Mardia’s test showed that the data is not multivariate normal, g1p = 34.23, χSkew = 1243.78, ppp
Bartlett’s sphericity test was significant, χ2(105) = 599.93, p
Parallel analysis with the unweighted least squares (ULS) estimator indicated that three factors should be retained (see Fig. 1). However, only two eigenvalues of the principal factors reached values greater than 1 or 0.7 (old and new Keiser criterion, respectively). Additionally, Hull’s method with CFI and RMSEA, and the lower bound of the 90% CI RMSEA, also support two-factor retention. Given these results and the conceptual framework surrounding the bifactor structure of the original scale, a two-factor solution was extracted.
An EFA with Promax rotation using the ULS estimator was performed. Respective factorial and R loadings2uniqueness and complexity per factor are shown in Table 1.
The two-factor solution accounted for 37% of the variance, with the PI factor explaining 18% and the GA factor 19% of the variance. The inter-factor correlation was 0.35. Interestingly, this factor solution mimics the two-factor solution expected and postulated in the literature (e.g., ), with the highest loading of each element saturated in the theoretically correct factor.
Items Q2, Q6, Q9 and Q13 had low loadings for the sample size (
In addition, two polytomous item response theory analyzes using a generalized partial credit model—one for each factor—were performed (see Table 2). The results showed that the five aforementioned items achieved discrimination values below acceptable (≥ 0.70) suggesting that they are not good at discriminating the latent trait and, therefore, supporting their deletion.
A CFA with WLSMV was used to confirm the 10-item bifactorial structure obtained from the EFA. The results revealed an acceptable overall fit, χ2(34) = 46.68; CFI = 0.93; TLI=0.91; RMSEA = 0.05, RMSEA 90% CI [0.00, 0.09]; SRMR = 0.07. Moreover, all items achieved high factorial weights and appropriate individual reliabilities on latent variables (see Fig. 3).
Several CFAs were also compared to check which factor structure best fits the data. In addition to the original factorial structure, and taking into account the cultural proximity, structural models from two Spanish studies evaluating the psychometric properties of the PVD were tested with our sample. All models are present in Table 3.
Models 1, 2.1 and 3.1 obtained inadequate global and local fit values. Models 2.2 and 4.1, on the other hand, obtained acceptable global fit values, but inappropriate local fit values. The remaining three models (i.e. 2.3, 3.2 and 4.2) achieved acceptable values of global and local fits. Considering the conceptual framework and the original structure of the PVD, the two-factor model obtained from the EFA (i.e. model 2.3) was adopted (see Supplementary File 2 for the original Portuguese version and end of the scale).
Convergent and discriminant validity of PVD factors
The values of the mean variance extracted (AVE) and the composite reliability (CR) for the two factors were as follows: AVEIP = 0.41, AVEGeorgia= 0.33, CRIP> 0.79, RCGeorgia> 0.74. Although these EAV values are lower than those generally considered adequate Fornell and Larcker  indicate that if AVE values are less than 0.5, but CR values are greater than 0.6, the convergent validity of the construct is still considered adequate. Additionally, both CR values are greater than 0.7, supporting the notion of proper build reliability. Thus, the convergent validity of the PVD factors was confirmed.
Moreover, both AVE values were greater than the square of the correlation between the two factors (0.12), indicating only 12.3% common information between them and confirming the discriminant validity of the two-factor model. .
Convergent and discriminant validity of the scale
Regarding convergent validity, as shown in Table 4, the two PVD subscales were significantly correlated with the DPSS-R, MOCI, MMPI-Hs and NEO-FFI Neuroticism subscales, while the DS total score -R, Core Disgust and Contamination-based Disgust subscales were only correlated with GEORGIA. Additionally, the DPSS-R Disgust Propensity subscale and MOCI were more strongly correlated with GA, and MMPI-Hs with PI, as expected.
Evidence of discriminant validity was also found, as the DS-R Animal-Reminder Disgust, SQ-R15, and NEO-FFI Extraversion and Openness subscales were not significantly correlated with the PVD factors. Similarly, the NEO-FFI subscales of Agreeableness and Conscientiousness showed weak correlations with PI and GA, respectively.
Both factors showed good levels of internal consistency Cronbach’s α ordinalIP= 0.82, G6(smc)IP= 0.81,median RIP= 0.54, Cronbach’s α ordinalGeorgia= 0.82, G6(smc)Georgia= 0.80, Median rGeorgia= 0.45.