What do alexithymia items measure? A discriminant content validity study of the Toronto-alexithymia-scale–20

Introduction

The alexithymia construct has been introduced in the early seventies by Sifneos (1972, 1973) to describe clinical observations of patients with classic psychosomatic diseases who had difficulty engaging in insight-oriented psychotherapy (e.g., MacLean, 1949; Marty & de M’Uzan, 1963; Ruesch, 1948; Sifneos, 1967). Since then alexithymia, defined as the inability to recognize and express emotions (Taylor, Bagby & Parker, 2016), has been considered a key construct in many theoretical models of health psychology (Lumley, Neely & Burger, 2007). Contemporary theories describe alexithymia as a multidimensional construct with four interrelated features: “(1) difficulty identifying feelings and distinguishing between feelings and the bodily sensations of emotional arousal, (2) difficulty describing feelings to other people, (3) constricted imaginal processes, as evidenced by a paucity of fantasies, and (4) a stimulus-bound, externally oriented cognitive style” (Taylor, Bagby & Parker, 1997, p. 29; see also Sifneos, 1994). Over the last decades, alexithymia has been recognized as a risk factor for various psychiatric and medical conditions (Corcos & Speranza, 2003; Taylor, Bagby & Parker, 1997). In particular, it has been theorized that alexithymia reflects a deficit in the cognitive processing and regulation of emotions (Taylor, Bagby & Parker, 1997). This deficit would increase one’s vulnerability for psychiatric and medical conditions (Fernandez, Jazaieri & Gross, 2016). This idea is furthermore supported by abundant research showing that levels of alexithymia are increased in patients suffering from illnesses, such as eating disorders (e.g., Taylor et al., 1996), posttraumatic stress disorders (e.g., Frewen et al., 2006), chronic pain (e.g., Pecukonis, 2009), cancer (e.g., Todarello et al., 1989), and many more (Luminet, Bagby & Taylor, 2018; for a review, see Taylor & Bagby, 2000; Taylor, 2004).

The Toronto Alexithymia Scale–20 (TAS-20; Bagby, Parker & Taylor, 1994a; Bagby, Taylor & Parker, 1994b) is worldwide the most frequently used measure of alexithymia in both research and clinical practice (Lane et al., 2015; Sekely, Bagby & Porcelli, 2018). Although the TAS-20 is considered to be a well validated self-report measure of alexithymia (e.g., Bagby, Parker & Taylor, 2020), some concerns about its validity remain (Bermond, Oosterveld & Vorst, 2015; Lane et al., 2015; Lumley, Neely & Burger, 2007). First, doubts have been raised about whether the TAS-20 measures alexithymia in a comprehensive and relevant manner. The TAS-20 contains three subscales, i.e., ‘difficulty identifying feelings’, ‘difficulty describing feelings’, and ‘externally-oriented thinking’. The items for assessing the daydreaming factor in the earlier revision of the original TAS (TAS-R; Taylor, Ryan & Bagby, 1985; Taylor, Bagby & Parker, 1992), were eliminated because of either low item-total correlations or high correlations with a social desirability measure (Bagby, Parker & Taylor, 1994a; Bagby, Taylor & Parker, 1994b). Bagby, Parker & Taylor (1994a, 1994b, see also Taylor, Bagby & Parker (2016)) motivated their decision by arguing that this factor may be indirectly measured by the factor ‘externally-oriented thinking’ (Bagby, Parker & Taylor, 2020). Furthermore, confirmatory and exploratory factor analyses show that at least half of the externally-oriented thinking items load poorly on their intended factor (factor loadings < 0.40; e.g., Kooiman, Spinhoven & Trijsburg, 2002; Taylor, Bagby & Parker, 2003). If aspects of the construct alexithymia are underrepresented by the TAS-20 items and/or TAS-20 items are not relevant for the construct, it reflects a lack of content validity.

