Factorial structure of the 20-item Toronto Alexithymia Scale: confirmatory factorial analyses in nonclinical and clinical samples

https://doi.org/10.1016/S0022-3999(01)00197-0Get rights and content

Abstract

Objective: The 20-item Toronto Alexithymia Scale (TAS-20) measures three intercorrelated dimensions of alexithymia: (1) difficulties identifying feelings (DIF), (2) difficulties describing feelings (DDF), and (3) externally oriented thinking (EOT). The aim of the study was to test the three-factor model of the TAS-20 using confirmatory factorial analyses (CFA). Method: 769 healthy subjects and 659 patients meeting the DSM-IV criteria for substance use disorders or eating disorders completed the TAS-20. The correlation matrices for each of the samples were analyzed with LISREL 7.16. Results: In each sample, the three-factor model was found to be replicable. Conclusion: The three TAS-20 subcales can be used to explore the distinct facets of the alexithymia construct.

Introduction

Alexithymia refers to a specific disturbance in psychic functioning characterized by difficulties in the capacity to verbalize affect and to elaborate fantasies. The alexithymia construct, formulated from clinical observations, is multidimensional and includes four distinct characteristics: (a) difficulty identifying and describing feelings, (b) difficulty distinguishing feelings from the bodily sensations, (c) diminution of fantasy, (d) concrete and poorly introspective thinking (externally oriented thinking, EOT). Alexithymia has been reported in various psychiatric disorders: psychoactive substance dependence, posttraumatic stress, eating and panic disorders [1]. Moreover, alexithymia is though to be one of several possible risk factors for a variety of medical and psychiatric disorders.

For several years, the 26-item Toronto Alexithymia Scale (TAS, 2) has been the only self-report measure of the alexithymia construct suitable for research and clinical practice. The TAS demonstrated a replicable four-factor structure that was congruent with the theoretical construct of alexithymia: Factor 1 (difficulty identifying and distinguishing between feelings and bodily sensations), Factor 2 (difficulty describing feelings), Factor 3 (reduced daydreaming), and Factor 4 (externally orientated thinking).

Although the psychometric properties of the TAS were an improvement over those of other measures of the alexithymia construct, at least two problems with the TAS have been identified. First, the factor assessing reduced daydreaming (Factor 3) correlated negatively with the factor assessing the ability to identify and distinguish between feelings and bodily sensations (Factor 1) and the items of the reduced daydreaming factor had low magnitude item–total correlation with the full TAS [1]. The authors concluded that the items assessing daydreaming have little theoretical coherency with the other dimensions of the alexithymia construct.

Secondly, Factor 1 assessing the difficulty to identify and distinguish between feelings and bodily sensations and Factor 2 assessing the difficulty to describe feelings were highly correlated and had several items that loaded significantly on both factors, thus questioning the independence of these two factors [3]. Indeed, two-factor analytic studies made with different samples (substance abusers, [4]; healthy and psychiatric subjects, [5]) yielded a three-factor solution for the TAS; the ability to describe feelings did not emerge as a separate factor.

Taken into account these problems, the authors have developed two revised versions [3], [6] to improve the psychometric properties of the original TAS.

The first one was a 23-item scale (TAS — revised, TAS-R), which eliminated all items assessing imaginal activity [3]. In contrast to the four-factor structure of the TAS, exploratory factorial analyses (EFA) of the TAS-R yielded a two-factor solution. The first factor comprised items assessing both the ability to identify and distinguish between feelings and bodily sensations associated with emotional arousal and the ability to describe feelings to others. The second factor comprised items assessing EOT. The two-factor solution was found in three different groups of subjects: 965 university students, 401 college students, and 218 general psychiatric oupatients.

The second revised version, the 20-item Toronto Alexithymia Scale (TAS-20), appears to be an improvement over it predecessors [6], [7]. The TAS-20 corresponds to TAS-R reduced by three items (one rating EOT and two rating ability to distinguish or describe feelings). Contrary to the TAS-R, the factor structure of the TAS-20 using confirmatory factorial analyses (CFA) indicated that a three-factor solution was a good representation of the data, the three factors being difficulties to identify feelings (DIF), difficulty to describe feelings (DDF), and EOT.

For the authors, the TAS-20 item content measures all clinical features of alexithymia. The authors argued that an external cognitive style reflects, in part, an impoverished inner fantasy life [7]. Moreover, the three-factor structure of the TAS-20 is congruent with alexithymia construct. This three-factor structure of the TAS-20 was demonstrated with clinical and nonclinical populations by the use of CFA.

