Filetype PDF | Posted on 06 Oct 2022 | 4 years ago
Authentication
digital resources
349x Filetype PDF File size 0.16 MB Source: gredos.usal.es
Braz J Psychiatry. 2019 Mar-Apr;41(2):122-130
doi:10.1590/1516-4446-2018-0006
Brazilian Psychiatric Association
00000000-0002-7316-1185
ORIGINAL ARTICLE
Psychometric properties of the Liebowitz Social Anxiety
Scale in a large cross-cultural Spanish and Portuguese
speaking sample
1 2 ´ 3 4 4
Vicente E. Caballo, Isabel C. Salazar, Vıctor Arias, Stefan G. Hofmann, Joshua Curtiss ; CISO-
AResearch Team
1 ´ ˜ 2 ´ ´ ˜ 3
Facultad de Psicologıa, Universidad de Granada, Granada, Espana. Centro de Psicologıa Clınica FUNVECA, Granada, Espana. Facultad
´ ˜ 4
dePsicologıa, Universidad de Salamanca, Salamanca, Espana. Department of Psychological and Brain Sciences, Boston University, Boston,
MA, USA. https://orcid.org/0000-0002-2767-8028
Objective: To examine the psychometric properties of the Liebowitz Social Anxiety Scale-Self Report
(LSAS-SR) based on a large sample recruited from 16 Latin American countries, Spain, and Portugal.
Methods: Two groups of participants were included: a non-clinical sample involving 31,243 com-
munity subjects and a clinical sample comprising 529 patients with a diagnosis of social anxiety
disorder (SAD). Exploratory factor analysis (EFA), confirmatory factor analysis (CFA) and exploratory
structural equation modeling (ESEM) were used in order to determine the psychometric properties of
the LSAS-SR.
Results: EFA identified five factors with eigenvalues greater than 1.00 explaining 50.78% of the cumula-
tive variance. CFA and ESEM supported this 5-factor structure of the LSAS-SR. The factors included:
1) speaking in public; 2) eating/drinking in front of other people; 3) assertive behaviors; 4) working/writing
while being observed; and 5) interactions with strangers. Other psychometric properties such as inter-
factor correlations, invariance, reliability, and validity of the scale were also found.
Conclusion:Psychometricdatasupporttheinternal consistency and convergent validity of the LSAS-
SR. It seems to be a valid and reliable measure of global social anxiety for Spanish and Portuguese-
speaking countries, although when considering a multidimensional approach (factor-based assessment)
it seems to be lacking some relevant social situations that are feared in those countries.
Keywords: Social anxiety disorder; questionnaires; psychometrics; cross-cultural comparison
Introduction be single, of a lower socioeconomic status, and under-
3
educated. The economic burden of this condition to both
5
Social anxiety disorder (SAD) or social phobia refers to society and the individual is also significant. Because
a ‘‘marked fear or anxiety about one or more social of its high prevalence and societal burden, SAD has
situations in which the individual is exposed to possible attracted increasing attention from researchers in recent
1
scrutiny by others.’’ SAD is among the most prevalent decades, leading to innovations in both the treatment
anxiety disorders, with 12-month prevalence rates of and assessment of this condition. Numerous measures
2
6.8%intheUnitedStatesofAmerica. Nevertheless, little have been developed to assess this condition, ranging from
is known about the prevalence of SAD in Spanish and semi-structured interviews to self-report inventories.
Portuguese-speaking countries. Extant epidemiological Oneof the most commonly used self-report SAD mea-
surveys suggest a 12-month SAD prevalence rate of 0.60% sures internationally is the Liebowitz Social Anxiety Scale
in Spain, 2.8% in Colombia, 2.60% among Mexican women, (LSAS).6 The original 24-item semi-structured interview
and 1.40% among Mexican men.2 The prevalence of this involves a two-factor model with separate subscales to
condition in Portugal and other Latin American countries assess fear and avoidance of situations involving social
remains unknown. interaction and performance/observation by others. How-
SADhasbeenassociatedwith substantial functional and ever, the two-factor model has been shown to provide
occupational impairment, a chronic, unremitting course, an an inadequate fit of the data, and a self-report version
elevated risk of comorbid depression,3 and lower social (LSAS-SR) divided into four subscales was proposed
4 7-9
skills. Specifically, individuals with SAD are more likely to instead. This four-dimensional factor structure has been
´
Correspondence: Vicente E. Caballo, Facultad de Psicologıa, Universi- Howtocite this article: Caballo VE, Salazar IC, Arias V, Hofmann
dad de Granada, Campus de la Cartuja, s/n, 18071 Granada, Spain. SG, Curtiss J; CISO-A Research Team. Psychometric properties of
E-mail: vcaballo@ugr.es the Liebowitz Social Anxiety Scale in a large cross-cultural Spanish
Submitted Aug 16 2017, accepted Jan 11 2018, Epub Oct 11 2018. andPortuguesespeakingsample. Braz J Psychiatry. 2019;41:122-130.