Second, there are doubts about whether the TAS-20 is sufficiently distinct from measures assessing related theoretical constructs. Some authors have argued that the TAS-20 is a measure of psychological distress rather than alexithymia (Leising, Grande & Faber, 2009). Indeed, significant and substantial correlations have been reported between the TAS-20 and measures of anxiety and depression in clinical samples (e.g., Marchesi et al., 2014) and in the general population (e.g., Honkalampi et al., 2010). Furthermore, Shahidi, Molaie & Dehghani (2012) found significant correlations between the TAS-20 scores and a measure of health anxiety. This study revealed that the ‘difficulty identifying feelings’ subscale predicted 52% of the total variance in health anxiety scores, and argued that this strong relationship is driven by particular items that measure difficulty in differentiating between bodily feelings and emotions (see also Barsky, 2001; De Gucht, Fischler & Heiser, 2004; Nakao et al., 2002). It is key that correlations between the TAS-20 and health anxiety are not (partially) explained by content overlap. If the TAS-20 is contaminated by content relevant to related constructs such as anxiety, depression, and health anxiety, relationships between the measures of these constructs may then simply be due to content overlap resulting in inflated explanatory power of alexithymia and hazardous theory building (Dixon & Johnston, 2019).

Despite these concerns, no study has examined the content validity and discriminant content validity of the TAS-20. In the current study, TAS-20 items are evaluated using the Discriminant Content Validity method (DCV; Johnston et al., 2014), a systematic and transparent way of investigating and reporting whether items are relevant for measuring target theoretical constructs (a key feature of content validity) and whether items are distinct from the content from other theoretical constructs (discriminant content validity). More specifically, we investigated to what extent items from the TAS-20 are (a) relevant for the construct ‘alexithymia’, and its key features, i.e., ‘difficulty identifying feelings’, ‘difficulty describing feelings’, ‘externally-oriented thinking’, and ‘limited imaginal capacity’ (content validity), and (b) distinct from related constructs, i.e., ‘anxiety’, ‘depression’, and ‘health anxiety’ (discriminant content validity).

Materials & methods

Results

Content validity of TAS-20 questionnaire and questionnaires of related constructs

TAS-20 questionnaire

The items of the TAS-20 questionnaire scored significantly higher on ‘alexithymia’ ( $\hat{\mu }$ = 1.12, 95% CI [−0.05 to 2.28]), compared to ‘anxiety’ ( $\hat{\mu }$ =−3.48, 95% CI [−4.65 to −2.31]; Δ = 4.60, 95% CI [4.16–5.05]), ‘depression’ ( $\hat{\mu }$ = −3.85, 95% CI [−5.03 to −2.68]; Δ = 4.98, 95% CI [4.53–5.42]), and ‘health anxiety’ ( $\hat{\mu }$ = −5.13, 95% CI [−6.31 to −3.97]; Δ = 6.25, 95% CI [5.80–6.69]). It should be noted though that the score for ‘alexithymia’ was not significantly different from zero. Findings are displayed in Fig. 1.

PROMIS-A questionnaire

The items of the PROMIS-A questionnaire scored significantly higher on ‘anxiety’ ( $\hat{\mu }$ = 7.93, 95% CI [6.37–9.28]), compared to ‘alexithymia’ ( $\hat{\mu }$ = −5.80, 95% CI [−7.39 to −4.45]; Δ = 13.74, 95% CI [12.85–14.62]), ‘depression’ ( $\hat{\mu }$ = 1.73, 95% CI [0.14–3.09]; Δ = 6.20, 95% CI [5.32–7.10]), and ‘health anxiety’ ( $\hat{\mu }$ = 3.86, 95% CI [2.26–5.21]; Δ = 4.08, 95% CI [3.19–4.97]; Fig. 1).

PROMIS-D questionnaire

Similar results were found for the PROMIS-D, showing that the items of the PROMIS-D scored significantly higher on ‘depression’ ( $\hat{\mu }$ = 8.40, 95% CI [6.31–9.77]), compared to ‘alexithymia’ ( $\hat{\mu }$ = −5.90, 95% CI [−8.00 to −4.46]; Δ = 14.30, 95% CI [13.45–15.16]), ‘anxiety’ ( $\hat{\mu }$ = 2.33, 95% CI [0.24–3.78]; Δ = 6.07, 95% CI [5.20–6.93]), and ‘health anxiety’ ( $\hat{\mu }$ = −2.15, 95% CI [−4.25 to −0.70]; Δ = 10.55, 95% CI [9.70–11.41]; Fig. 1).

SHAI questionnaire

The items of the SHAI questionnaire scored significantly higher on ‘health anxiety’ ( $\hat{\mu }$ = 8.95, 95% CI [7.98–9.81]), compared to ‘alexithymia’ ( $\hat{\mu }$ = −6.69, 95% CI [−7.66 to −5.78]; Δ = 15.64, 95% CI [14.84–16.42]), ‘anxiety’ ( $\hat{\mu }$ = 4.54, 95% CI [3.56–5.46]; Δ = 4.41, 95% CI [3.62–5.21]), and ‘depression’ ( $\hat{\mu }$ = −0.63, 95% CI [−1.61 to 0.29]; Δ = 9.58, 95% CI [8.78–10.37]; Fig. 1).

