Toronto Alexithimia Scale (TAS-20)
แบบวัดภาวะยากลำบากในการบอกหรือแสดงอารมณ์ของโตรอนโต
ฉบับภาษาไทย (พ.ศ. ๒๕๕๓) TAS-20 Thai.pdf
Rasch Validation of the Thai Toronto Alexithymia Scale in a Thai Sample
Abstract
This study evaluated the psychometric properties of the 20-item Thai Toronto Alexithymia Scale (Thai TAS) using Rasch rating-scale analysis. Data were obtained from 660 respondents who completed the 20-item scale with five response categories, resulting in 13,200 complete observations. The analysis examined person and item reliability, person–item targeting, category functioning, item fit, item polarity, dimensionality, local dependence, and global model fit.
Person reliability was acceptable (real reliability = .72; separation = 1.59), supporting use of the scale for group-level research and preliminary screening. Internal consistency was also acceptable (Cronbach’s alpha = .75). Item reliability was excellent (1.00), with item separation of 14.35, indicating a stable hierarchy of item endorsement difficulty. The Wright map showed imperfect person–item targeting, as the mean person measure (−0.94 logits) was below the item mean (0 logits), suggesting limited precision for respondents with very low alexithymia.
Using the predefined fit criterion of 0.50–1.50 for infit and outfit mean-square statistics, 18 of 20 items demonstrated acceptable fit. TAS12 showed substantial misfit (infit/outfit MNSQ = 1.83), while TAS16 showed marginal outfit misfit (MNSQ = 1.52) and the lowest point-measure correlation (.20). All items had positive point-measure correlations, indicating correct scoring direction. The five response categories had ordered thresholds; however, Category 5 was infrequently selected (5%) and showed outfit misfit (MNSQ = 1.67), suggesting weak differentiation between the two highest categories.
PCA of standardized residuals supported a dominant general alexithymia dimension, with Rasch measures explaining 37.2% of total raw variance and essential unidimensionality of 61.1%. However, strict unidimensionality was not fully supported because the first residual contrast had an eigenvalue of 4.00 and explained 12.5% of variance. Local dependence was observed, particularly between TAS10R and TAS19R (r = .44) and between TAS1 and TAS2 (r = .41).
Overall, the Thai TAS demonstrated acceptable measurement properties for use as a broad total-score measure of alexithymia in group-level research. Nevertheless, revision or further evaluation of TAS12 and TAS16, investigation of reverse-wording and local-dependence effects, and testing of a collapsed upper response category are recommended before use for individual clinical classification or high-stakes decisions.
Sample and model
A total of 660 university respondents were included in the study and completed the 20-item Thai Toronto Alexithymia Scale (Thai TAS). Age data were available for 653 respondents; seven respondents (1.1%) had missing age information. The mean age was 22.65 years (SD = 1.30), ranging from 21 to 37 years.
Of the 660 respondents, 354 (53.6%) were female and 306 (46.4%) were male. Therefore, the sample had a relatively balanced sex distribution, with slightly more female than male participants.
A Rasch rating-scale analysis was conducted on the full 20-item Thai Toronto Alexithymia Scale (Thai TAS) using responses from 660 participants. Each item had five response categories, producing 13,200 observations (660 × 20). There were no missing data. Two participants (0.3%) obtained minimum extreme scores; therefore, 658 non-extreme respondents contributed to the main fit estimation.
The analysis evaluated person and item measurement quality, person–item targeting, response-category functioning, item fit and polarity, dimensionality, local dependence, and global model fit.
1. Person and Item Measurement Quality
1.1 Person measurement
The average raw score was 41.9 across 20 items, and the mean Rasch person measure was −0.94 logits. The person-measure standard deviation was 0.57 logits, with person measures ranging from −5.29 to 0.83 logits.
Because the item mean was set at 0 logits, the negative mean person measure indicates that respondents, on average, had lower alexithymia levels than the average item endorsement level.
