Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=camh20
Aging & Mental Health
ISSN: 1360-7863 (Print) 1364-6915 (Online) Journal homepage: https://www.tandfonline.com/loi/camh20
Personality and suicidal ideation in the elderly:
factorial invariance and latent means structures
across age
Paolo Iliceto, Emanuele Fino, Ugo Sabatello & Gabriella Candilera
To cite this article: Paolo Iliceto, Emanuele Fino, Ugo Sabatello & Gabriella Candilera (2014)
Personality and suicidal ideation in the elderly: factorial invariance and latent means structures
across age, Aging & Mental Health, 18:6, 792-800, DOI: 10.1080/13607863.2014.880404
To link to this article: https://doi.org/10.1080/13607863.2014.880404
Published online: 30 Jan 2014.
Submit your article to this journal
Article views: 412
View related articles
View Crossmark data
Citing articles: 8 View citing articles
Personality and suicidal ideation in the elderly: factorial invariance
and latent means structures across age
Paolo Iliceto
a
, Emanuele Fino
b
*, Ugo Sabatello
c
and Gabriella Candilera
d
a
S&P Statistics and Psychometrics Ltd, Rome, Italy;
b
Department of Developmental and Social Psychology, Sapienza University of
Rome, Rome, Italy;
c
Department of Pediatrics and Child Neuropsychiatry, Sapienza University of Rome, Italy;
d
Clinical Psychologist,
Private Practice, Rome, Italy
(Received 13 September 2013; accepted 30 December 2013)
Objectives: Suicide among the elderly is a dramatic global health problem. Although fatal attempts are frequent in the
elderly, research indicated that they rarely present long-term elaboration of suicidal ideation and communicate their
intents. Consequently, risk factor detection and assessment are salient. Although evidence on the association between
personality and suicidal ideation in young adults is accumulating, little is known about its relevance in the elderly. The
purpose of the present study was to analyze the components of a measurement model that are invariant across young adults
and older adults and then investigate the relations among dimensions of personality and suicide risk. We postulated a
specific relation pattern a priori and tested the hypotheses statistically in order to examine the models for equivalency of
the factorial measurement.
Method: We investigated 316 young adults and 339 older adults, who were administered self-report questionnaires to
assess depression, hopelessness, alternative five-factor model of personality, and self–other perception.
Results: Multigroup confirmatory factor analyses were conducted, yielding a final model with excellent fit to the data. This
model showed a similar pattern of associations between suicidal ideation and personality across both groups.
Conclusions: Although the elderly are exposed to specific life stressors associated with suicidal ideation, our findings
suggest that the elderly and young adults may be similar on personality and psychopathology variables predicting suicidal
ideation than previously hypothesized. Implications are provided for enhanced assessment and intervention of the elderly
high in neuroticism, depression, hopelessness, and with negative self–other perception.
Keywords: elderly; neuroticism; extraversion; suicide; hopelessness
Introduction
Suicide among the elderly is a dramatic global health
problem (World Health Organization, 2012). In industrial-
ized countries, suicides in the elderly are more frequent
than in other age groups, and in the last decade such trend
has seen a substantial increase (Baker, Hu, Wilcox, &
Baker, 2013; Ciulla et al., 2014; Legleye, Beck, Peretti-
Watel, Chau, & Firdion, 2010). In Italy, suicide is the
third major cause of death in the adult population, and
rates tend to rise with age (Pompili et al., 2010). Recent
epidemiological studies indicate a death rate of 6.1/
100,000 inhabitants among individuals aged 25–44, 8.4
among those aged 45–64, and 11.3 among those aged >65
(Istituto Nazionale di Statistica, 2011).
Although fatal attempts in the elderly are more fre-
quent than in younger adults, there is evidence that the
elderly are less likely to present long-term elaboration of
suicidal thoughts and to communicate their intent and ide-
ation (Conwell et al., 1998). Consequently, risk detection
and timely intervention are particularly salient. Howeve r,
the progressive increase in average life expectancy and
the aging of the population exhort researchers to investi-
gate and assess risk factor s for suicidal ideation (SI) in the
elderly.