Content validity of the TAS-20 subscales

Difficulty identifying feelings subscale

Analyses indicated that the items of the ‘difficulty identifying feelings’ subscale scored significantly higher on ‘alexithymia’ ( $\hat{\mu }$ = 3.58, 95% CI [2.10–5.12]), compared to ‘anxiety’ ( $\hat{\mu }$ = −1.56, 95% CI [−3.05 to −0.02]; Δ = 5.14, 95% CI [4.31–5.97]), ‘depression’ ( $\hat{\mu }$ = −2.84, 95% CI [−4.32 to −1.30]; Δ = 6.42, 95% CI [5.60–7.25]), and ‘health anxiety’ ( $\hat{\mu }$ = −2.16, 95% CI [−3.65 to −0.62]; Δ = 5.75, 95% CI [4.92–6.57]; Fig. 2A).

Furthermore, the items of the ‘difficulty identifying feelings’ subscale scored highest on ‘difficulty identifying feelings’ ( $\hat{\mu }$ = 5.62, 95% CI [3.06–7.75]). Yet, compared to ‘difficulty describing feelings’, the difference was not significant ( $\hat{\mu }$ = 2.70, 95% CI [0.15–4.81]; Δ = 2.92, 95% CI [2.18–3.65]). Furthermore, items of the ‘difficulty identifying feelings’ subscale scored significantly higher compared to ‘externally-oriented thinking’ ( $\hat{\mu }$ = −4.53, 95% CI [−7.06 to −2.42]; Δ = 10.15, 95% CI [9.42 to 10.88]), and ‘limited imaginal capacity’ ( $\hat{\mu }$ = −3.59, 95% CI [−6.15 to −1.48]; Δ = 9.21, 95% CI [8.48 to 9.94]; Fig. 2B).

Difficulty describing feelings subscale

Analyses showed that the items of the ‘difficulty describing feelings’ subscale scored significantly higher on ‘alexithymia’ ( $\hat{\mu }$ = 4.42, 95% CI [2.42–6.27]), compared to ‘anxiety’ ( $\hat{\mu }$ = −4.08, 95% CI [−6.07 to −2.23]; Δ = 8.49, 95% CI [7.66–9.33]), ‘depression’ ( $\hat{\mu }$ = −3.57, 95% CI [−5.56 to −1.72]; Δ = 7.99, 95% CI [7.15 to 8.83]), and ‘health anxiety’ ( $\hat{\mu }$ = −6.84, 95% CI [−8.83 to −4.97]; Δ = 11.25, 95% CI [10.41–12.09]; Fig. 2A).

Furthermore, the items of the ‘difficulty describing feelings’ subscale scored significantly higher on ‘difficulty describing feelings’ ( $\hat{\mu }$ = 6.72, 95% CI [4.56–8.65]), compared to ‘difficulty identifying feelings’ ( $\hat{\mu }$ = 1.97, 95% CI [−0.19 to 3.91]; Δ = 4.74, 95% CI [3.82–5.67]), ‘externally-oriented thinking’ ( $\hat{\mu }$ = −5.06, 95% CI [−7.20 to −3.14]; Δ = 11.78, 95% CI [10.86–12.70]), and ‘limited imaginal capacity’ ( $\hat{\mu }$ = −3.45, 95% CI [−5.59 to −1.53]; Δ = 10.16, 95% CI [9.24 to 11.08]) (see Fig. 2B).

Externally-oriented thinking subscale

The items of the ‘externally-oriented thinking’ subscale scored highest on ‘alexithymia’ ( $\hat{\mu }$ = −3.05, 95% CI [−4.13 to −1.93]). Yet, there is no significant difference compared to ‘anxiety’ ( $\hat{\mu }$ = −4.76, 95% CI [−5.84 to −3.63]) (Δ = 1.71, 95% CI [1.11–2.30]), and ‘depression’ ( $\hat{\mu }$ = −4.89, 95% CI [−5.97 to −3.76]) (Δ = 1.83, 95% CI [1.24–2.43]). ‘Alexithymia’ scored significantly higher compared to ‘health anxiety’ ( $\hat{\mu }$ = −6.62, 95% CI [−7.70 to −5.49]) (Δ = 3.57, 95% CI [2.97–4.16]). However, note that $\hat{\mu }$ was negative for all constructs, indicating that the items of the ‘externally-oriented thinking’ subscale were not endorsed to measure ‘alexithymia’, nor ‘anxiety’, ‘depression’, or ‘health anxiety’ (Fig. 2A).