Person reliability and separation
|
Statistic
|
Real estimate
|
Model estimate
|
|
RMSE
|
.30
|
.28
|
|
True SD
|
.48
|
.50
|
|
Separation
|
1.59
|
1.82
|
|
Reliability
|
.72
|
.77
|
The real person reliability was .72, and the person separation index was 1.59. This indicates acceptable measurement precision for group-level research. The scale can distinguish approximately two to three broad strata of alexithymia:

Thus, the Thai TAS can broadly distinguish lower from higher alexithymia, but it does not have sufficient precision for high-stakes individual diagnosis or fine-grained person classification.
Cronbach’s alpha/KR-20 was .75, with a standard error of measurement of 4.26 raw-score points. This indicates acceptable internal consistency.
The correlation between raw total scores and Rasch person measures was .94, indicating that both scoring methods rank participants similarly. Nevertheless, Rasch person measures are preferable when available because they account for item difficulty and provide interval-like logit estimates.
1.2 Item measurement
The average item measure was fixed at 0 logits, as expected in Rasch calibration. Item measures ranged from −1.52 logits to +1.47 logits, indicating an item-difficulty range of approximately 2.99 logits.
|
Item measurement indicator
|
Result
|
|
Number of items
|
20
|
|
Mean item measure
|
0.00 logits
|
|
Item-measure SD
|
.73 logits
|
|
Easiest item
|
TAS16, −1.52 logits
|
|
Most difficult item
|
TAS3, +1.47 logits
|
|
Real item separation
|
14.35
|
|
Real item reliability
|
1.00
|
|
Model item separation
|
15.30
|
|
Model item reliability
|
1.00
|
Item separation was 14.35 and item reliability was 1.00, indicating excellent precision and a very stable hierarchy of item endorsement difficulty. The sample size was sufficient to estimate item locations accurately.
2. Person–Item Targeting: Wright Map
The Wright map showed that the mean person measure (−0.94 logits) was lower than the mean item measure (0 logits). Therefore, the Thai TAS was not perfectly targeted to this sample.
Most respondents were located approximately between −1.2 and −0.2 logits, while many items were concentrated between approximately −0.6 and +0.4 logits. Thus, the scale was most informative for respondents in the middle range of alexithymia.
At the lower end of the continuum, the easiest item was TAS16 at −1.52 logits. However, some respondents had measures far below this point, extending to −5.29 logits. There were no items available to distinguish among individuals with very low alexithymia. This suggests reduced measurement precision at the lower end.
At the upper end, the most difficult items were TAS3 (+1.47 logits), TAS13 (+1.16 logits), and TAS14 (+1.13 logits). The highest observed person measure was +0.83 logits. Thus, the most difficult items were above the alexithymia level of almost all respondents in this sample. These items may be more informative in a clinical sample with higher alexithymia levels.
Overall, the scale was best targeted to moderate levels of alexithymia and showed limited coverage of very low alexithymia levels.
3. Rating-Scale Category Functioning
The Thai TAS used five response categories. All categories were used by respondents.
|
Category
|
Frequency
|
Percentage
|
Infit MNSQ
|
Outfit MNSQ
|
|
1
|
5,380
|
41%
|
.91
|
.97
|
|
2
|
3,759
|
28%
|
1.05
|
.80
|
|
3
|
2,201
|
17%
|
.91
|
.81
|
|
4
|
1,147
|
9%
|
.97
|
1.02
|
|
5
|
713
|
5%
|
1.44
|
1.67
|
The observed average person measures increased monotonically across the response categories:

This indicates that higher category selections were generally associated with higher alexithymia levels.
The Andrich thresholds were also ordered correctly:

Therefore, the rating scale showed no threshold disordering.
However, Category 5 showed weaker functioning than the other categories:
- It was selected in only 5% of responses.
- It had outfit misfit (MNSQ = 1.67).
- Its category-to-measure coherence was only 2%.
- Its estimated discrimination was .19.
- The threshold distance between Categories 4 and 5 was only .21 logits.
- The observed average measure differed by only .05 logits between Categories 4 and 5.
These findings suggest that respondents may not clearly distinguish between Categories 4 and 5. The five-category format can be retained provisionally because thresholds were ordered, but collapsing Categories 4 and 5 should be evaluated in future studies.
