earlier
versions of the chapter. The author gratefully acknowledges financial
support from Greek State Scholarship Foundation (postdoctoral research
scholarship – No. IKY-801/2009).

REFERENCES
Andrich, D. (1996). A hyperbolic cosine latent trait model for unfolding polytomous responses:
Reconciling Thurstone and Likert methodologies. British Journal of Mathematical and
Statistical Psychology, 49, 347–365.
Andrich, D., & Luo, G. (1993). A hyperbolic cosine latent trait model for unfolding
dichotomous single-stimulus responses. Applied Psychological Measurement, 17,
253–276.
Andrich, D., & Styles, I. M. (1998). The structural relationship between attitude and behavior
statements from the unfolding perspective. Psychological Methods, 3(4), 454–469.
Austin, A. J. (2009). A reaction time study of responses to trait and ability emotional
intelligence test items. Personality and Individual Differences, 46, 381–383.
Baker, F. (2001). The basics of item response theory. College Park: ERIC Clearinghouse on
Assessment and Evaluation. University of Maryland.
Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic studies. New York:
Cambridge University Press.
Chernyshenko, O. S., Stark, S., Chan, K. Y., Drasgow, F., & Williams, B. A. (2001). Fitting
item response theory models to two personality inventories: Issues and insights.
Multivariate Behavioral Research, 36(4), 523–562.
Chernyshenko, O. S., Stark, S., Drasgow, F., & Roberts, B. W. (2007). Constructing personality
scales under the assumptions of an ideal-point response process: Toward increasing the
flexibility of personality measures. Psychological Assessment, 19, 88–106.
Coombs, C. H. (1964). A theory of data. New York: Wiley.
Cooper, A., & Petrides, K. V. (2010). A psychometric analysis of the Trait Emotional
Intelligence Questionnaire-Short Form (TEIQue-SF) using Item Response Theory.
Journal of Personality Assessment, 95(2), 449–457.
Cronbach, L., Gieser, G., Nanda, H., & Rajaratnam, N. (1972). The dependability of behavioral
measurements: Theory of generalizability for scores and profiles. New York: Wiley.
Measurement of Trait Emotional Intelligence with TEIQue-SF 313

Curran, P. J., West, S. G., & Finch, J. F. (1996). The robustness of test statistics to
nonnormality and specification error in confirmatory factor analysis. Psychological
Methods, 1, 16–29.
Davison, M. L. (1977). On a metric, unidimensional unfolding model for attitudinal and
developmental data. Psychometrika, 42, 523–548.
DiStefano, C., & Motl, R. W. (2006). Further investigating method effects associated with
negatively worded items on self-report surveys. Structural Equation Modeling: A
Multidisciplinary Journal, 13, 440–464.
Drasgow, F., Chernyshenko, O. S., & Stark, S. (2009). Test theory and personality
measurement. In: Oxford Handbook of Personality Assessment (Chapter 4, pp. 59–80).
London: Oxford University Press
Drasgow, F., & Hulin, C. L. (1990). Item response theory. In: M. D. D. L. M. Hough (Ed.),
Handbook of industrial and organizational psychology (2nd ed., Vol. 1, pp. 577–636). Palo
Alto, CA: Consulting Psychologists’ Press.
Drasgow, F., Levine, M. V., Tsien, S., Williams, B., & Mead, A. (1995). Fitting polytomous
item response theory models to multiple-choice tests. Applied Psychological Measure-
ment, 19, 143–165.
Embretson, S. E. (1996). The new rules of measurement. Psychological Assessment, 8(4),
341–349.
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Mahwah, NJ:
Lawrence Erlbaum.
Freudenthaler, H. H., Neubauer, A. C., Gabler, P., Scherl, W. G., & Rindermann, H.
(2008). Testing and validating the Trait Emotional Intelligence Questionnaire
(TEIQue) in a German-speaking sample. Personality and Individual Differences, 45,
673–678.
Johnson, M., & Junker, B. W. (2003). Using data augmentation and Markov chain Monte
Carlo for the estimation of unfolding response models. Journal of Educational and
Behavioral Statistics, 28, 195–230.
Liao, C., & Mead, A. D. (2009). Fit of ideal-point and dominance IRT models to simulated data.
Paper presented at the Twenty-fourth annual meeting of the Society for Industrial and
Organizational Psychology. New Orleans, LA.
Lord, F. M., & Novick, M. R. (1968). Statistical theories of mental test scores. Reading, MA:
Addison-Wesley.
Lorenzo-Seva, U., & Ferrando, P. J. (2006). FACTOR: A computer program to fit the
exploratory factor analysis model. Behavior Research Methods, 38(1), 88–91.
Luo, G. (2000). Joint maximum likelihood estimation for the hyperbolic cosine model for single
stimulus responses. Applied Psychological Measurement, 24, 33–49.
Luo, G., Andrich, D., & Styles, I. (1998). The JML estimation of the generalized unfolding
model incorporating the latitude of acceptance parameter. Australian Journal of
Psychology, 50(3), 187–198.
Matthews, G., Zeidner, M., & Roberts, R. D. (2007). Emotional intelligence: Consensus,
controversies, and questions. In: G. Matthews, M. Zeidner & R. D. Roberts (Eds.), The
science of emotional intelligence: Known and unknowns. New York: Oxford University
Press.
Meganck, R., Vanheule, S., & Desmet, M. (2008). Factorial validity and measurement
invariance of the 20-item Toronto Alexithymia Scale in clinical and nonclinical samples.
Assessment, 15(1), 36–47.
314 LEONIDAS A. ZAMPETAKIS

