Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Jan 19.
Published in final edited form as: Psychol Men Masc. 2021 Nov 18;23(1):133–142. doi: 10.1037/men0000373

Variance Composition, Reliability, Validity, and Measurement Invariance of the Toronto Alexithymia Scale-20 in Nonclinical Samples

Ronald F Levant 1, Mike C Parent 2, Shu Ling 1, Nicholas C Borgogna 3
PMCID: PMC10798665  NIHMSID: NIHMS1913923  PMID: 38250472

Abstract

The present study extended prior work on the Toronto Alexithymia Scale-20 (TAS-20). The TAS-20 total scale score is commonly used in research in the psychology of men and masculinities and in clinical practice, but most published confirmatory factor analyses (CFA’s) do not support this use. Using a sample of 913 men and women, variance composition was assessed, and model-based dimensionality and reliability coefficients were calculated, finding evidence for unidimensionality, which supports the use of the total scale raw score but not that of the subscales, although the model fit was less than desired. Convergent and concurrent evidence was found for the validity of the TAS-20 in men as a unidimensional scale (N = 505) by examining relationships with latent variables of several constructs in the nomological network. An assessment of measurement invariance by gender (men, women) found evidence for metric invariance, indicating that cisgender men and women understand the scale in the same way. It is recommended that the TAS-20 scale developers follow recommended procedures to trim composite measurement scales to improve the psychometric properties (i.e., model fit) of the TAS.

Keywords: Toronto Alexithymia Scale-20, assessment of variance composition, model-based reliability and dimensionality coefficients, validity, measurement invariance

The purpose of this study was to assess the evidence for using the total scale score of the Toronto Alexithymia Scale-20 (TAS-20; Bagby et al., 1994). This is an important task because recent research has pointed to the “hidden invalidity” of many psychological scales, with particular reference to structural validity—that is, dimensionality (factor structure) and measurement equivalence/invariance (Hussey & Hughes, 2020). The TAS-20 total raw scale score is often used in research in the psychology of men and masculinities (PMM), as well as in clinical practice, largely because of prior research establishing criteria for its use in diagnosing alexithymia (Taylor et al., 1988). However, empirical support for the use of the total raw scale score would require a demonstration that, despite the intended multidimensional nature of the scale, it can be used as unidimensional measure of alexithymia. This demonstration might involve confirmatory factor analytic (CFA) results confirming the existence of either a unidimensional structure, in which the items load only on a general alexithymia factor, or a bifactor structure, in which the items load on both their group factors (corresponding to the subscales) and a general alexithymia factor (corresponding to the total scale score), or a hierarchical structure, in which the items load only on their first-order factor (corresponding to the subscale scores), which in turn load on the second-order factor (corresponding to the total scale score). To conduct this investigation, the variance composition of the TAS-20 was examined, and model-based reliability and omega coefficients were calculated. In addition, evidence for the validity of a unidimensional TAS-20 for men was assessed. Finally, measurement invariance of the unidimensional TAS-20 by gender was assessed.

What Is Alexithymia and Why Is It Important to the Psychology of Men and Masculinities?

Alexithymia is a construct that evolved from clinical observations of patients with psychosomatic complaints in the 1940s. While Sifneos (1967) coined the term alexithymia to refer to two dimensions—difficulties identifying feelings and difficulties describing feelings, Nemiah et al. (1976) observed that patients with psychosomatic complaints were also preoccupied with the details of external events and objects (termed externally oriented thinking), and thus included that feature along with the other two in their theoretical definition of alexithymia. Alexithymia so defined was soon observed in patients suffering from other disorders, including eating, substance use, and posttraumatic stress disorders (Taylor, 2004).

Although alexithymia has been characterized as normally distributed in the general population (Mattila et al., 2010; Parker et al., 2008), it is actually nonnormally distributed due to its low prevalence; that is, the distribution is skewed to the low end. Furthermore, it affects men more often than women. Levant et al. (2009) conducted a meta-analysis of sex differences in alexithymia based on 41 samples and found consistent, although expectedly small, differences in mean alexithymia scores between women and men (Hedges’ d = .22), with men exhibiting more alexithymia.

Alexithymia has been investigated frequently in PMM studies not only because it is observed more frequently in men, but also because of its link with the traditional masculine norm and gender role conflict pattern of restrictive emotionality (Levant et al., 2006; Wester et al., 2012). Levant (1992) observed mild-to-moderate alexithymia in a nonclinical sample of men who were participants in a fatherhood education class. Levant (1992) hypothesized the existence of “normative male alexithymia” (reflecting the influence of masculine norms) to describe what he observed in these men: An inability to identify and describe their emotions, even when coached while watching videotaped replays of themselves in role-play exercises designed to stimulate emotions. Levant (2011) later theorized that social learning and social influence processes, guided by traditional masculinity ideology (TMI—the masculinity ideology of the dominant group in society, namely White, cisgender, heterosexual men), produced these results. Specifically, Levant (1992, 2011) theorized that socialization agents enforced the norm of restrictive emotionality. This norm restricted the expression of vulnerable emotions such as hurt, fear, or sadness, as well as attachment emotions (fondness, caring, loneliness) that reflect a need for another person because in TMI the most honored way of being a man is to never show vulnerability and never need anyone. However, boys were allowed and encouraged to express emotions associated with aggression (such as irritation, anger, and rage) and, later on, lust.

In support of this theory, Levant and Williams (2009) reviewed the emotion socialization literature in developmental psychology and concluded that boys were more emotionally expressive than girls as neonates and retained this advantage in emotional expressivity until at least 1 year of age. However, boys fell behind girls with respect to verbal expression of emotions by 2 years of age, and facial expression between the ages of 4 and 6, when they entered school. Moreover, Levant et al. (2014) conducted a semantic priming experiment and found that men with alexithymia showed more errors in lexical decision performance using target emotion words discouraged by masculine norms (e.g., sadness, fear) compared with men without alexithymia. In addition, men with and without alexithymia did not differ in their accuracy when using target emotion words that are encouraged by masculine norms (e.g., anger). Finally, Jardin et al. (2019) used event-related potentials (ERPs) to examine the locus of emotional processing in alexithymia. Using criteria established by the scale developers (Taylor et al., 1988), they tested men, both those scoring high (score > 61) and low (score < 51) on the TAS-20, on an emotional face discrimination task. Their results suggest that processing differences between men with and without alexithymia men occur both early in perceptual processing and later in conscious processing.