http://dx.doi.org/10.1590/1516-4446-2018-0006
Liebowitz Social Anxiety Scale 123
10,11
used in many studies. Nevertheless, different studies Rica, 1.44%Honduras,2.63%Bolivia,1.92%ElSalvador,
with various populations have detected a different number 0.63% Dominican Republic, and 1.22% Guatemala). The
12 8 13 14 15
of factors – three, four, five, six, or even eight. sample included 56.71% women (M = 25.10 years, SD =
These studies usually only factor-analyze the fear or 9.87) and 43.14% men (M = 26.00 years, SD = 10.43),
8,12,13
anxiety subscale, because factor analysis of the fear with 0.15% of participants not reporting their gender.
andavoidance subscales produces similar results.12,14 Thus Regarding education and type of occupation, 58.89% were
the fear or anxiety subscale is usually the only one analyzed. higher education students, 14.99% were workers with a
Regarding the psychometric properties of the LSAS- college diploma, 9.58% were secondary education students,
SR, the literature has reported good test-retest reliability, 8.69% were workers with no higher education, and 7.22%
adequate internal consistency, and adequate convergent did not match any of the former categories (e.g., housewife,
10,13
and discriminant validity, even in versions translated retired, or unemployed). No data on occupation were avai-
15-17
into other languages. The LSAS-SR has also been lable for 0.63% of the participants.
used to establish the convergent validity of other self- The second group of participants consisted of 529
report measures of social anxiety, such as the Social patients (M age = 31.73 years, SD = 11.96, range = 16-72)
Phobia Inventory (SPIN),18 the Social Phobia and Anxiety from 13 countries (164 Spain, 112 Mexico, 64 Brazil,
Inventory (SPAI),19 the Social Interaction Anxiety Scale 59Argentina, 44 Colombia, 31 Peru, 27 Chile, 11 Portugal,
20
(SIAS) and the Social Phobia Scale (SPS), and the 10Uruguay,threeVenezuela,twoBolivia,onePanama,and
short forms of the SIAS and the SPS.21 Furthermore, it is one Puerto Rico); there were 337 women (M = 32.61 years,
commonly used to assess treatment outcomes in SAD SD = 12.14) and 192 men (M = 30.18 years, SD = 11.49).
patients receiving pharmacological treatment6,22 or cog- For inclusion in this group, patients had to meet a pri-
23
nitive-behavioral therapy. mary diagnosis of SAD according to the criteria of the
Although the psychometric properties of the LSAS-SR 26 or ICD-10.27 Each center conducted its own
DSM-IV-TR
have been examined in different countries, few studies diagnostic assessment of individual patients. Patients with
have been conducted in Spanish or Portuguese-speaking aDSM-IV-TRorICD-10diagnosisofSADwereincludedin
17,19,24,25 The aim of this study was to examine
countries. the study even if they had other disorders in addition to
the factor structure, invariance, internal consistency, and SAD(Table 1).
convergent validity of the LSAS-SR with clinical and A second inclusion criterion was a score X 60 on the
non-clinical samples from Spain, Portugal, and 16 Latin 28 It should be noted that although Mennin et al.
LSAS-SR.
American countries. used a score of 60 as indicative of generalized social
anxiety disorder (GSAD) and of 30 as non-GSAD (NSAD),
Methods apreviousBrazilian study found scores between 52 and 81
for moderate phobia.17 Furthermore, it has been found that
Participants while 21% of a Spanish non-clinical sample scored higher
than 60 on the LSAS-SR, this percentage rose to 68% with
The first group of participants consisted of 31,243 non- acutoff score of 30.29 Given these results, a cutoff score of
clinical individuals (mean age [M] = 25.50 years, standard 60 was considered more appropriate than a score of 30 for
deviation [SD]= 10.13, range: 16-87 years) from 18 coun- the present sample.
