The 20-Item Toronto Alexithymia Scale: III. Reliability and factorial validity in a community population

https://doi.org/10.1016/S0022-3999(02)00578-0Get rights and content

Abstract

Objective: Some researchers have questioned the stability of the three-factor structure of the 20-Item Toronto Alexithymia Scale (TAS-20) or the reliability of one or more factors of the scale. The aim of this study was to assess the replicability of the factor structure of the TAS-20 in a large community sample and to determine also whether the same three-factor structure can be recovered in men and women. The study also assessed the reliability of the scale and the influence of gender, age, and education on TAS-20 scores. Method: The TAS-20 was administered to 1933 adults (880 men and 1053 women) residing in several small cities and towns in Ontario, Canada. The factor structure of the scale was assessed using confirmatory factor analysis (CFA). Results: The three-factor structure of the TAS-20 was replicable in the entire community sample and also separately in men and women. The TAS-20 and its three factors demonstrated internal reliability, and the variables of gender, age, and education accounted for relatively small or modest amounts of variability in total TAS-20 and factor scale scores. Conclusion: The results provide strong support for the reliability and factorial validity of the TAS-20 and indicate the importance of using CFA when assessing the replicability and theoretical integrity of the factor structure of the scale.

Introduction

The 20-Item Toronto Alexithymia Scale (TAS-20) was developed a decade ago [1], [2], [3] and has since become the most widely used instrument for assessing alexithymia in both research and clinical practice [4], [5], [6]. Over the years, there has been an accumulation of evidence that the scale is reliable and valid [7]. Some researchers, however, have questioned the stability of the factor structure [8], [9] or the reliability of one or more factors of the TAS-20 [10], [11], [12]. Others have expressed doubts that a self-report scale can adequately assess a construct that involves impairments in self-awareness [11], [13], [14], [15]. There is uncertainty also over the extent to which TAS-20 scores are influenced by gender, age, and education [2], [12], [14], [16], [17]. And there are questions as to whether the TAS-20 and the construct of alexithymia itself are valid across languages and cultures [18], [19], [20]. The purpose of this paper and a companion paper is to respond to some of these concerns. In the present paper, we report a study, which assesses the reliability and factorial validity of the TAS-20 in a large Canadian community population. In the companion paper [21], we review findings from a large number of cross-cultural studies that have evaluated different translations of the scale.

It is important to emphasize that the TAS-20, in contrast to previous measures of alexithymia, was developed using a combined empirical and rational method of scale construction with the items written to reflect the substantive domains of the alexithymia construct as defined originally by Nemiah et al. [22]. The TAS-20 yields three factors (difficulty identifying feelings, difficulty describing feelings, and externally oriented thinking), which are congruent with the salient facets of the construct [2]. Items assessing fantasy and imaginal activity, which are reduced in alexithymia, were eliminated during the development of the scale primarily because they had high correlations with measures of social desirability. There is evidence to suggest that reduced fantasy and imaginal activity are assessed indirectly by the externally oriented thinking factor, which correlates negatively with a measure of fantasy and imaginal activity [3]. The linking of reduced fantasy activity with externally oriented thinking is consistent with Marty and de M'Uzan's [23] concept of pensée opératoire from which the alexithymia construct is partly derived. Other more recently developed measures of alexithymia deviate from Nemiah et al.'s [22] definition of the construct; for example, the self-report Bermond-Vorst Alexithymia Questionnaire [24] includes a factor that assesses “emotionalizing” (which is defined as “the degree to which someone is emotionally aroused by emotion inducing events”) and the Observer Alexithymia Scale [25] includes factors that assess somatization, humorlessness, and rigidity. In our view, these additional characteristics should be considered correlates or physical and psychological sequelae of alexithymia rather than core features of the construct.

The three-factor structure of the TAS-20 has been replicated in both university student and general psychiatric outpatient populations in North America by the method of confirmatory factor analysis (CFA) [2], [26]. Some investigators have argued, however, that the three-factor structure is unstable or that the items are better represented by a two-factor solution. Loas et al. [27], for example, conducted principal components analysis on data collected from students at a French university and obtained a two-factor solution; the items assessing difficulty identifying feelings and difficulty describing feelings constituted a single factor and the items assessing externally oriented thinking comprised a second factor. Erni et al. [8] also obtained a two-factor solution when they used principal components analysis to determine the factor structure of a German translation of the TAS-20 with data collected from a group of Swiss medical students.