Prior Examination of the Psychometric Properties of the TAS-20

The TAS-20 is the most popular scale for alexithymia assessment (Gignac et al., 2007; Meganck et al., 2008; Preece et al., 2017; Reise, Bonifay, et al., 2013). It has been used extensively worldwide. However, significant psychometric problems with the scale have been reported. The first CFA reported a three-factor common (or correlated) factors model (in which items load only on the factors corresponding to the subscales), with the factors corresponding to the three theorized dimensions—difficulty identifying feelings (DIF), difficulty describing feelings (DDF), and externally oriented thinking (EOT) (Parker et al., 1993). However, Parker et al. (1993) used fit statistics that have been criticized for their dependence on sample size (Hu & Bentler, 1998), and with values that do not meet contemporary criteria (i.e., Goodness of Fit Index; GFI = .886, Adjusted Goodness of Fit Index; AGFI = .856). A decade later, Parker et al. (2003) reported a CFA using the common factors model with a mixed-gender sample, finding adequate fit [GFI = .98, AGFI = .98, Comparative Fit Index (CFI) = .97, root-mean-square error of approximation (RMSEA) = .06]. However, these results do not support the use of the total scale score and conflict with those of a series of other studies.

Several psychometric studies found that the first two factors (DIF, DDF) collapsed into one (Erni et al., 1997; Kooiman et al., 2002; Loas et al., 1996). Furthermore, evidence for the dimensionality and reliability of the EOT scale has also been questioned (Gignac et al., 2007; Meganck et al., 2008; Preece et al., 2017). In a review, Kooiman et al. (2002, p. 1083) noted that “in practically all studies the dimension ‘externally oriented thinking’ (EOT) appears to be unreliable.” As a result, several different approaches have been taken with regard to EOT, including breaking EOT into two subscales (Haviland & Reise, 1996; Meganck et al., 2008; Ritz & Kannapin, 2000), and also creating a method factor to account for the fact that all but one of the negatively keyed items load on the EOT factor (Gignac et al., 2007; Meganck et al., 2008). Thus, the dimensional structure of the TAS-20 is questionable.

Four studies assessed the dimensional structure of the TAS-20. First, Gignac et al. (2007) tested common factors and hierarchical models specifying the original three subscales, and four bifactor models. The latter had a general alexithymia factor plus one of several sets of group factors. Of all the models tested, only one model had acceptable levels of incremental fit and good levels of absolute fit; however, the required covariance link can only be accomplished using structural equation modeling (SEM), precluding the use of raw scores. Second, Meganck et al. (2008), using both clinical and student samples, found unacceptable fit for a unidimensional model. Two of three hierarchical models tested had worse fit than their respective common factors models. These models specified a second-order factor and one of two sets of first-order factors: (a) original three subscales and (b) combining DIF and DDF and splitting EOT into two factors. A third hierarchical model with separate DIF and DDF factors and splitting EOT into two factors had equivocal fit. Third, Preece et al. (2017) tested one unidimensional model, four common factors models, and three hierarchical models. They found that the hierarchical models resulted in decrements in fit as compared to their common factors counterparts in both samples, and all had unacceptable fit statistics. Finally, Tuliao et al. (2020) examined 191-factor structures in two samples, from the U.S. and the Philippines. Only one model in both samples met the criteria for acceptable fit criteria, and this was the only model that dropped items, suggesting that model trimming may be the way to improve the scale’s psychometric properties.

Finally, Reise, Bonifay, et al. (2013) took a very different approach to study the dimensionality of the TAS-20, deploying new psychometric indices (e.g., model-based dimensionality and reliability indices) to determine the extent that the TAS-20 total scale scores can be reliably reported. They found “compelling evidence that the item responses are ‘essentially unidimensional’” (p. 138). As will be discussed, while the present study replicates dimensionality and reliability analyses conducted by Reise, Bonifay, et al. (2013), it goes beyond it in also assessing evidence for validity in men using latent variables and measurement invariance by gender.

The Present Study

The first aim was to assess the dimensionality of the TAS-20 by estimating its fit to the data in unidimensional, common factors, bifactor, and hierarchical models. The second aim was to assess the scale’s dimensionality and the reliability of the total scale and subscale scores using the technique utilized by Reise, Bonifay, et al. (2013), namely calculating model-based dimensionality and reliability coefficients. Based on Reise, Bonifay, et al. (2013), we hypothesized (H1) that this analysis will provide evidence for a unidimensional structure. The third aim was to assess the validity of the TAS-20 in men. Validity information for the TAS-20 based on raw scores is abundant (cf., Taylor, 2004), but we are not aware of any studies that provide TAS-20 validity evidence for men using latent variables, which is a more stringent method as it removes many sources of error. Thus, this aim was to assess the convergent construct and concurrent criterion-related evidence for the validity of the TAS-20 using a latent variable approach. A CFA model which regressed the TAS-20 model on several related latent constructs was estimated.

To assess convergent construct evidence for validity, the gender-linked Normative Male Alexithymia Scale (NMAS) was used. The NMAS has been moderately associated with the TAS-20 (Levant et al., 2006). Hence, we hypothesized (H2) that the TAS-20 will be significantly, moderately, and positively related to the NMAS.

Conceptually, alexithymia should be inversely related to the ability to identify and describe one’s emotions toward others and oneself. Two scales were used to assess concurrent criterion-related evidence for validity by these dimensions. To assess the relationship between alexithymia and emotional identification/description toward others, we used the Revised Adult Attachment Scale (RAAS), which has three subscales: Close, Depend, and Anxiety (Collins & Read, 1990). Previous research has found a moderate, negative relationship between alexithymia and relationship satisfaction (Humphreys et al., 2009). Furthermore, the NMAS (closely linked to the TAS-20) has been found to have small negative correlations with relationship consensus, cohesion, satisfaction, and affectional expression, and a moderately positive one with fear of intimacy (Karakis & Levant, 2012). Therefore, based on the above-defined conceptualization of alexithymia and on previous research, we hypothesized that the TAS-20 would have significant, small, and negative correlations with the Close Scale of the RAAS (H3), which measures the extent to which a person is comfortable with closeness and intimacy and to the Depend scale of the RAAS (H4), which measures the extent to which a person feels they can depend on others to be available when needed. We also hypothesized that the TAS-20 would demonstrate a significant, small, and positive correlation with the Anxiety scale, which measures the extent to which a person is worried about being abandoned (H5).

We additionally wanted the extend the convergent evidence for the validity of the TAS-20 by examining how it would be related to a measure of emotional identification/description toward aspects of oneself. To do this, we used the Body Appreciation Scale (BAS; Avalos et al., 2005). Conceptually, individuals higher in alexithymia should demonstrate less emotional identification/description toward themselves, such as in regard to their body image. By extension, the TAS-20 should be inversely correlated with the body appreciation. Prior research supports this position, as Leone et al. (2015) showed that men likely to have alexithymia (as measured by the TAS-20) were also more likely to demonstrate symptoms of muscle dysphoria. Hence, we hypothesized (H6) that the TAS-20 would be significantly, small to moderately, and negatively correlated with the BAS.