tries (22.68% Mexico, 18.19% Colombia, 12.30% Spain, Patients were excludedfor several reasons (e.g., five or
9.82% Peru, 7.75% Brazil, 3.84% Argentina, 3.35% more unanswered items, presence of psychotic disorders,
Uruguay, 3.06% Venezuela, 1.75% Puerto Rico, 3.30% SAD not the primary diagnosis – which in fact was the
Portugal, 3.37% Chile, 1.91% Paraguay, 0.82% Costa main reason for exclusion). From a pool of 907 patients
Table 1 Distribution of patients by psychiatric disorder
n (%)
Psychiatric disorders Women Men Total
SAD 119 (22.49) 81 (15.31) 200 (37.81)
SAD+otheranxiety disorder 49 (9.26) 36 (6.80) 85 (16.07)
SAD+mooddisorder 104 (19.66) 37 (6.99) 141 (26.65)
SAD+otheranxiety disorder + mood disorder 6 (1.13) 6 (1.13) 12 (2.26)
SAD+avoidant personality disorder 1 (0.19) 4 (0.76) 5 (0.94)
SAD+personality disorder (except avoidant) 18 (3.40) 7 (1.32) 25 (4.72)
SAD+otheranxiety disorder + personality disorder (except avoidant) 4 (0.76) 1 (0.19) 5 (0.94)
SAD+eating disorder 17 (3.21) 0 (0) 17 (3.21)
SAD+eating disorder + personality disorder (except avoidant) 4 (0.76) 0 (0) 4 (0.76)
SAD+substance use disorder 0 (0) 8 (1.51) 8 (1.51)
SAD+mooddisorder + substance use disorder 3 (0.57) 3 (0.57) 6 (1.13)
SAD+oneotherdisorder (not included above) 8 (1.51) 7 (1.32) 15 (2.83)
SAD+twootherdisorders (not included above) 2 (0.38) 1 (0.19) 3 (0.57)
SAD+threeother disorders (not included above) 2 (0.38) 1 (0.19) 3 (0.57)
Total 337 (63.70) 192 (36.30) 529 (100)
Data presented as n (%).
SAD=social anxiety disorder.
Braz J Psychiatry. 2019;41(2)
124 VE Caballo et al.
diagnosed with SAD, 529 satisfied all the inclusion cri- administered to 529 patients with SAD and to 31,243 non-
teria. Regarding occupation, 25.14% were workers with clinical participants from the community. Given that we
a college diploma, 20.60% were workers with no college did not find significant differences among Spain, Portugal,
diploma, 17.58% were higher education students from andmostLatinAmericancountriesregardingassessment
different majors, 7.75% were secondary education students, of social anxiety,29,31 we grouped all these countries
1.89% were higher education psychology students, 0.76% together for analysis. For the assessment of the clinical
were psychologists, and 23.63% did not match any of the group, our collaborators administered the two question-
former categories (e.g., housewife, retired or unemployed). naires (LSAS-SR and SAQ) individually to patients. For the
No data on occupational status were obtained for the non-clinical sample, the questionnaires were administered
remaining 14 patients. to groups of subjects. Collaborators working in high schools,
colleges, or universities administered the questionnaires to
Measures people in classes and meetings of teachers or professors.
Those working in companies convened voluntary meetings
Liebowitz Social Anxiety Scale (LSAS)6 for workers. No compensation was provided to participants.
The two questionnaires were sent to each collaborator
The LSAS is a 24-item interviewer-rated instrument that and in order to derive the Portuguese version, both
assesses fear/anxiety and avoidance of specific social questionnaires were translated and back-translated from
situations. Each of the 24 items serves to assess both Portuguese to English (LSAS-SR) or Spanish (SAQ) until
variables. Respondents are asked to rate their fear/anxiety agreement was reached between translators. Both ques-
(LSAS-anxietysubscale)onafour-pointscalerangingfrom tionnaires were administered together, but the order of
0 (none) to 3 (severe), and avoidance (LSAS-avoidance administration was random.
subscale) on a four-point scale ranging from 0 (never) to 3
(usually). The total score is obtained by adding the scores Ethical considerations
obtained on both subscales. The LSAS has also been used
as a self-report instrument (LSAS-SR) in the literature with Participation in the study was voluntary and the ques-
10,13
these same characteristics. With regard to the psy- tionnaires were filled out anonymously. Informed verbal
chometric characteristics of the Spanish version, one study consent was obtained from all respondents, who were
25 and another study repor-
reported a four-factor structure free to withdraw at any time or to refuse to answer the
19
ted five-factors. A Portuguese version of the scale questionnaires. The study and all its procedures were
17
showedafive-factorstructure. Reportedreliability indices approved by Spain’s Ministry of Science and Technology.
are adequate.19,24,25 In this way, these last three studies This study does not break the agreements of the Helsinki
have found scores for internal consistency (Cronbach’s Declaration.
alpha) of 0.93, 0,83, and 0,87, respectively, for the LSAS-
anxiety subscale. Split-half reliability coefficient (Guttman) Data analysis
19 25
found has been 0.90, and test-retest reliability 0.89.