It is important to recognize that principal components analysis is an exploratory approach which is appropriate when a researcher is conducting initial forays into an area about which little is known [28]. CFA, on the other hand, is a hypothesis testing procedure that enables researchers to evaluate a priori theoretical models that form the basis of the factor structure of a test. It can also be used to compare the equivalence of factor structures in different samples [28], [29]. As such, CFA is more appropriate than an exploratory factor analysis (EFA) for assessing the replicability of the three-factor model of the TAS-20 across different samples and cultures. When CFA was applied to the French data collected by Loas et al. [27], the original three-factor structure of the scale was found to provide a better fit than a two-factor solution [30]. And although Erni et al. [8] did not subsequently report a CFA of their data collected from German-speaking Swiss medical students, Parker et al. [26] had previously replicated the three-factor structure of the scale by this method in a comparable sample of German university students.

In a more recent study, Kojima et al. [12] conducted a principal components analysis on data collected from a large sample of postmyocardial infarction patients of whom 95% were French Canadian. Although a three-factor solution was obtained, the item loadings did not fully correspond with those obtained in the original development of the scale. In contrast, in another recent study that was conducted in France, Loas et al. [31] used CFA and were able to replicate the original three-factor structure of the TAS-20 in both a large sample of normal adults and a large sample of psychiatric patients with eating disorders or substance use disorders.

In other research, Haviland and Reise [32] conducted confirmatory factor analyses on data sets from medical students and psychoactive substance-dependent inpatients in the United States, and reported that the three-factor solution provided a poor fit to the data in both samples. Examination of the results for the medical student sample, however, reveals that one of the goodness-of-fit indices (GFIs) met its criterion standard and two other indices were just below their criteria standards. Moreover, several other confirmatory factor analytic studies with US and Canadian student samples have yielded results that met the criterion standards for the same three GFIs [20], [26]. In addition, the substance-dependent patients constituted an unusual sample as they were recently abstinent from alcohol or psychoactive drugs and completed the TAS-20 within their first week of hospitalization [33]; as such, the results of the CFA have little generalizability.

Some researchers have expressed concern about the internal consistency of one or more subscales derived from the TAS-20 factors [10], [11], [12], [31], [34], especially since estimates of coefficient α in some studies have been below the generally recommended standard of >.70 [35]. In a study with US Air Force recruits, for example, Davies et al. [10] reported an α coefficient of .81 for Factor 1, but extremely low α coefficients for Factors 2 (.365) and 3 (.007). It was subsequently discovered, however, that these researchers failed to reverse the scoring for the negatively keyed items on the scale; once the correct scoring was used, the α coefficients increased to .76 for Factor 2 and .46 for Factor 3 [L. Stankov, personal communication, November 16, 1999].1 In studies with other North American populations, α coefficients range between .73 and .84 for the full TAS-20 and for the first two factors, and between .62 and .71 for the third factor [2], [26], [36]. Moreover, the mean interitem correlations for the three factors typically range between .20 and .40 [2], [36], which is the range recommended by Briggs and Cheek [37] for multifactor scales and indicates adequate homogeneity of the items for each factor.

Regretably, on the basis of the factor analytic studies by Haviland and Reise [32] and Loas et al. [27], or misleading information given by Davies et al. [10], some researchers (e.g., Jacob and Hautekeete [38]) have decided against using the TAS-20 and overlooked the substantial body of support for the reliability and validity of the scale. As Nunnally [35] points out, however, the results of factor analysis of a scale can be influenced by the type of subjects selected. Although the factor structure of translated versions of the TAS-20 has been cross-validated in normal adult samples in several different cultures [21], the replicability of the three-factor model for the original English version of the scale has been demonstrated only in relatively small samples of university students and psychiatric outpatients [2], [26]. In addition, some investigators have found TAS-20 scores to be associated with male gender and higher age [14], [17].

The aim of the present study was to assess the factor structure of the TAS-20 in a large adult community sample, and to determine also whether the same factor structure can be recovered in both men and women. The internal reliability of the TAS-20 and its three factors, and the influence of gender, age, and education on TAS-20 scores, were also reassessed.