The final aim was to assess the measurement invariance of the TAS-20 across gender (men, women).

Method

Participants (Sample 1)

Sample 1 was drawn from a larger project (Levant et al., 2014). A total of 913 university students were included in the analysis (73.4% men and 26.6% women2). Participants’ age ranged from 18 to 54 years, with a mean of 20.88 (SD = 4.44). Women and men did not differ in mean age. Most participants identified as White (79.4%), however 14.1% identified as Black, 3.6% as Middle Eastern, 3.5% as Asian/Asian American, 2.8% as Latinx, 1.9% as American Indians, 0.2% as Pacific Islanders, and 0.6% as other. Most participants were heterosexual (91.8%), yet 3.6% were gay/lesbian, 3.2% were bisexual, and 1.4% identified either as “other” or did not answer.

Participants (Sample 2)

Sample 2 was drawn from a larger project (Parent & Bradstreet, 2019). A total of 505 men were included in the analysis. Participant’s age ranged from 19 to 73 years, with a mean of 35.28 (SD = 11.08). Regarding race/ethnicity, most participants identified as White (73.9%), but 11.3% identified as Asian or Asian American, 5.1% as Black, 4.8% as multiracial, 3.2% as Hispanic, 1.6% as either American Indian or did not respond. Most (90.3%) participants identified as heterosexual, but 4.2% indicated they were bisexual, 3.6% indicated they were gay, and 2.0% indicated a different identity or did not respond.

Recruitment and Survey Procedure

For Sample 1, the study was approved by the first author’s university Institutional Review Board (IRB). University students were recruited using departmental undergraduate research participation pools. Using Qualtrics, after completing the informed consent, participants filled out the questionnaires and were provided with an educational debriefing. Following completion of the study, participants were redirected to another Qualtrics site where they could confidentially enter their information to receive course credit.

For Sample 2, the study was approved by the second author’s university IRB. Community-dwelling participants were recruited using Amazon’s Mechanical Turk (Mturk) service. Access to the survey was restricted to individuals in the United States who had 95% or better approval on prior Mturk tasks. Using Qualtrics, after completing the informed consent, participants filled out the questionnaires, and were provided with a debriefing. The survey contained two attention checks; those who did not answer the attention check items correctly were exited from the survey and their data were not used. Following completion, a payment of $1.00 was credited to participants’ Mturk accounts.

Sample Size Considerations

For the CFAs required for the assessment of variance composition and calculation of model-based reliability and omega coefficients, Kline (2016) recommended a minimum of 10 participants for every freely estimated parameter. The bifactor CFA had the largest number of parameters at 80, requiring 800 participants. Our n of 913 exceeded this number. The validity analysis using structural regression had 69 parameters, requiring 690 participants. Our n of 505 was less than this number and may not be adequate. However, using Soper’s (2013) a priori sample size calculator for structural equation models, with a small effect size of .20, power of .8, 6 latent variables, 18 observed variables, and p < .05, a sample size of 403 is sufficient. Finally, for the analysis of measurement invariance, the largest number of parameters for the basic test had 60 parameters, requiring 600 participants, for which our n of 913 was more than sufficient.

Measures

Toronto Alexithymia Scale-20

The TAS-20 (Parker et al., 1993) was created to assess alexithymia. Exploratory factor analysis supported a three-factor structure: DIF, DDF, and EOT, although (as detailed above) subsequent CFAs indicate that the dimensionality is not clear. Participants respond to items using a 5-point Likert scale (1 = strongly disagree; 5 = strongly agree). Internal consistency has been shown to be reliable (Parker et al., 1993). The TAS-20 was found to be positively correlated with measures of somatic complaints (Taylor et al., 1992).

Normative Male Alexithymia Scale

The NMAS (Levant et al., 2006) is a 20-item inventory designed to assess a gender-linked form of alexithymia—normative male alexithymia. Participants answered questions about their own experience of emotions on a 7-point scale (1 = strongly disagree; 7 = strongly agree). Seven items are reverse-scored. A mean is taken, with higher scores indicating greater normative male alexithymia. EFA and CFA using separate samples indicated that the NMAS consisted of a single 20-item factor. Men’s scores on the NMAS displayed very good internal consistency and test–retest reliability over a 1–2 month period. Results of analyses of gender differences, relations of the NMAS with other instruments, and its incremental validity in predicting masculinity ideology provided evidence supporting validity.

Revised Adult Attachment Scale

The RAAS (Collins & Read, 1990) is an 18-item inventory designed to assess adult romantic relationships along three dimensions: Close, Depend, and Anxiety. High scores on the Close scale characterize individuals who are comfortable with closeness and intimacy (Collins, 1996). High scores on the Depend scale characterize individuals who feel they can depend on others to be available when needed. High scores on the Anxiety subscale characterize individuals who are worried about being rejected or unloved. Participants rated their feelings about romantic relationships using a 5-point scale (1 = not at all characteristic of me, 5 = very characteristic of me). A review of 25 instruments on attachment found that compared with other instruments, RAAS had good test–retest, internal consistency, and inter-item reliability, and good convergent and discriminant validity (Ravitz et al., 2010). The test–retest reliabilities for the three subscales range from .52 to .71 (Collins & Read, 1990). The internal consistency of the subscales has been shown to be >.76 (Eng et al., 2001).

Body Appreciation Scale

The BAS (Avalos et al., 2005) is a 13-item measure that examines four related components of positive body image: Holding positive opinions of the body, acceptance of the body despite its imperfections, respect for the body, and protection of the body through the rejection of unrealistic ideals. All items were rated on a 5-point scale (1 = never; 5 = always). Higher scores indicate more positive body image. The internal consistency of the measure has been supported with a non-clinical sample: Cronbach’s α = .94 (Avalos et al., 2005). It shows good test-retest reliability (Avalos et al., 2005), and gender equivalence (Tylka, 2013). BAS scores have been shown to be correlated with men’s muscularity dissatisfaction, body fat dissatisfaction, height dissatisfaction, and overall body dissatisfaction (Tylka, 2013).

Data Analytic Procedure

We first estimated unidimensional, common factors, bifactor, and hierarchical models of the TAS-20. Second, following the recommendations of Reise, Bonifay, et al. (2013), we calculated model-based reliability and omega coefficients to assess the TAS-20’s dimensionality, which required entering the standardized factor loadings for both the unidimensional and bifactor models into a model-based reliability calculator developed by Dueber (2017). For the CFAs and the testing of Hypotheses 1–6, we used Mplus v.8.3 (Muthén & Muthén, 1998–2015) SEM software. The overall fit of any CFA models was assessed with the scaled chi-square goodness-of-fit test. However, because this statistic is dependent on sample size, it is overly sensitive to trivial sources of model misfit when sample sizes are large (Cheung & Rensvold, 2002). Thus, we used a set of alternative fit indices to determine whether a model demonstrates adequate fit (Kahn, 2006). Acceptable model fit is indicated by the Bentler comparative (CFI) and Tucker–Lewis fit indices (TLI) over .90, root mean square error of approximation (RMSEA) under .08, and standardized root-mean-square residual (SRMR) under .10. Good fit is indicated by a CFI and TLI over .95, RMSEA under .05, and SRMR under .05.