4. Item Fit and Item Polarity
Item fit was evaluated using infit and outfit mean-square statistics. The selected acceptable range was 0.50–1.50.
The mean item infit MNSQ was 1.06, and the mean outfit MNSQ was 1.03, indicating acceptable overall item-model fit.
4.1 Misfitting items
|
Item
|
Infit MNSQ
|
Outfit MNSQ
|
PTMEA correlation
|
Interpretation
|
|
TAS12
|
1.83
|
1.83
|
.38
|
Clear infit and outfit misfit
|
|
TAS16
|
1.32
|
1.52
|
.20
|
Marginal outfit misfit and weak item polarity
|
TAS12
TAS12 showed substantial misfit, with infit and outfit MNSQ values of 1.83. This indicates that respondents answered the item more unpredictably than expected by the Rasch model. Its observed exact match was 23.1%, lower than the expected exact match of 35.7%.
Although TAS12 had a positive point-measure correlation (.38), its fit results indicate that it may not operate consistently with the dominant general alexithymia dimension. TAS12 should be reviewed for wording, translation accuracy, cultural relevance, ambiguity, and theoretical alignment.
TAS16
TAS16 had marginal outfit misfit (1.52) and the lowest point-measure correlation of all items (.20, expected = .48). Thus, TAS16 was weakly related to the overall alexithymia measure.
TAS16 was the easiest item to endorse and contributed to low-end trait coverage. However, it should not be retained only because of its location on the Wright map. Its wording, translation, cultural interpretation, and construct relevance should be reviewed.
4.2 Items requiring monitoring
Some items fitted within the selected MNSQ range but had relatively high fit values or low point-measure correlations.
|
Item
|
Infit
|
Outfit
|
PTMEA
|
|
TAS20
|
1.28
|
1.39
|
.28
|
|
TAS3
|
1.33
|
1.00
|
.41
|
|
TAS10R
|
1.18
|
1.21
|
.22
|
|
TAS15
|
1.03
|
1.07
|
.28
|
|
TAS19R
|
1.05
|
1.06
|
.29
|
TAS20 and TAS10R had low point-measure correlations and should be monitored. TAS10R is reverse-scored and may be influenced by reverse-wording or translation effects.
4.3 Item polarity
All 20 items had positive point-measure correlations, ranging from .20 to .59. Therefore, all items were scored in the intended direction and none showed negative polarity.
The strongest item–measure relationships were observed for:
- TAS2: .59
- TAS9: .56
- TAS1: .55
- TAS7: .54
- TAS11: .54
- TAS13: .51
- TAS4R: .51
5. Dimensionality
PCA of standardized residuals was conducted to evaluate unidimensionality.
|
Indicator
|
Result
|
|
Raw variance explained by Rasch measures
|
37.2%
|
|
Essential unidimensionality
|
61.1%
|
|
First residual contrast eigenvalue
|
4.00
|
|
First contrast variance
|
12.5%
|
|
Second contrast eigenvalue
|
2.13
|
|
Third contrast eigenvalue
|
1.41
|
|
Fourth contrast eigenvalue
|
1.17
|
|
Fifth contrast eigenvalue
|
1.04
|
The Rasch dimension explained 37.2% of raw variance, supporting the presence of a meaningful general alexithymia dimension. The essential unidimensionality value of 61.1% also indicates that the general Rasch dimension was dominant.
However, strict unidimensionality was not fully supported. The first residual contrast had an eigenvalue of 4.00 and accounted for 12.5% of total variance. These values exceed commonly used strict screening guidelines.
The primary Rasch dimension was approximately three times larger than the first residual contrast:

Therefore, the Thai TAS can be described as essentially or predominantly unidimensional, rather than strictly unidimensional. A general total score can be used for group-level research, but it should be interpreted as a broad indicator of alexithymia rather than a perfectly pure single construct.
Possible sources of residual multidimensionality include:
- distinct alexithymia domains;
- reverse-wording effects;
- translation or cultural interpretation effects;
- local dependence;
- overlapping item content.