Furthermore, the items of the ‘externally-oriented thinking’ subscale scored highest on ‘external oriented thinking’ ( $\hat{\mu }$ = −2.16, 95% CI [−3.69 to −0.62]). Yet, no significant difference was found compared to ‘difficulty describing feelings’ ( $\hat{\mu }$ = −3.26, 95% CI [−4.79 to −1.73]; Δ = 1.10, 95% CI [0.37–1.82]), ‘difficulty identifying feelings’ ( $\hat{\mu }$ = −2.57, 95% CI [−4.09 to −1.02]; Δ = 0.41, 95% CI [−0.31 to 1.14]), and ‘limited imaginal capacity’ ( $\hat{\mu }$ = −3.28, 95% CI [−4.82 to −1.73]; Δ = 1.13, 95% CI [0.41–1.85]). Also here, note that $\hat{\mu }$ was negative for all constructs, indicating that the items of the ‘externally-oriented thinking’ subscale were not endorsed to measure ‘externally-oriented thinking’, nor ‘difficulty describing feelings’, ‘difficulty identifying feelings’, or ‘limited imaginal capacity’ (Fig. 2B).

Discussion

The present study investigated the content and discriminant content validity of the TAS-20, a widely used self-report measure of alexithymia (Bagby, Parker & Taylor, 2020). Using the Discriminant Content Validity method (DCV; Johnston et al., 2014), participants rated the extent to which each TAS-20 item was relevant for measuring ‘alexithymia’ and its key features (content validity), or related constructs, i.e., ‘anxiety’, ‘depression’, and ‘health anxiety’ (discriminant content validity). The results can be readily summarized. First, results showed that participants did not endorse the TAS-20 as measuring ‘alexithymia’, whereas the PROMIS-A, PROMIS-D, and SHAI did distinctively measure their intended construct. Second, the subscales ‘difficulty identifying feelings’ and ‘difficulty describing feelings’ measure ‘alexithymia’. This was not the case for the ‘externally-oriented thinking’ subscale. Additionally, results indicated that the ‘difficulty describing feelings’ subscale distinctively assessed its intended construct. This was not the case for the ‘difficulty identifying feelings’ subscale, which measures both ‘difficulty identifying feelings’ and ‘difficulty describing feelings’. Furthermore, the ‘externally-oriented thinking’ subscale assessed none of the included constructs. Finally, results showed that eight items distinctively measured their intended construct, and four items measured both the ‘difficulty identifying feelings’ and the ‘difficulty describing feelings’ constructs. Two items assessed (health) anxiety equally well or even better, and none of the items (except item 3) showed content overlap with ‘anxiety’ or ‘depression’.

To our knowledge, this study is the first to empirically investigate the content of the TAS-20. Until now content validity has been largely overlooked at the expense of other forms of validity, such as construct (i.e., convergent and discriminant validity) and criterion validity (i.e., predictive, concurrent, and retrospective validity) (e.g., Lumley, Neely & Burger, 2007; Parker, Taylor & Bagby, 2003; Bagby, Parker & Taylor, 2020). This is surprising as content validity is a fundamental property of any measure of any theoretical construct (Haynes et al., 1995) and key in theory testing, intervention design, and practical applications (Dixon & Johnston, 2019; Van Ryckeghem, 2021).