For the assessment of measurement invariance, the fits of nested CFA models were compared using a scaled chi-square difference tests, which were adjusted for the use of the maximum-likelihood estimation with robust standard errors (MLR; Satorra & Bentler, 2001). However, similar to the chi-square goodness-of-fit test, the scaled chi-square difference test (Δχ2) is affected by large sample sizes (Cheung & Lau, 2012; Cheung & Rensvold, 2002). Since the Δχ2 is expected to be statistically significant in samples larger than 300 (Kline, 2016), we also utilized a ΔCFI with a cut-off score of <.01 (Chen, 2007; Cheung & Lau, 2012; Cheung & Rensvold, 2002).

The descriptive statistics were calculated using SPSS 25. For the validity hypotheses, we followed the recommendations of Russell et al. (1998) and created three-to-four-item parcels from the manifest variables for each instrument or subscale that had six or more observed items. Item parcels were created by performing principle axis exploratory factor analyses with one-factor solutions for the items comprising each scale. Iterative assignment of items into one of the parcels was done to ensure that parcel loadings were balanced. Effect sizes were graded according to Ferguson’ s (2009) criteria for strength of association, where .2 = small, .5 = moderate, and .8 = large.

Results

Assessment of Dimensionality and Reliability

Before data analyses, both samples were checked for ineligible participants, outliers, normality, and missing data.

Dimensionality

As noted above, we first conducted CFA’s modeling the TAS-20 as common factors, bifactor, hierarchical and unidimensional models. In the bifactor model, all factors were constrained to be orthogonal to each other to allow the uncontaminated assessment of sources of variance in each item. As expected from prior research, none of the models fit well, although the bifactor model had the best fit. See Table 1 for fit statistics. Ordinarily, we would follow recommended procedures to trim composite measurement scales (cf., Goetz et al., 2013) to improve the model fit; however, copyright restrictions prevented us from doing that (Taylor, personal communication, 2/1/21).

Table 1.

Model Fit Statistics and Comparisons of Single-Group TAS-20 Models

Single-group model χ2(df) CFI, TLI RMSEA estimate and 90% CI SRMR
SG1: common factors    531.72 (167) .697, .655 .098 [.089, .108] .092
SG2: bifactor    731.66 (150) .888, .858 .065 [.060, .070] .052
SG3: hierarchical    531.72 (167) .697, .655 .098 [.089, .108] .092
SG4: unidimensional 1,499.74 (170) .744, .714 .093 [.088, .097] .086
Comparison Δχ2(df) p ΔCFI Conclusion
SG1 versus SG2 79.48 (17) .00001 +.19 Prefer SG2
Open in a new tab

Note. TAS-20 = Toronto Alexithymia Scale-20; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = root-mean-square error of approximation; CI = confidence interval; SRMR = standardized root-mean-square residual.

Given that the fit of the bifactor model was adequate in regard to RMSEA and SRMR and only marginally inadequate in regard to CFI, the standardized factor loadings for both the unidimensional and bifactor models were then entered into Dueber’s (2017) model-based dimensionality and reliability calculator. Table 2 summarizes the results. With regard to dimensionality, using the Reise, Scheines, et al. (2013) criteria, we missed the criterion of high Percentage of Uncontaminated Correlations (PUC) (>.80), which would have meant low risk of parameter bias if it had been met. However, Reise, Scheines, et al. (2013) also stated that low PUC (<.80; ours was .69), along with high explained common variance (ECV) (>.60; ours was .66) and high Omega Hierarchical (>.70; ours was .76) also means low risk of parameter bias. Hence, modeling the TAS-20 as a unidimensional instrument would not likely lead to significant measurement parameter bias (i.e., biased item factor loadings). There is also relative parameter bias—the difference between an item’s loading in the unidimensional solution and its general factor loading in the bifactor model, divided by the general factor loading in the bifactor model “Average parameter bias less than 10%–15% is acceptable and poses no serious concern” (Rodriguez et al., 2016, p. 145). Our average relative parameter bias is within this range at 11.3%. Finally, there is ECV, which was .66 for the general factor and .80 for EOT, but only .19 and .20 for DIF and DDF, suggesting that DIF and DDF essentially remeasure the general factor.3 In summary, the dimensional indices support the unidimensional structure of the TAS-20.

Table 2.

Explained Common Variance and Model-Based Reliability Estimates for the TAS-20

Model based coefficients General DIF DDF EOT
ECV .66 .19 .20 .80
Omega .89 .90 .80 .59
Omega H .76 — — —
Omega HS — .15 .16 .47
Relative Omega .86 .16 .20 .79
Open in a new tab

Note. TAS-20 = Toronto Alexithymia Scale-20; DIF = difficulty identifying feelings; DDF = difficulty describing feelings; EOT = externally oriented thinking; ECV = explained common variance. Omega = A model-based estimate of internal reliability of the multidimensional composite. Omega H (Hierarchical) = Percentage of variance in raw total scores that can be attributed to the individual differences on the general factor. Omega HS (Hierarchical Subscale) = Percentage of reliable variance of a subscale score after partitioning out variance attributed to the general factor. Relative Omega = Omega Hierarchical divided by Omega.

Reliability

Omega (ω) is a factor analytic model-based estimate of the internal reliability of the multidimensional composite scale. For the group factors, only those items that load on a factor are considered in the calculation, whereas for the general factor all items are taken into account. The ω values range from .80 to .90 for the general factor and for DIF and DDF, indicating these factors are reliable; however, ω for EOT was only .59, replicating prior research that has found EOT to be unreliable. Omega Hierarchical (ωH) reflects the percentage of variance in raw total scores that can be attributed to the general factor, whereas Omega Hierarchical Subscale (ωHS) reflects the percentage of reliable variance of a subscale score after removing variance attributed to the general factor. Definitive guidelines for evaluating ωH and ωHS do not exist; however, Reise, Bonifay, et al. (2013) indicated that “tentatively, we can propose that a minimum would be greater than .50, and values closer to .75 would be much preferred” (p. 137). From this, we can see that none of the group factors even met the .50 criterion, whereas the general factor met the .75 criteria.