Several CFA have been done on the correlation matrix of the TAS-20. All have used the LISREL 7.16 method and several criteria of goodness-of-fit. Among the studies, two have used the original version of the scale. These studies have found a three-factor structure [6], [7]. The Parker et al. study [7] was done on three groups of undergraduate students (306 students at a university of Germany, 292 American students, 405 Canadian students). The three-factor model was superior to a unidimensional model in all three samples. The Bagby et al. study [6] used separate undergraduate university student (n=401) and psychiatric outpatient samples (n=218). The three-factor model was superior to a unidimensional or a bidimensional (DIF and DDF) model in all two samples. Other studies have replicated the three-factor structure using translation of the TAS-20 into European and Asian languages. For example, the latest published study was done using a Swedish translation of the TAS-20 [8].

There are several studies which have failed to replicate the three-factor solution.

One study [9], using the English version of the TAS-20, was done on 219 medical students and 204 psychoactive substance dependent-abusing inpatients. The CFA showed that the three-factor solution provided poor fits to the data in both samples. Recently one study, using EFA, has shown in 277 medical students a two-factor solution where the items rating the difficulties to identify and to describe feelings constituted a single factor [10].

The main problem represented by the two- or three-factor solution of the TAS-20 is to know if in the alexithymia construct the DIF and the DDF to other people constitute one or two separate dimensions.

To elucidate this problem, we have explored the factorial structure of the TAS-20 using the French version of the scale. An EFA was done on 263 French university students and a two-factor model was found where the DIF and DDF items constituted a sole factor [11]. Moreover, a CFA of the same data [1] concluded that the data fitted well to the three-factor model. Further, this three-factor solution was superior to the two-factor solution (DIF and DDF).

It is important to note that the studies reporting the two-factor solution of the TAS-20 used only EFA whereas the three-factor solution were retained using only CFA. Moreover, the studies reporting the two-factor solution of the TAS-R, which corresponds to the TAS-20 increased by three items, used also only EFA. This important point is confirmed by our studies [1], [11] on the same sample that reported either a two-factor solution of the TAS-20 or a three-factor solution according to the respective use of EFA and CFA.

The discrepancy between the results of the EFA and the CFA could be explained by the limitations of the EFA. EFA lacks an empirical method for determination of how well a model fits the data or means for comparison of alternative models.

There are fundamental postulates of EFA. The indeterminacy related to the various number of factors is resolved by adopting the postulate of parsimony. When several factor models are consistent with the observed data, we accept on faith the more parsimonious model. It is important to note that such an assumption is not provable but is widely accepted in others field of research.

CFA is used as a means of testing specific hypotheses. For example, when several factor models are consistent with the observed data, CFA allows to test the hypothesis that one model is more superior than the others.

Moreover, the stability of the factorial structure must be replicated in various samples of subjects.

The present study has three aims. The first is to explore the factorial structure of the French version of the TAS-20 in large clinical and nonclinical groups by using CFA in order to help clarify the discrepancy between the previous studies. The second is to compare the two- and three-factor solutions in each sample. The third is to explore the reliability and the validity of the scale with the best factor solution.

Section snippets

Subjects

The data were generated from subjects who took part in a large multicenter collaborative study on addictive behaviors. A total of 659 inpatients or outpatients (270 males, 389 females) had to meet DSM-IV criteria for substance use disorders (n=372) or eating disorders (n=287). Patients ranged in age from 14 to 50 (mean=27.18, S.D.=8.64). Diagnoses were established by clinical interview.

Seven hundred sixty-nine healthy subjects (309 males, 460 females) were recruited by announcements and did not

Results

The parameter estimates for each of the items for both samples are presented in Table 1. For the three-factor model and the nonclinical sample, the chi-square goodness-of-fit was significant (χ2=639.03, df=167, P<.001), a result that is typically found with large sample sizes [18]. However, the GFI (0.92), AGFI (0.90), and RMS (0.061) all met the criteria standards for adequacy of fit. The pattern of results was similar for the clinical sample (χ2=557.41, df=167, P<.001; GFI=0.92; AGFI=0.90;

Discussion

The present study found the previously established three-factor model for the TAS-20 to be replicable in nonclinical and clinical samples. Moreover, the three-factor model was superior to a two-factor model in the two samples.

In the present study, it is interesting to note that both the two- and three-factor models met the goodness-of-fit criteria. However, only the difference in the values of chi-squares for goodness-of-fit has allowed to conclude to the superiority of the three-factor

References (31)

  • LS Mann et al.

    Factor analysis of the Toronto Alexithymia Scale: elucidation of a polythetic construct

    Psychother Psychosom

    (1992)
  • JDA Parker et al.

    Factorial validity of the 20-item Toronto Alexithymia Scale

    Eur J Pers

    (1993)
  • M Simonsson-Sarnecki et al.

    A Swedish translation of the 20-item Toronto Alexithymia Scale: cross validation of the factor structure

    Scand J Psychol

    (2000)
  • M.G. Haviland et al.

    Structure of the twenty-item Toronto Alexithymia Scale

    J Pers Assess

    (1996)
  • T Erni et al.

    Two-factor solution of the 20-item Toronto Alexithymia Scale confirmed

    Psychopathology

    (1997)
  • Cited by (0)

    View full text