The factor structure of the 12-item general health questionnaire (GHQ-12) in young Chinese civil servants

General Health Questionnaire (GHQ), developed by British scholar Goldberg in 1972, is one of the most popular and widely used screening instruments for recognition and measurement of mental health [2]. The original GHQ consists of 60 items, and now there are multiple versions, including GHQ-1, GHQ-12, GHQ-20, GHQ-28 and GHQ-30. Among these, the GHQ-12 is the most popular due to its simplicity [2, 3]. The GHQ-12 consists of 12 items, each of which is evaluated by four indexes. The two most commonly used scoring types are the bi-modal (0-0-1-1) and Likert scoring methods (0-1-2-3) [4].

The GHQ-12 has satisfactory reliability [5‐7] and good sensitivity and specificity [8, 9]. It has been applied to different populations in different countries to study its reliability and validity, and to explore the mental disorders characteristics of the populations in previous studies. These respondents have included both adolescents [4, 7, 10] and adult community residents [1, 11]. Numerous studies have focused on occupational groups, including nurses [12], academic staff [13], cohort [11] and civil servants [14] etc. Simultaneously, different language versions of the GHQ-12 have proven acceptable, (e.g., Arabic [9], Australian English [13], German [2], Spanish [10] and Swedish [15]). Although few studies have focused on the applicability of the Chinese version of the GHQ-12, it has been proven that the Chinese version is reliable and valid [16, 17]. Scholars have applied the Chinese version of the GHQ-12 to university students [16], and the results have shown that the instrument is acceptable. It has also been used to evaluate the mental disorders of secondary school students, reaching the same conclusion [17]. The studies mentioned above were focused on students, but there is scarce literature on the applicability of the GHQ-12 to specific Chinese occupational groups. Although previous studies have used the GHQ-12 to measure the mental health of civil servants and nurses, no research has been done on young civil servants in grassroots government agencies, a group that has specific occupational and physical characteristics.

Various populations in China suffered from mental disorders [18‐21]. Among these, civil servants are professionals with a high risk of mental disorders [22]. Civil servants are responsible for managing the state, and they link average citizens and the government [23]. The nature of their work and the organisational environment determines the specificity of their work, and their occupational health has specific characteristics. In fact, previous studies have shown that the health condition of this occupational group is not optimistic, and even can be worrisome [24].

Civil servants are under high work-related pressure, resulting in poor mental and physical health and lack of support [25, 26]. In addition, their high-intensity work makes them more vulnerable to disease. Most young civil servants who have recently entered the executive authorities are in lower-level positions. Marmot [27] conducted a study finding that mortality rate due to coronary heart disease of male civil servants in low-level positions was three to six times higher than that of male civil servants in higher positions in London. Civil servants in low-level positions have been employed for a short time [28], have significant employment pressure [29], tend not to play sports and are more susceptible to high blood pressure and blood sugar [30]. In addition, negative health behaviours and habits among civil servants are quite common, including addicted to tobacco and alcohol, lack of exercise and a tendency toward chronic diseases like insomnia [31‐34]. Thus, concerns about the health, especially the mental disorders, of special occupational groups are of great importance and urgency.

At present, the existing researches on the factor structure of the GHQ-12 have been full of controversy, as the results have been inconclusive. The majority of previous studies extracted one to three factors from 12 items. Initially, the GHQ-12 was designed as a unidimensional scale that all projects included in a main factor, but only a few scholars supported this original one-factor model in subsequent empirical studies. Banks [5] and his colleagues conducted three separate investigations, to test the validity for three types of samples: employees of an engineering firm, recent leavers and unemployed men. The results showed that the scaling properties of the GHQ-12 are sufficiently good to justify the use of a single-scale score. However,the original unidimensional model may be unable to provide researchers sufficient information because of its simplicity [5]. Since then,a number of studies have explored its possible factor structures and revealed different dimensions in the structure of the GHQ-12, mainly with two or three factors. A survey of teachers in Western Australia found that the GHQ-12 contains two dimensions, and they were described separately in a statement by positive and negative words [35]. This result also repeatedly showed the adequate fit to the data [15, 36, 37]. In another study, using a large sample of young Australians (n = 8998), Graetz [10] extracted three factors from the GHQ-12: Social Dysfunction (including items 1, 3, 4, 7, 8 and 12), Anxiety and Depression (including items 2, 5, 6 and 9), Loss of Confidence (including items 10 and 11). With the application of the confirmatory factor analysis, many studies have suggested that this three-factor model, compared to other models, has a better goodness of fit [7, 38, 39]. Simultaneously, other scholars have also found that the GHQ-12 has three dimensions [9, 40],but named the factors differently from Graetz [10]. However, the validity and usefulness of these multi-dimensional solutions have frequently been questioned because of the high degree of correlation between the factors [7, 41]. Other studies have focused on the connections between wording effect and negative vocabulary. Hankins [12] found through statistics from the Health Survey for England that the unidimensional model, with correlated errors on the negatively worded items, was superior to the two-factor (positively and negatively worded items) or three-factor models proposed [42]. The study results of Li [17] and Aguado are consistent with this. Both contend that as long as the wording effect is controlled, the GHQ-12 is a unidimensional model [17, 43]. In general, although there have been many studies on the factor structure of the GHQ-12, they have not reached a unanimous conclusion. Thus exploration and verification of the factor structure of the GHQ-12 are still necessary.

As it is a constantly improved and widely used measurement tool for mental health, the GHQ-12 has been applied to different populations in many countries. The research subjects have mostly been members of the general population and certain occupational groups. No paper that has made its sample young civil servants working at grassroots government agencies. What is more, the Chinese version of the GHQ-12 has not been researched extensively; few studies have examined the factor structure and psychological characteristics of this scale in Chinese occupational groups. In this paper, we use the GHQ-12 to measure the mental health of young Chinese civil servants and to analyse the GHQ-12 factor structure in this occupational population. This study has the following three aims: (1) Creatively applying the GHQ-12 to young civil servants working in grassroots government agencies to broaden the scope of use and acceptance of the GHQ-12; (2) Researching the GHQ-12 factor structure of occupational groups in China to provide more extensive clinical and empirical data for the improvement and development of the GHQ-12; (3) Through studying the reliability and validity of the GHQ-12 to promote its use in Chinese clinical measurements and individual self-assessment.

The subjects of this study are young civil servants working at 24 grassroots government agencies. We designed and conducted a cross-sectional study and selected six cities with comparatively developed economy in eastern China (Nanjing, Shanghai, Suzhou, Hangzhou, Yangzhou, Wenzhou). These cities are under better urban construction, and they have more complete and diverse government agencies, so they make representative sample regions. We randomly selected civil servants under 45 years old working in administrative departments below the county level. In the process of distributing the questionnaire, we asked whether respondents were willing to accept the investigation and state principles, such as the confidentiality and authenticity, in advance. Respondents were asked to complete the questionnaire independently. If they did not understand any item, they could consult the investigator on the spot. Finally, 1200 questionnaires were distributed, (i.e., 200 for each city). A total of 1051 questionnaires were returned, making the response rate 87.58 %.

This study used the GHQ-12 proposed by Goldberg [44] to measure the mental health of young Chinese civil servants. This questionnaire includes 12 items (six positively worded items (e.g., Have you been able to concentrate on whatever you are doing?) and negatively worded items (e.g., Have you lost much sleep over worry?)). We adopted the four-point Likert scale, with each item ranging from 0 to 3. For negatively worded items, ‘0’ indicated Not at all, ‘1’ indicated Seldom, ‘2’ indicated Usual and ‘3’ indicated More than usual, while positively worded items were reversely scored. All items were added to obtain the total score, making the score range 0-36 (with a higher score indicating worse mental health). Scores over the cut-off point of 12 could be classified as cases [8].

First, the score distributions for each item and the whole scale were calculated to understand the mental health of young Chinese civil servants. Then, we conducted a reliability analysis of the GHQ-12 scale. Finally, we used factor analysis to explore and validate the GHQ-12 factor structure of the respondents. In the factor analysis process, the total sample (n = 1051) was randomly split into an Exploratory Factor Analysis (EFA) sample (n = 525) and a Confirmatory Factor Analysis (CFA) sample (n = 526) by SPSS 19.0. Then, EFA was conducted to explore the underlying factor structure of the GHQ-12 in the EFA sample using the principal axis extraction method with varimax rotations. Parallel analysis was the method used to decide the number of factors extracted from EFA [45]. CFA with maximum likelihood estimation (MLR) was referred to for model fit.

The selection of the models for CFA was based on a literature review. We referred to a large number of previous studies and selected the model with the best fit from each validation study. As mentioned in the introduction, most of the studies extracted one to three factors from 12 items. The models most strongly supported the unidimensional model with wording effect by Hankins [41], Andrich and Van Schaubroecks’ two-dimensional model [35] and the three-dimensional model by Graetz [10], which have been supported and validated by many other scholars. Furthermore, an additional five two-dimensional and three-dimensional models [6, 9, 40, 46, 47] were also included, which have been proven to fit well by previous studies. Finally, coupling the unidimensional model from the original GHQ-12 designed by Goldberg and our model results of EFA, CFA was then performed by testing ten competing factor models with one-, two- and three-dimensional solutions of the CFA sample. The ten types of measurement models were as follows:(1) unidimensional model (original), (2) unidimensional model with correlated errors on the negatively worded items by Hankins [41], (3) two-dimensional model proposed by Andrich and Van Schaubroeck [35] (positively worded items (including items 1, 3, 4, 7, 8 and 12) and negatively worded items (including items 2, 5, 6, 9, 10 and 11)), (4) two-dimensional model by Schmitz et al. [46] (Anxiety/Depression (including items 1, 2, 6, 7, 10 and 11), Social Performance (including items 4, 5, 8, 9 and 12)), (5) two-dimensional model by Politi et al. [6] (Dysphoria (including items 2, 5, 6, 9, 10 and 11) and Social Dysfunction (including items 1, 3, 4, 7 and 8), (6) three-dimensional model by Graetz [10] (Social Dysfunction (including items 1, 3, 4, 7, 8 and 12), Anxiety and Depression (including items 2, 5, 6 and 9), Loss of Confidence (including items 10 and 11), (7) three-dimensional model by Farrell [47] (Anxiety (including items 2, 5, 10, 11 and 12), Depression (including items 1, 6, 7, 8 and 9) and Social Dysfunction (including items 3 and 4), (8) three-dimensional model by Daradkeh et al. [9] (General Dysphoria (including items 5, 6, 9, 10 and 11), Lack of Enjoyment (including items 1, 2, 7,8 and 12) and Social Dysfunction (including items 3 and 4), (9) three-dimensional model by Martin [40] (Cope (including items 1, 3, 4, 8 and 11), Stress (including item 2, 5 and 7) and Depression (item 6, 9 and 12), (10) unidimensional model obtained from EFA.

The reasons we conducted EFA before CFA are as follows: First, previous studies on the factor structure of the GHQ-12 have been full of controversy, and they have been inconclusive. The names of each factor in the same dimensions model also vary in the existing factor dimensional model. Second, so far, no studies have conducted EFA on Chinese occupational groups, although Ye [16] and Li [17] have applied EFA to Chinese students. Last, to the best of our knowledge, this is the first study of the GHQ-12 factor structure on young Chinese civil servants. Young civil servants are a specific occupational group at a high risk of mental disorders. They bare a greater work pressure and higher propensity for mental disorders [25, 26]. The results of the general population or other occupational groups may not be applicable to young civil servants.

We used EPIDATA3.1 double entry to test the sample quality. The significance level for all tests was 0.05. We performed the CFA model with AMOS 21.0 and EFA models and other statistical tests with SPSS 19.0. Thus, we adopted this short analysis to explore the reliability and validity of the GHQ-12 on young Chinese civil servants.

The demographic and socioeconomic characteristics of respondents mainly included gender, age, education, employment level and marital status. Of the 1051 respondents, 61.5 % were male, and 38.5 % were female. In terms of age, 41.5 % of respondents were aged 20–29 years, 48.4 % of respondents were aged 30–39 and 10.1 % were aged 40–45. Education levels of these young civil servants were divided into four categories: Junior college and bellow, Undergraduate, Graduate and Doctorate. Most respondents fell into the Junior college and below (32.7 %) and Undergraduate (42.6 %) categories, and most respondents were staff and below (41.4 %), and not married (46.8 %). In addition, the EFA sample and the CFA sample had similar distributions of age, education and employment grade. The CFA sample had a high ratio of respondents who were male or married. In general, the demographics of the respondents are representative of the young Chinese civil servants. Table 1 provides descriptive statistics of the sociodemographic characteristics of the respondents in this study.

Table 2 shows the overall and individual item scores of the GHQ-12. The GHQ-12 items were to the left on both sides of the score distribution except item 10, and the kurtosis coefficients were less than 0. The average score of the GHQ-12 was 23.62 (SD = 7.92), which was far higher than the cut-off point of 12 [8], and of all the respondents, 86.49 % scored greater than or equal to 12 and could be used as cases. This indicated that the health, especially the mental health, of the respondents was in very poor condition. In particular, the highest average scores were for items 1, 9 and 11, which were more than 2.10. Of these items, the average score of item 9 was 2.14 (SD = 1.08), as the highest, indicating that the majority of respondents felt unhappy and depressed. While the average score of item 12 was 2.06 (SD = 1.06), the majority of respondents (74.3 %) scored 2 or 3 points, and only 13.8 % of respondents scored 0, showing that in general the respondents did not feel happy.

Cronbach’s alpha reliability was obtained using the Spearman-Brown formula. Table 3 shows the reliability and correlations between items and for the overall GHQ-12 scale. Cronbach’s alpha of the GHQ-12 was 0.844 (>0.8), indicating that the scale has a good reliability and correlations. Item 11 had the highest correlation coefficient, which was 0.609. Correlation coefficients for the rest of the items and the total score ranged from 0.324 to 0.606. The Cronbach’s alpha coefficients when an item was deleted were over 0.8, but below the total coefficients, they range from 0.824 to 0.837 when an item was deleted, indicating that each item was necessary and of equal importance.

Previous studies on the factor structure of the GHQ-12 have been full of controversy, and they have been inconclusive. In the factor analysis process, the total sample (n = 1051) was randomly split into the EFA sample (n = 525) and the CFA sample (n = 526) by SPSS 19.0. Using the data from the EFA sample, EFA was firstly used to identify the possible latent variables of the GHQ-12. Table 4 shows the results of the EFA of the scale using the principal axis extraction method with varimax rotation. Kaiser-Meyer-Olkin (KMO) measured the sampling adequacy of the GHQ-12 as 0.908 (>0.9), and the Bartlett test results passed the significance test (approximated chi-square = 1249.186, df = 66, P < 0.001), indicating the adequacy of the sample. We extracted three factors: Factor 1, Factor 2 and Factor3. The eigenvalue of each factor was greater than 1, and the load factor was greater than 0.4 for each item. Factor 1 included six items, items 4, 6, 9, 10, 11 and 12, and explained 21.206 % of the variance. Factor 2 included four items, item 3, 5, 7 and 8, and explained 17.339 % of the variance. Factor 3 included two items, items 1 and 2, and explained 11.675 % of the variance. The three factors together explained 50.220 % of the total variance.

Table 5 shows the goodness of fit indexes for the CFA model. The statistics results indicate that all the models were well fitted with the actual data. In CFA, the fit indices of all models were relatively similar, and RMSEA was less than 0.05 (approximately 0.03). CFI and TLI were also greater than 0.9, and the chi-square/degrees of freedom ratio was less than 2, ranging from 1.465 to 1.561. This indicated that all of the ten confirmatory models for the GHQ-12 did equally well and had satisfactory levels of fit.