Relative Omega (ωH/ω) is Omega Hierarchical or Omega Hierarchical Subscale divided by Omega. For the general factor, this represents the percentage of reliable variance in the multidimensional composite that is due to the general factor; for the group factors, it represents the percentage of reliable variance in the subscale composite that is independent of the general factor. The relative ω for the general factor was .86, indicating that 86% of the reliable variance in the TAS-20 total scale score was due to the general factor. Thus, model-based reliability estimates support the use of the raw TAS-20 total score to represent general alexithymia. The relative ω for the group factors were DIF 16%, DDF 20%, and EOT 79% in terms of the proportion of the variance of the items loading on their factors that were independent of the general factor, indicating that (with the exception of EOT) the group factors largely remeasure the general factor.

Thus, the analysis of model-based dimensionality and reliability coefficients support the unidimensionality of the TAS-20 and therefore support the use of the raw total scale score, which can be said to represent the general alexithymia construct, supporting Hypothesis H1. Given the evidence for the scale’s unidimensionality and for the reliability of the general factor but not for the subscales, we recommend that only the total raw scale score be used in research and in clinical practice.

Descriptive Statistics

Raw-score-based correlation coefficients, α coefficients, means, and standard deviations for the TAS-20, NMAS, RAAS subscales, and the BAS are presented in Table 3.

Table 3.

Zero-Order Correlations, Means, Standard Deviations, and Coefficient Alphas of Study Variables Raw Scores for Men

Variable 1 2 3 4 5 6 M SD α
1. TAS-20 — .53** −.15** −.28** .34** −.21** 2.35 .78 .89
2. NMAS — −.62** −.55** .38** −.31** 3.91 1.08 .92
3. RAAS-close — .71** −.50** .26** 3.24 .83 .82
4. RAAS-depend — −.62** −.23** 3.03 .83 .81
5. RAAS-anxiety — −.30** 2.37 1.02 .92
6. BAS — 42.87 10.73 .94
Open in a new tab

Note. N = 505. TAS-20 = Toronto Alexithymia Scale-20 (scores ranged from 1 to 5); NMAS = Normative Male Alexithymia Scale (scores ranged from 1 to 6.9); BAS = Body Attitudes Scale (scores ranged from 13 to 65); RAAS-Close = the Close subscale of the Revised Adult Attachment Scale (scores ranged from 1 to 5); RAAS-Depend = the Depend subscale of the Revised Adult Attachment Scale (scores ranged from 1 to 5); RAAS-Anxiety = the Anxiety subscale of the Revised Adult Attachment Scale (scores ranged from 1 to 5).

**

p < .01.

Convergent and Concurrent Evidence for the Validity of the TAS-20

Based on the Chen et al.’s (2006) guidelines, we assessed the convergent and concurrent evidence for the unidimensional TAS-20’s validity. In this model latent factors for all scales were created using parcels as discussed above. A latent factor representing the general alexithymia factor was regressed on latent factors representing the validity measures of the NMAS, the three RAAS subscales, and the BAS. The CFA of the measurement model produced good fit to the data, χ2(174) = 554.58, p < .001, CFI = .961, TLI = .953, RMSEA = .056, 90% confidence interval (CI) = [.049, .062], SRMR = .04. All parcels had significant loadings of at least .40 on their respective factors. Next, regression paths were added from each of the validity factors to the TAS-20 factor. The fit statistics for this model were identical to the measurement model, and thus also showed good fit. With regard to the validity hypotheses, Hypothesis H2, predicting that the TAS-20 will show a significant moderate coefficient when it is regressed on the latent factor of the NMAS, was supported, B = 0.46, SE = .07, 90% CI [.32, .60]; β = .75, p < .001, 95% CI [.53, .97], providing convergent construct evidence for the validity of the TAS-20. Hypothesis H3, which stated that the TAS-20 will show a significant small negative coefficient when it is regressed on the latent factor of RAAS Close subscale was not supported, B = .17, SE = .23, 95% CI [−.28, .62]; β = .16, p = .47; 90% CI [−.28, .59]. Hypothesis H4, predicting that the TAS-20 will show a significant small negative coefficient when it is regressed on the latent factor of RAAS Depend subscale was not supported, B = .13, SE = .14,90% CI [−.13, .40]; β = .16,p = .326, 90% CI [−.16, .49]. Hypothesis H5, predicting that the TAS-20 will show a significant small positive coefficient when it is regressed on the latent factor of RAAS Anxiety subscale was supported, B = .20, SE = .04, 90% CI [.12, .28], β = .31, p < .001, 90% CI [.19, .43], providing concurrent criterion-related evidence for validity. Finally, Hypothesis H6, predicting that the TAS-20 will show a significant small–moderate negative coefficient when it is regressed on the latent factor of BAS was not supported, B = −.02, SE = .04,90% CI [−.09, .05], β = −.02, p = .615, 90% CI [−.12, .07]. Thus, two out of five validity hypotheses were supported, one providing convergent evidence, and one providing criterion evidence.

Assessment of Measurement Invariance by Gender

Prior to conducting invariance analyses, we tested the fit of the unidimensional TAS-20 in separate analyses for men and women. To identify the model in the single- and multigroup CFA’s, we constrained the latent factor variance to 1 and the mean to 0. Given that the unidimensional model did not fit the data well for the whole sample, it is not surprising that it also did not fit well for the unisex samples. For men, χ2(df) = 2,100.44 (170), p < .0001, CFI = .543; TLI = .489; RMSEA = .133, 90% CI = [.128, .138]; SRMR = .106. For women, χ2(df) = 917.25 (170), p < .0001, CFI = .545, TLI = .492, RMSEA = .137, 90% CI = [.128, .146], SRMR = 0.108. However, given the limitations we are working under (in terms of not being able to trim the model to improve fit), we proceeded to assess for invariance between genders by fitting multigroup CFA’s to assess for configural, metric, and scalar invariance. The results are shown in Table 4. Comparing the configural to the metric invariance model, although the Δχ2 was statistically significant, the ΔCFI was .007, under the cutoff of .01, thus supporting metric invariance. However, comparing the metric to the scalar model, the Δχ2 was statistically significant and the ΔCFI was .012, over the cutoff of .01; hence scalar invariance failed.

Table 4.

Model Fit Statistics and Comparisons of Nested Multiple Gender Group Models of the TAS-20

Invariance model χ2(df) CFI, TLI RMSEA [90% CI] SRMR BIC
Configural 1,695.23 (340) 0.746, 0.716 0.093 [0.089, 0.098] 0.088 53,558.88
Metric 1,750.52 (359) 0.739, 0.724 0.092 [0.088, 0.096] 0.097 53,481.85
Scalar 1,834.99 (379) 0.727, 0.727 0.092 [0.088, 0.096] 0.099 53,428.79
Model comparison Δχ2(df) p ΔCFI Conclusion
Config. versus metric 50.95 (19) <.001 0.007   Prefer metric, metric invariance supported
Metric versus scalar 81.90 (20) <.001 0.012   Significant fit degradation, reject scalar invariance
Open in a new tab

Note. TAS-20 = Toronto Alexithymia Scale-20; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = root-mean-square error of approximation; CI = confidence interval; SRMR = standardized root-mean-square residual; BIC = Bayes Information Criterion.