6. Local Dependence
Local independence requires that, after controlling for alexithymia, responses to different items should not remain strongly related. The largest standardized residual correlations were:
|
Item pair
|
Residual correlation
|
Interpretation
|
|
TAS10R – TAS19R
|
.44
|
Strongest local-dependence signal
|
|
TAS1 – TAS2
|
.41
|
Potential local dependence
|
|
TAS2 – TAS9
|
.38
|
Moderate local dependence
|
|
TAS5R – TAS10R
|
.34
|
Possible local dependence
|
|
TAS7 – TAS13
|
.34
|
Possible local dependence
|
|
TAS7 – TAS9
|
.33
|
Possible local dependence
|
|
TAS1 – TAS9
|
.32
|
Possible local dependence
|
|
TAS13 – TAS14
|
.31
|
Possible local dependence
|
|
TAS18R – TAS19R
|
.31
|
Possible local dependence
|
The strongest residual correlation was between TAS10R and TAS19R (.44). Because both items are reverse-scored, this may reflect a reverse-wording method effect rather than only substantive alexithymia variation.
The correlations among TAS1, TAS2, and TAS9 suggest possible content overlap or a narrow item cluster. Their lower MNSQ values also suggest highly predictable, potentially redundant response patterns.
The presence of local dependence may inflate reliability estimates because dependent items provide partly overlapping information.
7. Global Model Fit
|
Global statistic
|
Result
|
|
Equal-item-discrimination test
|
χ²(19) = 310.48, p < .001
|
|
van den Wollenberg Q1 test
|
Q1(285) = 1219.17, p < .001
|
|
Log-likelihood chi-square
|
30,380.54; df = 13,138; p < .001
|
|
Pearson global chi-square
|
13,561.87; df = 13,138; p < .001
|
|
Pearson chi-square/df ratio
|
1.03
|
|
Global RMS residual
|
.936
|
|
Expected RMS residual
|
.911
|
|
Standardized residual mean
|
−.024
|
|
Standardized residual SD
|
1.015
|
The equal-discrimination and Q1 tests were statistically significant, suggesting that the items did not have perfectly equal discrimination or fully parallel response functions, as expected in a strict Rasch model.
However, these tests are sensitive to the large number of observations. The Pearson chi-square/df ratio was close to 1.00, standardized residuals approximated the expected normal distribution, and the global RMS residual was only slightly higher than expected.
Therefore, the full 20-item Thai TAS demonstrated reasonable but imperfect overall Rasch-model fit. The observed deviations are consistent with the misfit of TAS12 and TAS16, local dependence among selected item pairs, residual multidimensionality, and weak Category 5 functioning.
8. Use of the Total Score
The results support use of the 20-item total score as a broad indicator of general alexithymia, with clear qualifications.
The total score is appropriate for:
- group-level research;
- descriptive reporting;
- comparison of mean alexithymia levels between groups;
- correlation, regression, and predictive analyses;
- preliminary non-diagnostic screening.
The total score should be used cautiously for:
- individual-level interpretation;
- fine distinctions between respondents;
- creation of new clinical cut-off scores;
- diagnostic or high-stakes decisions.
The justification for retaining the total score is that the Thai TAS showed a dominant general alexithymia dimension, acceptable reliability, positive item polarity, and generally satisfactory item fit. However, strict unidimensionality was not achieved, and local dependence and item-level concerns remain.
For routine research, the raw total score can be calculated by summing all 20 item responses after reverse scoring the reverse-worded items. If the categories are coded from 1 to 5, the possible total-score range is:

For advanced psychometric analyses, Rasch person measures may be preferred because they account for item difficulty and provide individual standard errors.
9. Discussion
The full 20-item Thai TAS showed acceptable psychometric performance for group-level research. Person reliability (.72) and Cronbach’s alpha (.75) indicate acceptable internal consistency. Item reliability was excellent (1.00), showing that item difficulties were estimated very precisely.
The main limitation was imperfect targeting. Respondents, on average, had lower alexithymia than the average item endorsement level. The scale was most effective in the middle range of alexithymia and less precise for very low alexithymia levels. Future research may consider developing or adapting additional well-functioning items that target lower levels of alexithymia.
Although most items fitted the model, TAS12 and TAS16 require further evaluation. TAS12 was the clearest problematic item because it showed both infit and outfit misfit. TAS16 had marginal outfit misfit and the weakest point-measure correlation. These items may reflect translation issues, cultural interpretation differences, ambiguous wording, or imperfect alignment with the general alexithymia construct.
The five response categories were ordered, which supports the existing response structure. However, Category 5 was rarely used and poorly differentiated from Category 4. This suggests that a four-category response scale, created by combining Categories 4 and 5, should be examined in subsequent validation studies.
The evidence supports essential rather than strict unidimensionality. This is not unexpected because alexithymia is often conceptualized as having related facets. The first residual contrast may reflect the theoretical components of alexithymia, reverse-wording effects, or local dependence. The total score remains useful as a broad overall measure, but the scale should not be presented as measuring a perfectly homogeneous single trait.
10. Recommendations
- Use the total score for group-level research.
The 20-item Thai TAS total score may be used as a broad indicator of alexithymia, provided that limitations are reported.
- Review TAS12 as the highest priority.
Conduct expert review and cognitive interviews to examine wording, translation equivalence, clarity, cultural relevance, and theoretical alignment.
- Review TAS16.
Its weak item polarity and marginal outfit misfit warrant review, even though it contributes to lower-end targeting.
- Examine reverse-worded items.
Review TAS10R and TAS19R for possible reverse-wording or translation-related method effects.
- Evaluate collapsing Categories 4 and 5.
Re-run the Rasch analysis using a four-category scoring structure and compare category fit, reliability, item fit, dimensionality, local dependence, and targeting.
- Investigate local dependence using testlets.
Consider theoretically justified testlets for TAS10R–TAS19R and TAS1–TAS2, with possible inclusion of TAS9 if supported by content review.
- Evaluate the theoretical domain structure.
Use confirmatory factor analysis, bifactor analysis, or subscale-specific Rasch analyses to examine whether domain scores provide useful additional information.
- Replicate in other Thai populations.
Include clinical samples and groups with higher levels of alexithymia to evaluate item targeting and functioning of Category 5.
- Conduct differential item functioning analysis.
Test whether items operate equivalently across sex/gender, age, education, clinical status, and other relevant groups.
- Avoid using unvalidated diagnostic cut-offs.
Any clinical cut-off should be validated against external clinical criteria before use.
11. Conclusion
The full 20-item Thai TAS demonstrated acceptable reliability and generally satisfactory Rasch measurement properties in 660 respondents. Person reliability was .72, Cronbach’s alpha was .75, and item reliability was 1.00. Eighteen of 20 items showed acceptable fit to the Rasch model. All items had positive point-measure correlations, and the five response categories were ordered appropriately.
However, the scale did not fully meet the expectations of a strict Rasch rating-scale model. TAS12 showed substantial misfit, TAS16 showed marginal outfit misfit and weak alignment with the overall trait, Category 5 functioned weakly, and several item pairs demonstrated local dependence. PCA residual analysis supported a dominant general alexithymia dimension but not strict unidimensionality.
Therefore, the Thai TAS total score may be used as a broad measure of general alexithymia for group-level research and preliminary screening. Interpretation should remain cautious, particularly for individual-level decisions. Future work should focus on reviewing TAS12 and TAS16, improving response-category functioning, investigating reverse-wording and local-dependence effects, and replicating the analysis in diverse Thai populations.
Note: Although TAS12 and TAS16 showed item-level concerns, immediate deletion of both items was not considered warranted. TAS12 demonstrated substantial infit and outfit misfit and was identified as the strongest candidate for revision or removal in future refinement. TAS16 showed only marginal outfit misfit but had weak point-measure correlation; because it was the easiest item to endorse and contributed to measurement at lower alexithymia levels, it was retained provisionally. The original 20-item total score was used for the primary analyses to preserve comparability with the established instrument. Sensitivity analyses using a 19-item score excluding TAS12, and an 18-item score excluding TAS12 and TAS16, are recommended to assess the robustness of substantive findings.