Tocross-validate the factor structure of the LSAS-SR, the
Social Anxiety Questionnaire for Adults (SAQ)29-31 total sample of 31,243 individuals was randomly split in
two halves (n =15,566; n =15,677). There were missing
The SAQ is a 30-item self-reported questionnaire which 1 2
data in 2.19% of the responses to the LSAS-SR. Given
was empirically developed in Spanish and Portuguese the low rate of missing data, and also the absence of evi-
speaking countries to assesses five dimensions of social dence incompatible with a missing completely at random
anxiety: 1) speaking in public/talking with people in authority; structure, pairwise deletion was used to handle the miss-
2) interactions with the opposite sex; 3) assertive expres- 33,34
ing data.
sion of annoyance, disgust, or displeasure; 4) criticism and First, we conducted an exploratory analysis with
embarrassment; and 5) interactions with strangers. Each parallel analysis and exploratory factor analysis (EFA) to
item is answered on a five-point Likert scale to indicate examine the internal structure of the LSAS-SR. Parallel
the level of unease, stress, or nervousness in response to analysis was implemented with LSAS-SR anxiety sub-
each social situation: 1 = not at all or very slight; 2 = slight; scaledatafromsubsample1usingtheMonteCarlopro-
3 = moderate; 4 = high; and 5 = very high or extremely cedure with 1,000 replications. Parallel analyses compared
high. Cronbach’s alpha for the total scale has been shown the eigenvalues extracted from the observed correla-
to range from 0.88 to 0.93,29-31 with split-half reliability
32 tion matrix to be analyzed with the eigenvalues obtained
coefficients (Guttman) ranging from 0.90 to 0.93. Regard- from uncorrelated normal variables (parallel components
ing the five dimensions, Cronbach’s alpha ranged from 0.74 derived from random data). EFA(unweighted leastsquares
to 0.90,29-31 with split-half reliability coefficients (Guttman) 35
32 with direct oblimin oblique rotation) wascomputedonthe
rangingfrom0.57to0.95. first subsample of non-clinical participants. We conduc-
ted separate EFAs on the fear/anxiety and the avoidance
Procedure items. Because we obtained similar results with both
subscales and because their distinctiveness has been
Our CISO-A Research Team consists of researchers questioned,8,19 we decided to continue the analysis with
and psychologists from Spain, Portugal, and most Latin only the fear/anxiety subscale. We also conducted an EFA
American countries. The LSAS-SR and the SAQ were with the clinical sample.
Braz J Psychiatry. 2019;41(2)
Liebowitz Social Anxiety Scale 125
Then, we tested the factor structure of the LSAS-SR country has negligible influence on the measurement model
anxiety subscale unveiled by EFA results using confirma- would not be rejected. Conversely, if the M12 fit is sub-
tory factor analysis (CFA)36 and exploratory structural stantially better than that of M10 and M11, possibly certain
equation modeling (ESEM).37 ESEM models have been items would be at risk of differential functioning.
recently developed to address a common limitation of Finally, we analyzed the mean differences between
CFAmodels, which often produce overly restrictive mea- clinical and non-clinical samples in the factors and sub-
surement models that do not provide acceptable good- scales of the LSAS-SR, reporting their effect sizes
ness of fit for most psychological instruments.38 The (Cohen’s d), as well as the differences in latent means
ESEM model is a special case of CFA in which the expressed in SDs of the clinical group from the general
assumption that the cross-loadings are 0 is relaxed, so group, as estimated from the scalar invariance model. These
that both models can be considered nested and their fit 36
differences are directly interpretable as a Cohen’s d.
37 The CFA and ESEM were completed with All statistical analyses were performed using Statistica
comparable.
41 42 43
the second non-clinical subsample of participants (n = version 12, SPSSversion 22, and MPlus version 7.4.
2
15,677) using weighted least squares with adjusted
means and variances (WLSMV) estimation. Four models Results
weretested: M1 = unifactorial model; M2 = two correlated
factors model (social interaction and performance situa- Exploratory and CFA of the LSAS-SR
tions of the original model); M3 = four-correlated factors
model (social interaction, public speaking, observation by The results of parallel analyses of subsample 1 with the
others, and eating and drinking in public of the Safren’s LSAS-SR anxiety subscale showed that the five-factor
model); M4 = five-correlated factors model; and M5 = solution was the best fit for the data, given that only the
ESEM. These same five models were also examined in eigenvalues of these five factors were greater than the
the clinical sample. randomly generated eigenvalues.
To appraise overall model fit, a number of fit indices Furthermore, the EFA with the LSAS-SR anxiety sub-
were examined, including the root mean square error of scale identified five factors with eigenvalues greater than
approximation (RMSEA), comparative fit index (CFI), and 1.00, explaining 50.78% of the cumulative variance. The
Tucker-Lewis index (TLI). Values of the RMSEA exceed- five factors were the following: 1) speaking in public (six
ing 0.07 indicate poor fit, whereas values X 0.90 would items); 2) eating/drinking in front of other people (four
indicate acceptable fit for the CFI and TLI. items); 3) assertive behaviors (four items); 4) working/
Subsequently, we estimated the correlations among the writing while being observed (two items); and 5) interac-
factors of the LSAS-SR, internal consistency (Cronbach’s tions with strangers (five items) (Table 2). All items loaded
alpha), and the convergent validity of the instrument with above 0.40 on only one factor except for items 7 and 13,
another self-report measure of social anxiety (i.e., the SAQ) which loaded above 0.40 on two factors. Thus, items
for the non-clinical and clinical samples separately. 7 and 13 were assigned to factors 2 and 3, respectively,
In the case of invariance by country, and given the because those are the individual factors with which these
large number of groups involved, Multiple Indicators items were most strongly associated. Items 1, 14, and
39 were used, according
Multiple Causes models (MIMIC) 18 did not load above 0.40 on any factor. In order to
40 In a MIMIC
to the procedure described in a former study. empirically test the redundant contribution of the avoid-
model, the grouping variable (previously dummy-coded) ancesubscale,anEFAwasalsoperformed.Weidentified
acts as a predictor of either the latent or the observed the same five factors explaining 48.58% of the cumulative
variables. Three nested MIMIC models were estimated: variance. Eighteen items from the EFA of this subscale
1) A null model (M10) in which the paths between the loaded above 0.40 on the same factors as the anxiety
grouping variable and the other terms of the model were subscale.Item7loadedstronglyontwofactorsanditem1
set to 0 (zero); M10 reflects the null hypothesis that the did not load above 0.40 on any factor.
country of origin has no effect either on thresholds (i.e., Thebestfactorsolutionfortheclinicalsamplebasedon
scalar invariance holds) or on latent variables (i.e., there the scree test was again a five-factor structure with eigen-
is no country-dependent differences in LSAS-SR scores). values greater than 1.00, explaining 47.23% of the cumu-
2) An invariant model (M11) in which the paths leading from lative variance. The first factor, interactions with strangers
the grouping variable to the factors were freely estimated, (eigenvalue: 5.91, items: 1, 10, 11, 12), explained 24.61%
setting the paths to the observable variables at 0 (zero). of the variance. The second factor, speaking in public
3) A saturated model (M12) in which the paths leading to (eigenvalue: 1.84, items: 5, 6, 15, 16, 20), explained 7.68%
factors were set to 0 (zero), but the paths from the of the total variance. Factor 3, working/writing while being
grouping variable to all observable indicators were freely observed (eigenvalue: 1.52, items: 8, 9), explained 6.33%
estimated (i.e., hypothesizing the non-invariance of of the variance. Factor 4, assertive behaviors (eigenvalue:
measurement relative to the country of origin). 1.41, items: 13, 17, 18, 22, 24), explained 5.89% of the
variance. Finally, factor 5, eating/drinking in front of other
Once the three models were estimated, M10 was people (eigenvalue: 1.32, items: 2, 3, 4, 7), explained
compared with M11 and M12. Since M11 and M12 are 5.51%ofthe variance. All items loaded above 0.40 on only
more parameterized than M10, they should tend to a onefactor. Items 14, 19, and 21 did not load above 0.40 on
better fit. However, if the M10 fit is not substantially any factor. Seventeen of the 24 items loaded on the same
worse than that of the other models, the hypothesis that the factors as observed for non-clinical subsample 1.
Braz J Psychiatry. 2019;41(2)
no reviews yet
Please Login to review.