Discussion

The purpose of this study was to assess the evidence for using the total scale score of the TAS-20. Common factors, bifactor, hierarchical, and unidimensional models were estimated, and none met contemporary standards for good fit. However, the analysis of model-based dimensionality and reliability coefficients supported the unidimensionality of the raw TAS-20 total score, which can be said to represent the general alexithymia construct. Convergent and concurrent evidence for validity was found in men using latent variables. It is an important reminder that assessing validity using latent variables is a more stringent approach than using raw scores because it allows for the separation out of most sources of error. For example, in the present study, the raw-score correlations of the TAS-20 with the five validity variables were all significant; yet, only two regression coefficients were significant using the latent variable approach. Finally, the assessment of measurement invariance by gender supported metric invariance, indicating that cisgender men and women understand the scale in the same way, including the meaning of the scale scores and the distance between the scores.

Although evidence was found for modeling the TAS-20 as a unidimensional scale, and thus for the use of the total scale score, it should be noted that the fit of the unidimensional model is not optimal. We thus recommend, first, that researchers and clinicians use only the total scale score and not the subscales, and second, that the TAS-20 scale developers follow recommended procedures to trim composite measurement scales (cf., Goetz et al., 2013) in order to improve the psychometric properties of the TAS-20 and in particular the fit to the data. While we were awaiting an editorial decision on this manuscript, we became aware that the TAS-20 scale developers had very recently published an article online in which they too recommended treating the TAS-20 as a unidimensional scale:

In general, our findings were consistent with previous studies indicating that TAS-20 total scores can be considered indicative of a single construct. The replication of these earlier results from previous investigations provides additional support for the use of a total TAS-20 score and questions the utility of using TAS-20 subscale scores. Based on these results, we recommend that researchers and clinicians use a single total TAS-20 score and not subscale scores (Carnovale et al., 2021, p. 1).

We wish to acknowledge some limitations of the present study. First, this was a cross-sectional study using self-report measures. The self-report nature of the surveys introduces the possibility that respondents approached the measures in a manner that make them look more socially desirable than they are (socially desirable responding [SDR]). SDR was not measured in our study; however, SDR is not always a problem. As Tracey (2016, p. 229) put it:

Perhaps the biggest recommendation is for researchers not to implicitly assume that SDR is harmful bias. There is an extensive literature that indicates that across many domains of psychology there is no strong support for such a view ….

Second, we did not assess discriminant, predictive, and incremental evidence for validity, and test–retest evidence for reliability, which would be important tasks for future research. Third, our samples were collected from community and undergraduate participant pool subjects who self-selected to participate in this study. Recruitment via the internet or university classes also excluded potential participants of lower socioeconomic status. Further, there is less representation from communities of color with online samples. In addition, there were no transgender or nonbinary people in our sample. We know from prior research that an intensive effort, including partnering with Lesbian, Gay, Bisexual, Transgender, and Queer (LGBTQ) groups, is necessary to do this, and it should be done in future research. In effect, although participants were diverse in terms of age, they were predominantly White, Christian, educated beyond high school, cisgender, and heterosexual. Our results may thus not be representative of the general population and important differences may emerge with other samples. Additional research is certainly needed, especially work using more sophisticated sampling procedures to gather a representative sample of the United States population.

In conclusion, there is evidence for the use of the total raw (i.e., observed) scale score but not that of the subscales of the TAS-20 in both research and practice, which is important because of prior work that allowed for the establishment of criteria for diagnosing alexithymia (Taylor et al., 1988). However, in keeping with idea of “hidden invalidity” of psychological scales (Hussey & Hughes, 2020), the fit of the TAS-20 modeled as a unidimensional scale is suboptimal because of the poor fit to the data (replicating most studies on the TAS-20).

Public Significance Statement.

Evidence was provided showing that the Toronto Alexithymia Scale-20 (TAS-20) may be used as a unidimensional instrument, and that men and women understand the scale in the same way. However, the use of a TAS-20 total score may be more efficacious if the scale were trimmed.

Acknowledgments

We gratefully acknowledge the technical assistance of Dr. Joseph Hammer at the University of Kentucky, and of Eric McCurdy, MA, Kathleen Alto, MA, Zach Gerdes, MA, Stefan Jadaszewski, MA, Baron Rogers, MA, and Shana Pryor, doctoral students in counseling psychology at the University of Akron, in recruiting participants.

Footnotes

Both data sets from which this manuscript was developed were used in prior publications. One of these was a scale development article on the Normative Male Alexithymia Scale-Brief Form, which used the TAS-20 as a validity scale. Another one explored the relationships between alexithymia and muscularity and related variables. The third was a study involving evoked response potentials using the TAS-20 as a screener. None of these papers focused on the psychometric properties of the TAS.

1

Although the title references only 18 models, there was also a 19th model labeled “bifactor with negatively worded items dropped.”

2

Although the sex question included the response options “transgender” and “other—please specify,” none of the participants selected either of these options.

3

It should be noted, in terms of potential scale trimming, that the item ECV scores for the four of the five reverse-scored items were miniscule (the largest was .009), and two of them had elevated absolute relative parameter bias scores.