Section snippets

Subjects

The sample was comprised of 1933 adults (880 men and 1053 women) residing in several small cities and towns in Ontario, Canada. The participants were invited to take part in the study by means of advertisements posted in the local community. The mean age of the sample was 35.47 years (S.D.=12.55); the mean level of education was 14.75 years (S.D.=2.42). For ethnicity, 88.1% of the participants identified themselves as “White,” 3.0% as “Black,” 1.1% as “Asian,” 1.4% as “Native American,” and

Confirmatory factor analysis

The item to factor parameter estimates from the CFA with the three-factor model for the TAS-20 items are presented in Table 1; the GFIs are presented in Table 2. The GFI (.98), AGFI (.98), CFI (.97), RMSR (.05), and the RMSEA (.06), all met the criteria standards for adequacy of fit. All parameter estimates were significant at P<.05. The parameter estimate between F1 and F2 was .73 (P<.05), between F1 and F3 was .49 (P<.05), and between F2 and F3 was .63 (P<.05).

In evaluating whether the TAS-20

Discussion

The results of this study with a large English-speaking adult community sample provide strong support for the validity of the three-factor structure of the TAS-20. The three-factor model provided a better fit to the data than both a unidimensional model and a two-factor model. In addition, the parameter estimates for the relationships among the three factors provide evidence that the factors reflect three separate, yet empirically related, facets of the alexithymia construct. The three-factor

Acknowledgements

This study was supported by research grants to the first author from the Social Sciences and Humanities Research Council of Canada (SSHRC) and the Ontario Government's Premier's Research Excellence Award program. The third author was supported by operating grants from the Ontario Mental Health Foundation (OMHF) and SSHRC, as well as a Senior Research Fellowship Award from OMHF. The authors wish to thank Tonya Bauermann, Barbara Bond, Terri Collin, Marjorie Hogan, Patricia Kloosterman, Sarah

References (59)

  • MG Haviland et al.

    An observer scale to measure alexithymia

    Psychosomatics

    (2000)
  • G Loas et al.

    Factorial structure of the 20-Item Toronto Alexithymia Scale. Confirmatory factorial analyses in nonclinical and clinical samples

    J Psychosom Res

    (2001)
  • MG Haviland et al.

    Alexithymia in women and men hospitalized for psychoactive substance dependence

    Compr Psychiatry

    (1994)
  • M Bach et al.

    Independency of alexithymia and somatization

    Psychosomatics

    (1996)
  • KL Dion

    Ethnolinguistic correlates of alexithymia: toward a cultural perspective

    J Psychosom Res

    (1996)
  • S Jacob et al.

    Alexithymia is associated with low self-estimated affective intensity

    Pers Individ Differ

    (1999)
  • C Marchesi et al.

    Are alexithymia, depression, and anxiety distinct constructs in affective disorders

    J Psychosom Res

    (2000)
  • RM Bagby et al.

    Reliability and validity of the revised 20-Item Toronto Alexithymia Scale

  • GJ Taylor et al.

    Disorders of affect regulation: alexithymia in medical and psychiatric illness

    (1997)
  • GJ Taylor et al.

    Assessment of alexithymia: self-report and observer-rated scales

  • T Erni et al.

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

    Psychopathology

    (1997)
  • M Davies et al.

    Emotional intelligence: in search of an elusive construct

    J Pers Soc Psychol

    (1998)
  • J Ciarrochi et al.

    Measuring emotional intelligence

  • RD Lane et al.

    Impaired verbal and nonverbal emotion recognition in alexithymia

    Psychosom Med

    (1996)
  • JD Mayer et al.

    Selecting a measure of emotional intelligence: the case for ability scales

  • LJ Kirmayer

    Languages of suffering and healing: alexithymia as a social and cultural phenomenon

    Transcult Psychiatr Res Rev

    (1987)
  • R Prince

    Alexithymia and verbal psychotherapies in cultural context

    Transcult Psychiatr Res Rev

    (1987)
  • CG Loiselle et al.

    Cross-cultural validation of the Toronto Alexithymia Scale (TAS-20) in US and Peruvian populations

    Transcult Psychiatry

    (2001)
  • Taylor GJ, Bagby RM, Parker JDA. The Twenty-Item Toronto Alexithymia Scale: IV. Reliability and factorial validity in...
  • Cited by (535)

    View all citing articles on Scopus
    View full text