Depression and hopelessness have been most consis-
tently indicated as major predictors of SI in the elderly
(Pompili et al., 2008 ; Szanto, Prigerson, & Reynolds III,
2001) as well as in younger adults (Haaga et al., 2002;
Vrshek-Schallhorn, Czarlinski, Mineka, Zinbarg, &
Craske, 2011). However, different life stressors mark the
emergence of such conditions in the two groups, i.e.
mainly retirement, social isolation, and loss of a partner
in the elderly (Juurlink, Herrmann, Szalai, Kopp, &
Redelmeier, 2004), and negative childhood experiences,
poverty, ill-treatment in youth, and drug abuse (Pompili
et al., 2011) in young adults. In particular, hopelessness
has been reported as the most common emotion experi-
enced among suicidal individuals (Shneidman, 1996), and
research supports a positive relation between scores at the
Beck Hopelessness Scale (BHS) (Beck, Weissman,
Lester, & Trexler, 1974) and measures of depression,
suicidal intent, and ideation in clinical and nonclinical
populations.
Furthermore, to date, few studies have investigated
suicidality in the elderly in relation to specific personality
traits. Lynch et al. (1999) found in a sample of suicidal
old adults that chronic behavioral and personality prob-
lems were related to previous episodes and early onset of
depression. Useda, Duberstein, Conner, and Conwell
(2004) found associations between depression, hopeless-
ness, and SI in the elderly within the broader framework
of neuroticism and introversion, and Tsoh et al. (2005)
*Corresponding author. Email: emanuele.fino@uniroma1.it
Ó 2014 Taylor & Francis
Aging & Mental Health, 2014
Vol. 18, No. 6, 792–800, http://dx.doi.org/10.1080/13607863.2014.880404
reported a positive association between depression and
neuroticism, and a negative association between depres-
sion and extr aversion in old adults with previous suicide
attempt. In the same vein, Wiktorsson et al. (2013) found
that suicide attempters aged 75 and above scored higher
on neuroticism than comparisons, and lower on the extra-
version scale.
Research has recently addressed individual differences
in SI to internal working models (Davaji, Valizadeh, &
Nikamal, 2010; Sheftall, Mathias, Furr, & Dougherty,
2013). According to Bowlby’s (1969, 1973, 1980) theory
of personality development, early life experiences of
attachment impact later self–other perceptions, determin-
ing significant variations in relationships functioning
across life stages. Consistently, previous studies showed
that adults with less secure attachment styles are charac-
terized by less self-confidence and higher levels of nega-
tive effect, and are more lik ely to ideate and attempt
suicide (Fraley & Shaver, 2000).
In addition, although evidence on the role of personal-
ity dimensions in suicide in young adults is accumulating,
little is known about their relevance in the elderly. How-
ever, to date most of the research on SI has employed the
NEO Personality Inventory (NEO-PI; Costa & McCrae,
1992). The NEO-PI represents a lexical approach to the
assessment of personality. More recently, theorists have
focused on the association between personality traits and
relevant behavioral and biological characteristics (Block,
1995; Zuckerman, 1992), going beyond the descriptive
analysis of personality, toward a causal and psychobiolog-
ical appro ach. In particular, Zuckerman elaborated an
alternative five-factor model of personality, in which traits
present psychophysiological correlates and reliability
across cultures (Zuckerman, Kuhlman, Joireman, Teta, &
Kraft, 1993). Understanding the interplay between such
personality characteristics and psychopathology will aid
clinicians in the identification and assessment of the
elderly who are at risk of suicide.
Consequently, the purpose of the present study was to
analyze the components of a measurement model that are
invariant across the young adults and the elderly, and then
investigate the relations among dimensions of personality
and SI risk. We postulated a specific relationship pattern a
priori and then tested the hypotheses statistically in order
to examine the models for equivalency of the factorial
measurement. Specifically, nine scales representing
depression, hopelessness, attachment representations of
the self and the other, and the alternative five-factor model
of personality, as well as the underlying latent structure of
these observed variables characterizin g dimens ions of
personality and SI risk, were tested in young adults and
older adults.
Method
Participants
From November 2012 to May 2013, 655 participants aged
18–75 years were contacted at universities, markets,
supermarkets, shops, banks, public parks, post offices, and
senior centers in three Italian regions: Lo mbardia, Veneto,
and Lazio, and their respective districts. These three non-
randomly selected regions are highly representative of the
current demographic background of Italy. Lombardia and
Veneto are located in the north part of the country, with
approximately 9 million and 5 milli on residents, respec-
tively, while Lazio is located in the central part of the
country, with more than 5.5 million residents. All partici-
pants came from lower to upper middle class, with various
educational and socioeconomic backgrounds, representing
well enough the Italian population. The respon dents vol-
untarily participated in this study after providing written
informed consent. We created two 17-year age groups on
the basis of participa nts’ age: the Young Adults and the
Older Adults. The Young Adults (N ¼ 339) group ranged
from 18 to 35 years (mean ¼ 26.2; SD ¼ 3.8) and the
Older Adults group (N ¼ 316) ranged from 58 to 75 years
(mean ¼ 66.38; SD ¼ 3.9).
Measures
The Beck Depression Inventory Second Edition (BDI-II;
Beck, Steer, & Brown, 1996) is a 21-item self-report mea-
sure of symptoms of depression. The internal consistency
was assessed by means of Cronbach’s alpha (.92).
Respondents choose statements that reflect how they have
felt over the past 2 weeks. BDI-II scores range between 0
and 63; categorical depression ratings are ‘minimal’ (0–
13), ‘mild’ (14–19), ‘moderate’ (20–28), and ‘severe’
(29–63). The authors found, in their assessed clinical sam-
ple, a cut-off of 17 or greater with a 93% true-positive
rate and an 18% false-positive rate.
The BHS (Beck & Steer, 1987; Beck et al., 1974;
Pompili et al., 2009) is a 20-item true or false self-report
scale developed to operationalize the const ruct of hope-
lessness. Beck originally used this scale with adult psychi-
atric patients in order to predict who would commit
suicide and who would not. Responding to the 20 true or
false items on the BHS, individuals can either endorse a
pessimistic statement or deny an optimistic statement.
Research consistently supports a positive relation between
BHS scores and measures of depression, suicidal intent,
and current SI. Instead of the response format that
includes the true/false endorsement, in this case, to
increase the response sensitivity, we used a Likert-type
scale with 5-point format having two extreme options of
‘very strongly disagree’ (0) and ‘very strongly agree’ (4).
To obtain the measure of hopelessness, we reversed the
scoring of positive items, and then we summed the 20
items to yield a total score ranging from 0 to 80. In this
sample, the instrument showed a good reliability
(Cronbach’s alpha ¼ .83). This change is consistent with
previous research measuring hopelessness in the elderly
(Abraham, 1991; Neufeld, O’Rourke, & Donnelly, 2010).
Several authors have noted that although generally
accepted in clinical and research contexts, the original
dichotomous response format is likely to constrict
measurement variance and determine lower sensitivity
(Hayslip, Lopez, & Nation, 1991; Hill, Gallagher, Thompson,
& Ishida, 1988; Neufeld et al., 2010). In particular,
Aging & Mental Health 793
Neufeld et al. (2010, p. 752) asse ssed SI in a sample of
older adults, changing the response format of the BHS to
a Likert-type scale. The results of the study provided sup-
port to the revised response format of the BHS, showing
good psychometric properties and enhancing measure-
ment sensitivity of SI among older adults.
The 9 Attachment Profile (9AP; Candilera, 2007) is a
semi-projective test for assessing the quality of the inter-
personal relationships based on self–other perception and
internal working models of adult attachment. Bowlby’s
notion of attachment representation involves ideas regard-
ing both the self and others, whereas a person’s representa-
tion of the self and the other could be characterized by one
of the two orientations, i.e. positive or negative. This mea-
sure consists of seven basic pictures. Each picture repre-
sents a situation with one black figure and one or more
white figures in different environments. The presentation
of each picture is accompanied by two equal lists of
nine differential semantic scales anchored by opposed
terms. In the first list, participants are asked to rate their
self-perception on a 9-point scale for each differential
semantic, and in the second list their perception of the
others. 9AP provides 18 bipolar scales regarding psycho-
logical and emotional constructs, nine self-related and nine
other-related: acceptance–rejection, friendliness–hostility,
power–submission, security–insecurity, availability–
unavailability, calm–agitation, satisfaction–dissatisfaction,
independence–dependence, lack of competition–competi-
tion. Higher scores correspond to the first term of each
bipolar scale (positive representation), and lower scores
to the second term (negative representation). In summary,
we used the two general indicators, self-perception
(Cronbach’s alpha ¼ .91) and other-perception
(Cronbach’s alpha ¼ .89), scored by adding up the score of
the nine self-related and the nine other-related scales,
respectively.
The Zuckerman–Kuhlman–Aluja Personality Ques-
tionnaire (ZKA–PQ; Aluja, Kuhlman, & Zuckerman,
2010) is a 200-item questionnaire based on the theoretical
constructs of the alternative five-factor model of personal-
ity. The instrument measures aggressiveness (physica l
aggression, verbal aggression, anger, hostility), activity
(work compulsion, general activity, restlessness, work
energy), extraversion (positive emotions, social warmth,
exhibitionism, sociability), neuroticism (anxiety, depres-
sion, dependency, low self-esteem), and sensation seeking
(thrill and advent ure seeking, experience seeking, disinhi-
bition, boredom susceptibility/impulsivi ty). The authors
reported that alphas for aggressiveness, activity, extraver-
sion, neuroticism, and sensation seeking were .78–.81,
.76–.73, .75–.75, .74–.79, and .70–.72 for the Spanish and
American samples, respectively.
Statistical an alyses
We used two-tailed t-tests for continuous variables, and
chi-square tests with Yates’ correction where appropriate
for categorical variables. Factorial invariance and latent
mean structure were tested by structural equation model-
ing (SEM). SEM relies on several statistical tests to
determine the adequacy of model fit to the empirical data.
In SEM, it is possible to analyze relations between
observed variables and latent variables in addition to a
measurement model. The measurement model specifies
hypotheses about the relations between a set of observed
variables and the unobserved variables or constructs that
they were designed to measure. Confirmatory factor anal-
ysis (CFA) allows for a test of specific hypotheses con-
cerning the relation between observed variables and their
underlying latent constructs. On the basis of existing liter-
ature and consistent with theory, we anticipated several
relation patterns a priori and then tested the hypotheses
statistically. CFA seeks to determine if the number of fac-
tors and the loadings of measured (indicator) variables
conform to what is expected by the pre-established theory.
Our a-priori assumption was that each factor would be
associated with a specified subset of indicator variables.
The CFA implies the formal specification of the measure-
ment instrument in terms of a factor model, the statistical
fitting of the factor model to the observed data (variances
and covariances or correlations), the assessment of fit, and
the interpretation of the results if the model is consistent
with the data (Bollen, 1989; Byrne, 2010 ).
A series of multigroup CFA-nested models were con-
structed to examine the evidence of measurement invari-
ance (i.e. configural, metric, scalar, strict) and then the
latent mean structures (Vandenberg & Lance, 2000). In
testing for invariance, it is preferable first running a model
in which only the factor loadings are constrained equal
(i.e. a measurement model), and accordingly, provided
with evidence of group equivalence, these factor-loading
parameters remain constrained, and equality constraints
are then placed on the factor variances and covariances
(i.e. structural model) (Byrne, 2010).
Configural invariance requires that each common fac-
tor is associated with identical measurement sets across
groups, examining the strength of the relation between the
observed variables and their underlying latent constructs.
This model has no equality constraints imposed on the
estimated parameters, thus permitting different parameter
values across groups. This multigroup model serves two
important functions: it allows for invariance tests to be
conducted across the groups simultaneously, and in test-
ing for invariance, the fit of this configural model provides
the baseline value against which subse quent specified
invariance models can be compared.
Metric invariance is tested by imposing equality con-
straints on corresponding factor loadings and com paring
the fit of the constrained model to the configural model.
Metric invariance suggests that the observed variables
have identical meanings across groups. Scalar invariance
requires that the intercepts of the observed variables are
the same across groups and is tested by imposing equality
constraints on the intercepts and assessing model fit in
comparison to the metric invariant model. Strict invari-
ance assesses whether the data support equality of varia-
bles and factor residual variances across groups.
We used the following criteria to evaluate the overall
goodness of fit. The x
2
value close to 0 indicates little dif-
ference between the expected and observed covariance
794 P. Iliceto et al.
matrices, with the probability leve l greater than .05,
evidencing the absence of meaningful unexplained vari-
ance. Moreover, to estimate a better goodness of fit, due
to the fact that x
2
is sensitive to sample size, we calculated
the ratio of x
2
to degrees of freedom that should be less
than 3 as acceptable data-model fit. In addition to the
x
2
/df test, we utilized the comparative fit index (CFI;
Bentler, 1990), the Tucker–Lewis index (TLI; Tucker &
Lewis, 1973), the root mean square error of approximation
(RMSEA; Steiger, 1990) and the standardized root mean
square residual (SRMR; J
€
oreskog & S
€
orbom, 1996). Indi-
cators of a well-fitting model are evidenced by CFI and
TLI greater than .95, RMS EA less than .06, and SRMR
less than .08 (Browne & Cudeck, 1989; Hu & Bentler,
1998, 1999). We compared nested models using the x
2
difference test, and the change in CFI. A critical ratio
(CR), as z statistic, equal or greater than 1.96 indicates a
difference betwee n latent means (Cheung, 2008; Cheung
& Rensvold, 2002).
We conducted multigroup CFA-nested models to
examine whether or not the components of the measure-
ment model and the underlying theoretical struct ure were
invariant across the two groups of interest (i.e. the Young
Adults and the Older Adults) to test the hypothesis if the
loadings of the observed variabl es on the factors conform
to what would be expected on the basis of pre-established
theory (Byrne, Shavelson, & Muth
en, 1989). Specifically,
we tested for equivalency of the factorial measurement,
the scales representing the observed variables, and the
underlying latent structure as well as the relations among
dimensions of personality and SI risk across the Young
Adults and the Older Adults.
To test for factorial equivalence, given that the estima-
tion of baseline models involves no between-group con-
straints, the data can be analyzed separately for each
group. Then we used the nine scales to measure the under-
lying constructs of personality and suicidal ideation,
which provided the basis for the hypothesized model in
the determination of the baseline model for each group
separately. If this model fits the data well for both the
groups, it will remain the hypothesized model under the
test for equivalence across the two groups.
We examined the configural invariance to investigate
multigroup representation of the baseline models with
freely estimated factor loadings for each of the groups
simultaneously. This configural model provides the base-
line value against which all subsequent specified invari-
ance models were compared. Provided the evidence of
invariance between the two groups, we estimated latent
mean differences, that is unobserved means derived from
the observed variable means loading on the factor. We
chose the Young Adults as reference group, and fixed to
zero the means of the latent factors, and the Older Adults
as comparison group, and let the means of the latent fac-
tors vary freely.
All analyses were carried out using SPSS 17.0 (SPSS
Inc., Chicago, IL, USA). CFA was applied with the use of
AMOS 16.0 (AMOS: analysis of moment structures) and
maximum likelihood estimation (Arbuckle, 2007).
Results
As expected, there was a significant difference between the
age of the two groups (t
(653)
¼ 133.01; p < .001). As
regards gender, the Young Adults comprised 179 (52.8%)
(mean ¼ 25.8; SD ¼ 3.7) males and 160 (47.2%) females
(mean ¼ 26.2; SD ¼ 3.9). No difference was found
between the age of the participants in this group (t
(337)
¼
.81; p ¼ .41). The Older Adults comprised 153 (48.4%)
(mean ¼ 66.1; SD ¼ 3.7) males and 163 (51.6%) females
(mean ¼ 66.6; SD ¼ 4.1), and no difference was found
between the age of the participants in this group (t
(314)
¼
1.04; p ¼ .29). No gender difference was found in the two
groups (x
2
(1)
¼ 1.25; p ¼ .26), while significant differences
were found in the years of education (x
2
(2)
¼ 48.6; p
.001), as among the Older Adults there were more individ-
uals with less years of education (23.7%), and less individ-
uals with more years of education (25.9%). On the basis of
the working status, participants were divided into two
groups: Unemployed and Employed. No differences were
found in the working status between the Young Adults and
the Older Adults (x
2
(1)
¼ .05; p ¼ .80). The sociodemo-
graphic characteristics are summarized in Table 1.
Baseline models
When we tested a model with all nine observed variables,
namely depression, hopelessness, and neuroticism as indi-
cators of suicidal ideation, and self–other perception,
aggressiveness, activity, extraversion, and sensation
Table 1. Sociodemographic characteristics of subjects.
Characteristics Young Adults (N ¼ 339) Older Adults (N ¼ 316) Statistics p
Age (years) 26.02 3.8
a
66.38 3.9
a
t
(653)
¼ 133.01 <.001
Sex x
2
(1)
¼ 1.25 .26
Males (%) 52.8 48.4
Females (%) 47.2 51.6
Education x
2
(2)
¼ 48.6 <.001
8 (%) 5.0 23.7
13 (%) 58.4 50.3
>13 (%) 36.6 25.9
Working status x
2
(1)
¼ .05 .80
Employed% 35.4 34.5
Unemployed% 64.6 65.5
a
Values shown as mean SD.
Aging & Mental Health 795