References

  1. Avalos L, Tylka TL, & Wood-Barcalow N (2005). The Body Appreciation Scale: Development and psychometric evaluation. Body Image, 2, 285–297. 10.1016/j.bodyim.2005.06.002 [DOI] [PubMed] [Google Scholar]
  2. Bagby RM, Parker JD, & Taylor GJ (1994). The twenty-item Toronto Alexithymia Scale—I. Item selection and cross-validation of the factor structure. Journal of Psychosomatic Research, 38, 23–32. 10.1016/0022-3999(94)90005-1 [DOI] [PubMed] [Google Scholar]
  3. Carnovale M, Taylor GJ, Parker JDA, Sanches M, & Bagby RM (2021, July). A bifactor analysis of the 20-item Toronto Alexithymia Scale: Further support for a general alexithymia factor. Psychological Assessment, 33, 619–628. 10.1037/pas0001000 [DOI] [PubMed] [Google Scholar]
  4. Chen FF (2007). Sensitivity of goodness-of-fit indexes to lack of measurement invariance. Structural Equation Modeling, 14,464–504. 10.1080/10705510701301834 [DOI] [Google Scholar]
  5. Chen FF, West SG, & Sousa KH (2006). A comparison of bifactor and second-order models of quality of life. Multivariate Behavioral Research, 41, 189–225. 10.1207/s15327906mbr4102_5 [DOI] [PubMed] [Google Scholar]
  6. Cheung GW, & Lau RS (2012). A direct comparison approach for testing measurement invariance. Organizational Research Methods, 15, 167–198. 10.1177/1094428111421987 [DOI] [Google Scholar]
  7. Cheung GW, & Rensvold RB (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9, 233–255. 10.1207/S15328007SEM0902_5 [DOI] [Google Scholar]
  8. Collins NL (1996). Working models of attachment: Implications for explanation, emotion and behavior. Journal of Personality and Social Psychology, 71, 810–832. 10.1037/0022-3514.71.4.810 [DOI] [PubMed] [Google Scholar]
  9. Collins NL, & Read SJ (1990). Adult attachment, working models, and relationship quality in dating couples. Journal of Personality and Social Psychology, 58, 644–663. 10.1037/0022-3514.58A644 [DOI] [PubMed] [Google Scholar]
  10. Dueber DM (2017). Bifactor Indices Calculator: A Microsoft Excel-based tool to calculate various indices relevant to bifactor CFA models. University of Kentucky. 10.13023/edp.tool.01 [DOI] [Google Scholar]
  11. Eng W, Heimberg RG, Hart TA, Schneier FR, & Liebowitz MR (2001). Attachment in individuals with social anxiety disorder: The relationship among adult attachment styles, social anxiety, and depression. Emotion, 1, 365–380. 10.1037/1528-3542.1.4.365 [DOI] [PubMed] [Google Scholar]
  12. Erni T, Lötscher K, & Modestin J (1997). Two-factor solution of the 20-item Toronto Alexithymia Scale confirmed. Psychopathology, 30, 335–340. 10.1159/000285079 [DOI] [PubMed] [Google Scholar]
  13. Ferguson CJ (2009). An effect size primer: A guide for clinicians and researchers. Professional Psychology: Research and Practice, 40, 532–538. 10.1037/a0015808 [DOI] [Google Scholar]
  14. Gignac GE, Palmer BR, & Stough C (2007). A confirmatory factor analytic investigation of the TAS-20: Corroboration of a five-factor model and suggestions for improvement. Journal of Personality Assessment, 89, 247–257. 10.1080/00223890701629730 [DOI] [PubMed] [Google Scholar]
  15. Goetz C, Coste J, Lemetayer F, Rat AC, Montel S, Recchia S, Debouverie M, Pouchot J, Spitz E, & Guillemin F (2013). Item reduction based on rigorous methodological guidelines is necessary to maintain validity when shortening composite measurement scales. Journal of Clinical Epidemiology, 66, 710–718. 10.1016/j.jclinepi.2012.12.015 [DOI] [PubMed] [Google Scholar]
  16. Haviland MG, & Reise SP (1996). Structure of the twenty-item Toronto Alexithymia Scale. Journal of Personality Assessment, 66, 116–125. 10.1207/s15327752jpa6601_9 [DOI] [PubMed] [Google Scholar]
  17. Hu L, & Bentler PM (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424–453. 10.1037/1082-989X.3A424 [DOI] [Google Scholar]
  18. Humphreys TP, Wood LM, & Parker JDA (2009). Alexithymia and satisfaction in intimate relationships. Personality and Individual Differences, 46, 43–47. 10.1016/j.paid.2008.09.002 [DOI] [Google Scholar]
  19. Hussey I, & Hughes S (2020). Hidden invalidity among 15 commonly used measures in social and personality psychology. Advances in Methods and Practices in Psychological Science, 3(2), 166–184. 10.1177/2515245919882903 [DOI] [Google Scholar]
  20. Jardin E, Allen PA, Levant RF, Lien M-C, McCurdy ER, Villalba A, Mallik P, Houston JR, & Gerdes ZT (2019). Event related brain potentials reveal differences in emotional processing in alexithymia. Journal of Cognitive Psychology, 31(5–6), 619–633. 10.1080/20445911.2019.1642898 [DOI] [Google Scholar]
  21. Kahn JH (2006). Factor analysis in counseling psychology research, training, and practice: Principles, advances, and applications. The Counseling Psychologist, 34, 684–718. 10.1177/0011000006286347 [DOI] [Google Scholar]
  22. Karakis EN, & Levant RF (2012). Is normative male alexithymia associated with relationship satisfaction, fear of intimacy and communication quality among men in heterosexual relationships. The Journal of Men’s Studies, 20, 179–186. 10.3149/jms.2003.179 [DOI] [Google Scholar]
  23. Kline RB (2016). Principles and practice of structural equation modeling (4th ed.). Guilford. [Google Scholar]
  24. Kooiman CG, Spinhoven P, & Trijsburg RW (2002). The assessment of alexithymia: A critical review of the literature and a psychometric study of the Toronto Alexithymia Scale-20. Journal of Psychosomatic Research, 53(6), 1083–1090. 10.1016/S0022-3999(02)00348-3 [DOI] [PubMed] [Google Scholar]
  25. Leone JE, Wise KA, Mullin EM, Harmon W, Moreno N, & Drewniany J (2015). The effects of pubertal timing and alexithymia on symptoms of muscle dysmorphia and the drive for muscularity in men. Psychology of Men & Masculinity, 16, 67–77. 10.1037/a0035920 [DOI] [Google Scholar]
  26. Levant R (1992). Toward the reconstruction of masculinity. Journal of Family Psychology, 5, 379–402. 10.1037/0893-3200.5.3-4.379 [DOI] [Google Scholar]
  27. Levant RF (2011). Research in the psychology of men and masculinity using the gender role strain paradigm as a framework. American Psychologist, 66, 765–776. 10.1037/a0025034 [DOI] [PubMed] [Google Scholar]
  28. Levant RF, Allen PA, & Lien M-C (2014). Alexithymia in men: How and when does the deficit in the processing of emotions occur? Psychology of Men & Masculinity, 15, 324–334. 10.1037/a0033860 [DOI] [Google Scholar]
  29. Levant RF, Good GE, Cook S, O’Neil J, Smalley KB, Owen KA, & Richmond K (2006). The Normative Male Alexithymia Scale: Measurement of a gender-linked syndrome. Psychology of Men & Masculinity, 7, 212–224. 10.1037/1524-9220.7A212 [DOI] [Google Scholar]