The results of the current study call for reflection. First, overall, the TAS-20 was not considered relevant for measuring ‘alexithymia’. Furthermore, the ‘difficulty identifying feelings’ subscale assessed both the ‘difficulty identifying feelings’ and the ‘difficulty describing feelings’ constructs, and the ‘externally-oriented thinking’ scale was not identified as measuring ‘externally-oriented thinking’, nor ‘alexithymia’. These findings put a threat on the interpretation of earlier and future studies using the TAS-20 as findings are potentially flawed due to a lack of content validity. Nonetheless, the finding that only the ‘difficulty identifying feelings’ subscale and the ‘difficulty describing feelings’ subscale were content valid for ‘alexithymia’ is not surprising. Sifneos (1973) (p. 256) stated that “for lack of a better term”, he proposed the term “alexithymic” (from Greek stems a = lack, lexis = word, and thymos = mood or emotion) to denote “the most striking characteristic”, namely the inability of these patients to find appropriate words to describe their feelings (Nemiah & Sifneos, 1970). Although the literal meaning of the term alexithymia-‘without words for feelings’-refers to this particular characteristic (Apfel & Sifneos, 1979), Sifneos made repeatedly clear that the term ‘alexithymia’ is the name of a construct that encompasses multiple characteristics (e.g., Nemiah, Freyberger & Sifneos, 1976; Sifneos, 1994, 1996). Therefore, to define the alexithymia construct in our study, we chose not to use the literal meaning but instead turn to the definition that is used in scientific literature (Taylor, Bagby & Parker, 2016) and understood in lay terms (online Oxford Living Dictionaries for English)—‘The inability to recognize one’s own emotions and to express them, especially in words.’ Although this definition has a broad scope, its focus is on only two out of the four key features, namely the ‘difficulty identifying feelings’ feature referring to the inability to recognize and the ‘difficulty describing feelings’ feature referring to the inability to express. In line with this reasoning, we see a plausible explanation for the finding that the ‘externally-oriented thinking’ subscale was not identified to measure the alexithymia construct. However, by including each of the definitions of the key features, we expected that the ‘externally-oriented thinking’ subscale would be identified as a measure of the externally-oriented thinking construct, and potentially as an indirect measure of the limited imaginal capacity construct. This was not the case. Analyses of the individual externally-oriented thinking items corroborated this finding. Only two items of this scale were perceived as measuring externally-oriented thinking and none as measuring limited imaginal capacity. Closer inspection of these results showed that four of the items that did not perform well, were reverse-scored items. However, this is not a full explanation because the two other items were regularly keyed. This observation is in line with the findings of Preece et al. (2018a), who performed confirmatory factor analysis and showed the presence of a reverse-scored item method factor, but also poor factor loadings for the regular-keyed items. Our findings are important as they signal the need of revising the items designed to measure externally-oriented thinking so that they represent their intended construct more accurately. Furthermore, our findings contradict Bagby, Parker & Taylor (1994a, 1994b) assumption on the representation of the limited imaginal capacity feature in externally-oriented thinking items. The present results indicate that caution is warranted in using the TAS-20 in its entirety: only two out of the four key features of the alexithymia construct are represented in the item pool.

We also observed that multiple difficulty identifying feelings items had higher scores on the ‘difficulty identifying feelings’ construct than on the ‘difficulty describing feelings’ construct. Furthermore, one difficulty describing feelings item had higher scores on the ‘difficulty describing feelings’ construct than on the ‘difficulty identifying feelings’ construct. These findings are in line with the results of various studies, revealing that difficulty identifying feelings items and difficulty describing feelings items are closely related, and subscale scores often correlate highly (e.g., r = 0.43–0.80; Kooiman, Spinhoven & Trijsburg, 2002). Current findings suggest that content overlap between both subscales may (at least partly) be the basis of observed correlations. Together with the fact that some studies showed that the items of these subscales merge into one single factor (e.g., Erni, Lötscher & Modestin, 1997; Loas et al., 1996), current findings provide support for the idea that part of the items on these scales probably represent the same aspect of alexithymia (Kooiman, Spinhoven & Trijsburg, 2002, but see Gignac, Palmer & Stough, 2007). Future research is needed to examine why the wording and phrasing of some of these items is perceived as measuring both constructs.

