Participants

This study initially involved 552 adults from the general Spanish population recruited by students of the Psychology Degree of the Complutense University of Madrid, who participated in a voluntary seminar on personality and post-traumatic stress disorder (PTSD). Each student applied 11 short questionnaires to six relatives, of which only three questionnaires were relevant to this study. Of that initial sample, 519 participants completed the SAS without leaving any items unanswered, comprising the final sample. These participants were between 18 and 90 years of age (M = 40.5, SD = 15.7), 58.7% were women, and 59.7% were working at the time of assessment. Concerning education, 45.8% of the participants had secondary education, 42.2% had university studies, 11.2% had primary studies, and the remaining 1% had no regulated studies. Most participants (56%) were married or living with a stable partner, 36.5% were single, 6.2% were divorced or separated, and 1.4% were widowed.

Instruments

Beck Depression Inventory-II (BDI-II; ^{Beck et al., 1996}; Spanish adaptation in ^{Beck et al., 2011}). The BDI-II is a 21-item self-report designed to assess the presence and severity of depressive symptoms. The BDI-II presents evidence of reliability and validity in its original version and its Spanish adaptation (^{Beck et al., 2011}; ^{Sanz, 2013}; ^{Sanz & García-Vera, 2013}). In the sample of participants of the present study, the BDI-II obtained a Cronbach alpha coefficient of .88.

PTSD Checklist for DSM-5 (PCL-5; ^{Weathers et al., 2013}; Spanish adaptation of ^{Sanz et al., 2023}, unpublished manuscript). The PCL-5 is a 20-item instrument that assesses the presence and severity of PTSD symptoms according to the DSM-5. Both its original version and its Spanish adaptation present evidence of reliability, convergent validity and diagnostic validity (^{Blevins et al., 2015}; ^{Sanz et al., 2023}, unpublished manuscript). Participants completed the PCL-5 concerning the most traumatic event they had experienced, and their scores had a Cronbach alpha coefficient of .94.

Self-Absorption Scale (SAS; ^{McKenzie & Hoyle, 2008}). We used the Spanish version of ^{Perona-Garcelán et al. (2013}, ²⁰¹⁴⁾, which replaces the original instructions that ask about the degree to which the persons consider the behaviors presented in its 17 items characteristic of themselves with instructions that ask about the frequency with which these behaviors have happened to them. The SAS provides scores of a total scale and two subscales, one that evaluates the Private Self-absorption factor (8 items) and the other that evaluates the Public Self-absorption factor (9 items). All three scores have shown adequate indices of reliability and validity in different populations (^{DaSilveira et al., 2011}; ^{McKenzie & Hoyle, 2008}).

Procedure

Each participant signed an informed consent to collaborate in a broader investigation on personality and PTSD. Subsequently, the student who had invited the participant applied several short questionnaires in this order: the BDI-II, a dysfunctional attitudes questionnaire, an optimism questionnaire, the SAS, a meta-cognition questionnaire, a traumatic events questionnaire, and the PCL-5. The students were rewarded with a subject's grade, but the participants did not receive any compensation. The first two authors of the study taught a seminar to train students in the questionnaires' content, characteristics and mode of application.

Data Analysis

An exploratory factor analysis (EFA) was performed on the scores in the SAS items of a first subsample (Subsample 1) comprising 50% of the cases of the total sample, leaving a second subsample with the remaining cases (Subsample 2) for cross-validation with a confirmatory factor analysis (CFA). The Solomon method was used to divide the sample, carried out with the FACTOR program, v. 12.01.02 (^{Ferrando & Lorenzo-Seva, 2017}) because, compared to random partitioning, it optimizes the equivalence of the subsamples in terms of the amount of their common variance. We also calculated the commonality ratio index S, which indicates the degree of equivalence obtained (^{Lorenzo-Seva, 2021}).

Given that, in Subsample 1, 14 of the 16 items of the SAS presented kurtosis or skewness values higher in absolute value than the range of values indicative of a normal distribution of scores (± 1), the EFA was performed on the matrix of polychoric correlations, as it is usually more appropriate when a high percentage of items do not meet the assumption of normality. The Bartlett sphericity test and the Kaiser-Meyer-Olkin (KMO) test were calculated to analyze the adequacy of the data to factor analysis. Five procedures were used to determine the number of factors to be extracted: Cattell's scree plot, Hull's method, Velicer's MAP test, classical parallel analysis, and Timmerman and Lorenzo-Seva's optimal parallel analysis. We extracted as many factors as recommended by most of these procedures, using the unweighted least squares (ULS) estimation method because it does not assume a multivariate normal distribution of the data.

For each recommended factorial solution, we calculated the following goodness-of-fit indices, with their corresponding criteria for an adequate fit in parenthesis (^{West et al., 2012}): 1) χ²/df (≤ 5); 2) goodness of fit index or GFI (≥ .95); 3) Bentler's comparative fit index or CFI (≥ .95); 4) non-normalized fit index or NNFI (≥ .95), and 5) root mean square error of approximation or RMSEA (≤ .08). In the case of extracting a single factor, three additional fit indices were calculated for a one-dimensional solution: unidimensional congruence index or UniCo (> .95), percentage of explained common variance or ECV (> .85), and mean of item residual absolute loadings or MIREAL (< .30).

We assessed the results of these indices in the context of the psychological interpretation of the factorial loading matrix of the different factorial solutions. In the case of the multifactor solutions, the matrix was rotated with a promin oblique procedure. Following the recommendations of ^{Lloret-Segura et al. (2014)} , for the psychological interpretation, we considered the content of the items that presented factorial loadings ≥ .40 in one factor and lower loadings in the rest.

The factorial solution or solutions of the SAS found in Subsample 1 were validated in Subsample 2 with CFA, performed with the JASP program (^{JASP Team, 2020}). For the CFA, the polychoric correlation matrix, the diagonally weighted least squares (DWLS) method and the aforementioned goodness-of-fit indices were used.

If multifactorial solutions were obtained, we performed Schmid-Leiman analyses with FACTOR on the total sample. We calculated the hierarchical omega coefficients (ωH) with the formulas of ^{Rodriguez et al. (2016)} to obtain, in a hierarchical model, the proportion of variance of the items that a general factor would explain and the variance that the lower-order factors would explain, controlling for the variance due to the general factor.

Finally, for the total scale of the SAS and the subscales defined by the most appropriate factorial solution, we examined the internal consistency using Cronbach's alpha and ^{McDonald's (1999)} omega coefficients calculated with JASP, the ordinal alpha coefficient calculated with CATPCA of the SPSS program, and the ordinal omega coefficient calculated with FACTOR. We also calculated the means and standard deviations of the SAS items, the correlations of their items with the scores of the total scale or with the subscales without the concurrence of the corresponding item (item-total and item-subscale corrected correlations), as well as the correlations of the total scale and the subscales with the measures of depression and post-traumatic stress symptomatology.

Evidence of Internal Structure

For Subsample 1, the results of Bartlett's sphericity test (2220, p < .001) and the KMO (.83) indicated that the matrix of polychoric correlations was adequate for EFA. The scree plot (Figure 1), the two parallel analyses and the MAP test indicated that two factors should be extracted in Subsample 1, but Hull's method indicated only one. Consequently, one and two factors were extracted in Subsample 1 to compare their goodness-of-fit indices and their psychological interpretation. These indices are shown in Table 1.

Figure 1. Cattell Scree Test of the Matrix of Polychoric Correlations Between the Items of the SAS in Subsample 1 

Both the unifactorial and bifactorial solutions obtained good fit indices, as all five general fit indices showed acceptable or good values (Table 1). However, two of the three indices that specifically evaluated the fit of a single-factor solution (UniCo and ECV) suggested that this solution did not fit the data well. The matrix of factorial loadings of the unifactorial solution indicated that all the items of the SAS loaded more than .40 in this solution, except for Item 4, which showed a factorial loading of .35 (Table 2). This pattern of factor loadings corroborated the good general fit indices of the unifactorial solution, but the pattern of factorial loadings of the rotated bifactorial solutions of Table 2 corroborated the good general fit indices of the bifactorial solution and the psychological coherence of this solution with the theory underlying the SAS. Thus, as originally intended by the authors of the SAS, Items 1, 2, 3, 7, 8, 13, 16, and 17 presented relevant factorial loadings (> .40) on the same first factor, which could be considered public self-absorption, and irrelevant factorial loadings, even less than .30, on the second factor. In addition, Items 4, 5, 6, 9, 11, 12 and 15 presented relevant factorial loadings on the second factor, which could be considered private self-absorption, and factorial loadings of less than .30 on the first factor.