  30. Levant RF, Hall RJ, Williams C, & Hasan NT (2009). Gender differences in alexithymia: A meta-analysis. Psychology of Men & Masculinity, 10, 190–203. 10.1037/a0015652 [DOI] [Google Scholar]
  31. Levant RF, & Williams CM (2009). The psychology of men and masculinity. In Bray JH & Stanton M (Eds.), The Wiley-Blackwell handbook of family psychology (pp. 588–599). Wiley Blackwell; 10.1002/9781444310238.ch41 [DOI] [Google Scholar]
  32. Loas G, Otmani O, Verrier A, Fremaux D, & Marchand MP (1996). Factor analysis of the French version of the 20-item Toronto Alexithymia Scale (TAS-20). Psychopathology, 29, 139–144. 10.1159/000284983 [DOI] [PubMed] [Google Scholar]
  33. Mattila AK, Keefer KV, Taylor GJ, Joukamaa M, Jula A, Parker JDA, & Bagby RM (2010). Taxometric analysis of alexithymia in a general population sample from Finland. Personality and Individual Differences, 49, 216–221. 10.1016/j.paid.2010.03.038 [DOI] [Google Scholar]
  34. Meganck R, Vanheule S, & Desmet M (2008). Factorial validity and measurement invariance of the 20-item Toronto Alexithymia Scale in clinical and nonclinical samples. Assessment, 15,36–47. 10.1177/1073191107306140 [DOI] [PubMed] [Google Scholar]
  35. Muthén LK, & Muthén BO (1998–2015). Mplus user’s guide (7th ed.). [Google Scholar]
  36. Nemiah JC, Fryberger H, & Sifneos PE (1976). Alexithymia: A view of the psychosomatic process. In Hill OW (Ed.), Modern trends in psychosomatic medicine, 13, 430–439. Butterworths. [Google Scholar]
  37. Parent MC, & Bradstreet TC (2019). The strong, silent type(ology): Examining intersections of alexithymia and the drive for muscularity. The Journal of Men’s Studies, 27, 66–88. 10.1177/1060826518782976 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Parker JA, Bagby RM, Taylor GJ, Endler NS, & Schmitz P (1993). Factorial validity of the 20-item Toronto Alexithymia Scale. European Journal of Personality, 7, 221–232. 10.1002/per.2410070403 [DOI] [Google Scholar]
  39. Parker JD, Taylor GJ, & Bagby RM (2003). The 20-Item Toronto Alexithymia Scale. III. Reliability and factorial validity in a community population. Journal of Psychosomatic Research, 55, 269–275. 10.1016/S0022-3999(02)00578-0 [DOI] [PubMed] [Google Scholar]
  40. Parker JDA, Keefer KV, Taylor GJ, & Bagby RM (2008). Latent structure of the alexithymia construct: A taxometric investigation. Psychological Assessment, 20, 385–396. 10.1037/a0014262 [DOI] [PubMed] [Google Scholar]
  41. Preece D, Becerra R, Allan A, Robinson K, & Dandy J (2017). Establishing the theoretical components of alexithymia via factor analysis: Introduction and validation of the attention-appraisal model of alexithymia. Personality and Individual Differences, 119, 341–352. 10.1016/j.paid.2017.08.003 [DOI] [Google Scholar]
  42. Ravitz P, Maunder R, Hunter J, Sthankiya B, & Lancee W (2010). Adult attachment measures: A 25-year review. Journal of Psychosomatic Research, 69, 419–432. 10.1016/j.jpsychores.2009.08.006 [DOI] [PubMed] [Google Scholar]
  43. Reise SP, Bonifay WE, & Haviland MG (2013). Scoring and modeling psychological measures in the presence of multidimensionality. Journal of Personality Assessment, 95, 129–140. 10.1080/00223891.2012.725437 [DOI] [PubMed] [Google Scholar]
  44. Reise SP, Scheines R, Widaman KF, & Haviland MG (2013). Multidimensionality and structural coefficient bias in structural equation modeling a bifactor perspective. Educational and Psychological Measurement, 73, 5–26. 10.1177/0013164412449831 [DOI] [Google Scholar]
  45. Ritz T, & Kannapin O (2000). The construct validity of a German version of the Toronto Alexithymia Scale. Zeitschrift für Differentielle und Diagnostische Psychologie, 21(1), 49–64. 10.1024//0170-1789.21.1.49 [DOI] [Google Scholar]
  46. Rodriguez A, Reise SP, & Haviland MG (2016). Evaluating bifactor models: Calculating and interpreting statistical indices. Psychological Methods, 21, 137–150. 10.1037/met0000045 [DOI] [PubMed] [Google Scholar]
  47. Russell DW, Kahn JH, Spoth R, & Altmaier EM (1998). Analyzing data from experimental studies: A latent variable structural equation modeling approach. Journal of Counseling Psychology, 45, 18–29. 10.1037/0022-0167.45.1.18 [DOI] [Google Scholar]
  48. Satorra A, & Bentler PM (2001). A scaled difference chi-square test statistic for moment structure analysis. Psychometrika 66, 507–514. 10.1007/BF02296192 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Sifneos PE (1967). Clinical observations on some patients suffering from a variety of psychosomatic diseases. Acta Medicina Psychosomatica, 7, 1–10. [Google Scholar]
  50. Soper D (2013). A priori sample size calculator for structural equation models [Computer software]. https://www.danielsoper.com/statcalc/calculator.aspx?id=89
  51. Taylor GJ (2004). Alexithymia: Twenty-five years of theory and research. In Nyklicek I, Temoshok L, & Vingerhoets A (Eds.), Emotional expression and health: Advances in theory, assessment and clinical applications (pp. 137–153). Brunner-Routledge. [Google Scholar]
  52. Taylor GJ, Bagby RM, Ryan DP, Parker JDA, Doody KF, & Keefe P (1988). Criterion validity of the Toronto Alexithymia Scale. Psychosomatic Medicine, 50, 500–509. 10.1097/00006842-198809000-00006 [DOI] [PubMed] [Google Scholar]
  53. Taylor GJ, Parker JDA, Bagby RM, & Acklin MW (1992). Alexithymia and somatic complaints in psychiatric out-patients. Journal of Psychosomatic Research, 36, 417–424. 10.1016/0022-3999(92)90002-J [DOI] [PubMed] [Google Scholar]
  54. Tracey TJG (2016). A note on socially desirable responding. Journal of Counseling Psychology, 63, 224–232. 10.1037/cou0000135 [DOI] [PubMed] [Google Scholar]
  55. Tuliao AP, Klanecky AK, Landoy BVN, & McChargue DE (2020). Toronto Alexithymia Scale–20: Examining 18 competing factor structure solutions in a U.S. sample and a Philippines sample. Assessment, 27(7), 1515–1531. 10.1177/1073191118824030 [DOI] [PubMed] [Google Scholar]
  56. Tylka TL (2013). Evidence for the Body Appreciation Scale’s measurement equivalence/invariance between U.S. college women and men. Body Image, 10, 415–418. 10.1016/j.bodyim.2013.02.006 [DOI] [PubMed] [Google Scholar]
  57. Wester S, Vogel D, O’Neil JM, & Danforth L (2012). Development and evaluation of the Gender Role Conflict Scale Short Form (GRCS-SF). Psychology of Men & Masculinity, 13(2), 199–210. 10.1037/a0025550 [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES