Psychometric properties of the Brief Symptom Inventory support the hypothesis of a general psychopathological factor

Abstract Introduction The existence of a general factor related to psychiatric symptoms is supported by studies using a variety of methods in both clinical and non-clinical samples. Objectives This study aims to evaluate the replicability of the internal structure of the Brief Symptom Inventory in a large Brazilian sample. Methods Participants were 6,427 Brazilian subjects (81% female). Mean age was 42.1 years (standard deviation [SD] = 13.6, Min = 13, Max = 80). All participants completed the online version of the Brief Symptom Inventory. This scale presents a general score (GSI) and nine specific clusters of symptoms (depression, anxiety, phobic anxiety, interpersonal sensibility, psychoticism, paranoid ideation, obsessive-compulsive behavior, hostility, and somatization symptoms). Results Confirmatory factor analysis was performed to assess the factor structure of the BSI. The results showed that the best-fitting model was a bifactor solution and the general factor was the main dimension explaining most of the reliable variability in the data. Conclusion The findings suggest that the BSI’s internal structure was replicated in a non-clinical sample and that the general factor is the most reliable score. However, it is necessary to better understand the meaning of the general factor scores in a non-clinical sample to increase interpretability of scores.


Introduction
The frontiers between psychiatric illnesses are much less established than those conceived in the diagnostic manuals of mental disorders. The lack of precise boundaries between mental illnesses has modified the process of psychiatric diagnosis, with a gradual change from a categorical perspective to a dimensional one. 1 Current categories of mental disorders are highly comorbid with each other and this phenotypic covariance should not be neglected in clinical practice. Some authors even argue that the symptomatic similarity between patients with specific disorders suggests a shared common core between mental disorders. 2 Their study found a modified bifactor model with three correlated specific factors (internalizing, externalizing, thought disorder) and one general psychopathology factor, the "p" factor. The "p" factor has since been corroborated by studies with children and adolescents 6 and with adults, 2 a pattern that is likely to be stable over time. 6,7 In effect, if the "p" factor is not an erroneous finding, it will consistently appear in the psychometric modeling of instruments measuring different types of psychopathological dimensions. Therefore, as in studies concerning constructs like the general intelligence factor, also called "g," psychometric analyses of instruments evaluating psychiatric disorders or psychological distress are a useful way to assess the hypothesis of the existence of a "p" factor.
The Brief Symptom Inventory (BSI) 8 is a selfreport instrument developed to assess psychological distress and psychopathological symptoms in nine dimensions: depression, anxiety, somatization, obsession-compulsion, interpersonal sensitivity, phobic anxiety, hostility, paranoid ideation, and psychoticism.
The inventory also produces a Global Severity Index (GSI), which includes all symptoms assessed by the scale. The scale was developed before the third edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-III) 9 and has remained in use for clinical and research purposes with psychiatric patients and in non-clinical samples. 10,11 The BSI has been translated into several languages in the decades since its development. [12][13][14][15][16][17][18] The nine-factor structure of the BSI has been replicated through confirmatory and exploratory factor analysis in samples from countries like Italy 19 and Azerbaijan, 20 but a unidimensional factor structure has also been found in other countries, like the United States 21 and Greece. 22 These inconsistencies are usually due to the communality of items measuring psychiatric symptoms, which could vary between samples from different cultures. Thomas  female) participants self-reported previous lifetime psychiatric diagnoses for at least one condition. All the participants included in the analyses gave their consent and had at least one valid answer for a BSI item.

Procedure
Participants were recruited via the internet with a social media campaign run from May to June 2020, using a snowball sampling procedure. The SurveyMonkey platform delivered all the questionnaires. Participants gave informed consent before starting to answer the tests and questionnaires. Ethical procedures were approved by the National Research Ethics Commission (process CAAE 30823620.6.0000.5149) and comply with the Helsinki Declaration.

Measure
The BSI is a 53-item instrument designed to identify relevant psychological symptoms. 8  A confirmatory factor analysis was conducted with the lavaan package 26 in R software 27 using weighted least squares mean and variance adjusted estimation with Satorra-Bentler correction, to correct the standard errors and chi-square estimates. 28 Global model fit was evaluated using the comparative fit index (CFI), Tucker-Lewis index (TLI), and the root mean square error of approximation (RMSEA). To interpret model fit, values equal to or greater than 0.95 for CFI and TLI, and equal to or less than 0.05 for RMSEA were considered acceptable. 29 The quality of the models was verified using several indices. The H index was developed to evaluate construct replicability, measuring the degree to which the indicators appropriately represent the latent variables.
A threshold of 0.70 is generally accepted as a criterion for this index. 30 Omega (ω) and omega hierarchical (ωH) coefficients were calculated. The omega hierarchical coefficient is useful for bifactor models for assessing the percentage of common variance attributable to the general factor. Reise et al. 31 argue that the higher the omega hierarchical value, the higher the relevance of the general factor to explain the variance of the data. In that case, the general factor could reflect an essentially unidimensional structure that explains the variance in respondents' scores.
To evaluate the unidimensionality of the factors, explained common variances were calculated for general (ECV), specific (ECV_SG and ECV_GS), and item levels (I-ECV). The ECV index evaluates the proportion of common variance explained by the general factor.
The ECV_SG and ECV_GS indicate common variance explained related to specific factors and the variance in each factor due to the general factor, respectively. This indicates the proportion of the items' variance that could be explained by the general factors. 30 The percentage of uncontaminated correlations (PUC) specifies the possible data bias of interpreting multidimensional data into unidimensional data and PUC > 0.90 means that ECV, ω, and ωH can be interpreted directly. The semPlot 32 and BifactorIndicesCalculator 33 in R and Jamovi software 34 were also used in these analyses.

Results
Adequate solutions were found for all models and these results are presented in Table 1.  (Table 4). For GSI, omega was 0.98 and omega hierarchical was 0.95, which suggests that around 97% of the reliable variance is due to the general factor, 3% is due to the specific factors, and 2% squarely to random error. 30 For the specific factors, omega values range from 0.83 to 0.93 and are higher than their omega hierarchical values. These results    suggest an essentially unidimensional structure as a result of a strong general factor that explains most of the reliable variance and is less affected by the multidimensionality induced by specific factors.
The ECV of the GSI explains 77% of the variance and, in conjunction with the PUC of 0.918, common variance might be interpreted as essentially unidimensional.

Nevertheless, the comparison of ECV_SG and ECV_GS
implies that most of the explained variance on the specific factors is due to the general factor and not to the item composition of the dimensions themselves.
Also, most of the BSI items showed high communality  Thus, evidence suggests that one general factor is sufficient to explain the score variability of the BSI.

Discussion
The present results provide replication of BSI internal structure models previously reported in different countries and samples. Five models were examined and the bifactor model was the best representation for Brazil. Our results strongly support the hypothesis of a unidimensional structure in the assessment of psychiatric symptoms using the BSI.
They are in line with results previously reported, 15,23 reinforcing the bifactor nature of the BSI regardless of cultural influences and mental health conditions. These results also support the hypothesis raised by Loutsiou-Ladd et al. 22 suggesting that the BSI is unidimensional, at least in non-clinical samples.
Our results suggest that the general symptom index presents the most robust psychometric properties, rather than the specific factors. The idea of a "p" factor is supported by previous psychometric studies, which argue that a bifactor structure of symptoms Therefore, future studies should assess whether this factor structure will remain relatively unchanged in a similar community sample in a post-pandemic scenario.
Investigation of evidence of the validity of BSI scores is also important to understand exactly what they represent and how they might be interpreted.