Finally, two TAS-20 items of the ‘difficulty identifying feelings’ scale that are developed to measure the difficulty in differentiating between bodily feelings and emotions showed to measure other constructs. One TAS-20 item was identified to measure ‘anxiety’ and ‘health anxiety’ (i.e., “I have physical sensations that even doctors don’t understand.”), the other item was identified to measure both ‘health anxiety’ and ‘difficulty identifying feelings’ (i.e., “I am often puzzled by sensations in my body”). Due to this content overlap, these TAS-20 items may result in a misleading evaluation of patients suffering from medically unexplained symptoms. A total score on the TAS-20 may overestimate the prevalence or severity of alexithymia in patients with medical conditions. To avoid unduly psychologization, caution is warranted in interpreting the TAS-20 overall score as a straightforward measure of alexithymia in these populations. One possibility would be to make a separate scale of these items allowing to check/control for their contribution in the TAS-20 total score. A recent study of Fournier et al. (2019) provides a potential starting point. In line, with current findings they found that both item 3 and item 7 form a new latent factor difficulty in interoceptive abilities that is specifically related to health and personality trait outcomes. Alternatively, current findings may indicate that a standalone use of this self-report questionnaire may be insufficient to assess alexithymia. There is a need to take into account the context. Indeed, the use of a multimethod approach including the TAS-20, accompanied by a clinical interview and expert judgement may be required to assess whether elevation of self-reported alexithymia levels can be attributed to different aspects of the presentation of alexithymia, the context of the person, or comorbid medical conditions (see also Bagby et al., 2006; Taylor, Bagby & Parker, 2016). Future research is warranted to further scrutinize this topic.

Finally, the TAS-20 items did show discriminant content validity with ‘anxiety’ (except item 3) and ‘depression’. This supports the idea that the TAS-20 is not merely a measure of negative affect (Lumley, 2000; Bagby, Parker & Taylor, 2020), indicating that high correlations between alexithymia and anxiety/depression are not due to content overlap between scales of both constructs. Yet, it remains possible that the high correlations between alexithymia and negative affect are due to the particular formulation of a substantial part of the TAS-20 items. Indeed, a substantial number of items of the ‘difficulty identifying feelings’ subscale and the ‘difficulty describing feelings’ subscale are negatively phrased (e.g., “I find it hard to describe how I feel about people”). It is known that people high in negative affectivity tend to manifest a general tendency towards a self-effacing response style or self-criticism, thus, tend to report negative things about themselves on self-report questionnaires generally (Lumley, 2000).

This study has some limitations. First, healthy lay people, and no experts or patients were involved. The nature of the discriminant validity method is designed to allow lay people without scientific background (and thus knowledge biases) to judge whether items assess a certain construct (see also Crombez et al., 2020). However, no agreement exists whether experts should be used who are familiar with the theoretical constructs, or whether non-biased lay people should be used who are the putative respondents of the measure (Dixon & Johnston, 2019). Second, judges were mainly female, which precluded the examination of gender effects. Third, the DCV method provides a quantitative analysis of content validity. Other methods are possible, and may provide insight in how participants mentally process and respond to items. One promising procedure to provide a qualitative analysis of content validity is cognitive interviewing (Willis, 2015). Fourth, the statistical analyses differed from the analyses performed in typical research with the DCV method (Johnston et al., 2014, but see Crombez et al., 2020). These studies generally involve a low number of participants (at least 15 participants are recommended) and use primarily one-sample or paired t-tests. Hence, results then strongly depend on sample size and statistical power. The analyses in the present research were performed using Bayesian hierarchical models, which has several advantages (see earlier), but also requires a larger number of participants. Therefore, the sample size of current study largely exceeds previous studies resulting in narrow confidence intervals. Fifth, we have only included the TAS-20. Other measures exist such as the Bermond Vorst Alexithymia Questionnaire (BVAQ; Vorst & Bermond, 2001), the Psychological Treatment Inventory-Alexithymia Scale (PTI-AS; Gori et al., 2012), and the Perth Alexithymia Questionnaire (PAQ; Preece et al., 2018b). Although promising results on the psychometric properties of the latter measures have been reported (de Vroege et al., 2018; Bagby et al., 2009; Vorst & Bermond, 2001), future research is needed to corroborate the (discriminant) content validity of these questionnaires as well.

Conclusions

The TAS-20, currently the most utilized instrument to assess alexithymia and its key features is found to be only partially content valid. Particularly, current findings indicate problems with the content validity of the TAS-20 questionnaire, and some of its subscales do not measure the intended key features. Indeed, only the subscales ‘difficulty identifying feelings’ and ‘difficulty describing feelings’ represented ‘alexithymia’ and their intended construct. This was not the case for the ‘externally-oriented thinking’ subscale, which assessed none of the alexithymia key features or related constructs. Finally, some items of the TAS-20 are contaminated with content measuring (health) anxiety. Due to described problems with (discriminant) content validity, revision of the TAS-20 is recommended to adequately assess (all key features of) alexithymia. Furthermore, caution is warranted when assessing the TAS-20 in people suffering from medical conditions.

Supplemental Information