In summary, from a statistical point of view, both the unifactorial and bifactorial solutions seemed adequate in Subsample 1. However, the bifactorial solution appeared to be slightly better and conformed to the distinction between types of self-absorption postulated by the creators of the SAS, namely private and public self-absorption. Therefore this latter solution seemed the most appropriate from a psychological point of view. The two self-absorption factors had a high and statistically significant correlation, with a value of .69 (p < .001). This high correlation justified performing a second-order EFA. In this analysis, the two factors showed very high factor loadings on a higher-order factor, specifically, loadings of .918 and .748 for Public and Private self-absorption factors, respectively.

However, the bifactorial solution could represent a symmetrical bifactor model instead of a hierarchical model of two correlated factors that would reflect the existence of a general second-order factor. In a symmetrical bifactor model, the variance observed in the SAS would be decomposed into one part that is common to all the SAS items-that is, a general factor-and another part that is specific to different groups of items-in this case, two specific groups corresponding to the Public and Private Self-absorption factors.

To answer this question, CFAs were performed in Subsample 2 in which three models were tested: unifactorial, hierarchical with two primary factors and one second-order factor and symmetric bifactor model. The goodness-of-fit indices of the three models were generally adequate (Table 3), although better for the second-order model and the bifactor model, especially the latter. In fact, when comparing the three models through analysis of structural equation models performed with JASP, a statistically significant reduction was found in χ² of the bifactor model compared with the unifactorial model (Δχ² = 187.06, p < .001) and the second-order model (Δχ² = 103.27, p < .001), suggesting that the bifactor model had a more appropriate fit to the data.

In summary, the results of the factor analyses indicated that the SAS had a symmetrical bifactor structure of two specific factors and one general factor. The results of the Schmid-Leiman analysis performed with the total sample and the hierarchical omega coefficients obtained from these results are presented in Table 4. These coefficients revealed that the general factor explained 78.6% of the variance of the items, whereas, controlling for the variance due to the general factor, the Public self-absorption factor explained 12.7% of the variance of its items, and the Private self-absorption factor explained 36.4% (Table 4). These results support the presence of a relatively strong overall Self-absorption factor, as it explained more than 70% of the variance of the items, with an explained common variance greater than .70 (ECV = .73). In addition, in the Schmid-Leiman analysis, 13 of the 16 items loaded higher on the general Self-absorption factor than on their respective specific factors (Table 4). This pattern of results was also replicated, in general, when performing the Schmid-Leiman analyses in 12 subsamples created, two of them with the SOLOMON method and the remaining 10 with the random method of the SPSS. In 10 of these 12 subsamples, the general factor explained more than 70% of the variance of the items -approximately, an average of 76.8%-, the Public Self-absorption factor, an average of 13.6%, and the Private Self-absorption factor, an average of 37.3%.

Evidence of Internal Consistency

The finding of a hierarchical bifactorial structure with a higher-order factor empirically supported obtaining and validating a total SAS score and the scores for the two SAS subscales initially proposed by the instrument's authors. The results of the internal consistency analyses of all these scores revealed that, according to the standards of ^{Hernández et al. (2016)}, the internal consistency coefficients were excellent (alpha/omega ≥ .85) for the total scale and the Public Self-absorption subscale, and good (.80 ≤ alpha/omega < .85) or adequate (.70 ≤ alpha/omega < .80) for the Private Self-absorption subscale (Table 5).

Table 6 shows, for each item of the SAS, its mean, standard deviation, and item-total and item-subscale correlations for the total sample of participants. The item-total and item-subscale correlations indicated good internal consistency indices for all the items because they exceeded in all cases the value of .30, except for Item 4, whose item-total correlation was .28, although its item-subscale correlation (.31) did exceed the standard of .30.

Evidence of Validity Based on Relationship with Other Variables

In the total sample, the mean score on the measure of depressive symptoms (BDI-II) was 9.34 (SD = 7.51), and on the measure of post-traumatic stress symptoms (PCL-5), it was 14.25 (SD = 14.71). Statistically significant and moderate correlations (.30 ≤ r < .50) were found between the SAS measures and measures of depression and post-traumatic stress symptoms (Table 7). According to the standards of ^{Hernández et al. (2016)} , these correlations are considered good evidence of validity for the total measure of self-absorption of SAS (.45 ≤ r < .55) and, in general, adequate evidence of validity for the measures of private and public self-absorption of the SAS (.35 ≤ r < .45).