Mikolajczak, M., Menil, C., & Luminet, O. (2007). Explaining the protective effect of trait
emotional intelligence regarding occupational stress: Exploration of emotional labor
processes. Journal of Research in Personality, 41, 1107–1117.
Morizot, J., Ainsworth, A. T., & Reise, S. (2007). Toward modern psychometrics: Application
of item response theory models in personality research. In: R. W. Robins, R. C. Fraley,
& R. F. Krueger (Eds.), Handbook of research methods in personality psychology
(pp. 407–423). New York: Guilford.
O’Connor, B. P. (2000). SPSS and SAS programs for determining the number of components
using parallel analysis and Velicer’s MAP test. Behavior Research Methods, Instruments, &
Computers, 32(3), 396–402.
Petrides, K. V. (2009). Technical manual for the Trait Emotional Intelligence Questionnaires
(TEIQue). London: London Psychometric Laboratory.
Petrides, K. V., & Furnham, A. (2001). Trait emotional intelligence: Psychometric investigation
with reference to established trait taxonomies. European Journal of Personality, 15,
425–448.
Petrides, K. V., & Furnham, A. (2006). The role of trait emotional intelligence in a gender-
specific model of organizational variables. Journal of Applied Social Psychology, 36,
552–569.
Petrides, K. V., Pérez-González, J. C., & Furnham, A. (2007). On the criterion and incremental
validity of trait emotional intelligence. Cognition and Emotion, 21(1), 26–55.
Petrides, K. V., Pita, R., & Kokkinaki, F. (2007). The location of trait emotional intelligence in
personality factor space. British Journal of Psychology, 98, 273–289.
Petrides, K. V., Vernon, P. A., Schermer, J. A., Ligthart, L., Boomsma, D. I., & Veselka, L.
(2010). Relationships between trait emotional intelligence and the Big Five in the
Netherlands. Personality and Individual Differences, 48, 906–910.
Reckase, M. D. (1979). Unifactor latent trait models applied to multifactor tests: Results and
implications. Journal of Educational Statistics, 4, 207–230.
Roberts, J. S., Donoghue, J. R., & Laughlin, J. E. (2000). A general item response theory model
for unfolding unidimensional polytomous responses. Applied Psychological Measure-
ment, 24, 3–32.
Roberts, J. S., Donoghue, J. R., & Laughlin, J. E. (2002). Characteristics of MML/EAP
parameter estimates in the generalized graded unfolding model. Applied Psychological
Measurement, 26, 192–207.
Roberts, J. S., Fang, H., Cui, W., & Wang, Y. (2006). GGUM2004: A Windows-based program
to estimate parameters of the generalized graded unfolding model. Applied Psychological
Measurement, 30, 64–65.
Roberts, J. S., & Laughlin, J. E. (1996). A unidimensional item response model for unfolding
responses from a graded disagree-agree response scale. Applied Psychological Measure-
ment, 20, 231–255.
Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores.
Psychometrika Monograph Supplement, 34, 100–114.
Scherbaum, C. A., Finlinson, S., Barden, K., & Tamanini, K. (2006). Applications of item
response theory to measurement issues in leadership research. Leadership Quarterly, 17,
366–386.
Sevdalis, N., Petrides, K. V., & Harvey, N. (2007). Trait emotional intelligence and decision-
related emotions. Personality and Individual Differences, 42, 1347–1358.
 Measurement of Trait Emotional Intelligence with TEIQue-SF 315

Smith, L., Heaven, P. C. L., & Ciarrochi, J. (2008). Trait emotional intelligence, conflict
communication patterns, and relationship satisfaction. Personality and Individual
Differences, 44, 1314–1325.
Stark, S. (2001). MODFIT: A computer program for model-data fit. Unpublished manuscript.
University of Illinois at Urban, Champaign.
Stark, S., Chernyshenko, O. S., Drasgow, F., & Williams, B. A. (2006). Examining assumptions
about item responding in personality assessment: Should ideal point methods be
considered for scale development and scoring? Journal of Applied Psychology, 91, 25–39.
Weekers, A. M., & Meijer, R. R. (2008). Scaling response processes on personality items using
unfolding and dominance models. An illustration with a Dutch dominance and
unfolding personality inventory. European Journal of Psychological Assessment, 24(1),
65–77.
Zampetakis, L. A. (2010). Unfolding the measurement of the creative personality. Journal of
Creative Behavior, 44(2), 105–123.
Zampetakis, L. A., Kafetsios, K., Bouranta, N., Dewett, T., & Moustakis, V. (2009). On the
relationship between Emotional Intelligence and entrepreneurial attitudes and inten-
tions. International Journal of Entrepreneurial Behavior & Research, 15(6), 595–618.

View publication stats