Participants
Multistage cluster sampling was conducted to recruit 1390 first-year college students in mainland China. Five colleges and five universities were randomly sampled out of 1949 colleges and universities nationwide. The sampled schools included one out of 190 Project 211 universities (Project 211 is the Chinese government's new endeavor aimed at strengthening about 100 institutions of higher education and key disciplinary areas as a national priority for the 21st century), one out of 389 normal universities (those that train teachers, chiefly for the elementary grades), one out of 91 medical universities, one out of 160 business and finance colleges, one out of 130 colleges of science and technology, one out of 81 art colleges, one out of 299 universities administered by the Ministry of Education, one out of 118 provincial colleges, one out of 169 universities of agriculture and forestry, and one out of 403 other colleges. Then, a cluster sampling procedure was conducted to select classes in each of the 10 colleges and universities. Once selected, we sought to recruit all individuals within the cluster that had been chosen. Specifically, 24 classes were selected randomly from the potential pool of all classes. From the 24 classes, 1390 first-year students aged 16–24 years were identified as possible subjects.
Pooling all classes, the response rate was 84.7% (1177 out of 1390). Reasons for non-response included non-attendance of the survey class (207 students) and withdrawing before the questionnaire was completed (6 students). Cases marked by missing data on any of the variables under study were deleted from the analysis. Complete data on all variables were available from 1168 respondents (84.0%).
Procedure
Data were collected from September to December 2010. Approval was obtained from the Ethical Committee of Southern Medical University. Before the formal study, a pilot study was conducted in 400 first-year medical students to help formulate study hypotheses and modify questionnaires. We took the following steps to control the quality of the survey and to address the ethical issues in the formal study: 1) improved the survey instrument through pilot study (e.g., based on the pilot survey, we re-arranged sensitive items regarding suicidality in the front to middle part of the survey); 2) standardized instruction and trained ten surveyors, one in each college; 3) carefully selected appropriate survey time to avoid the influence of unexpected stressful life events such as examinations; 4) cooperated with the administrative team of the sampled colleges to ensure that the survey environment was quiet and the participants sat at one-seat intervals; 5) assured the survey was kept voluntary, anonymous, independent, and confidential; 6) asked students for verbal consent to participate and gave them the option to withdraw at any time; 7) clarified ambiguous items before distribution of the questionnaire (e.g., political beliefs); 8) asked participants to re-check for missing responses before collection of questionnaires; 9) encouraged participants to list at least one of three means of contact, including email, phone number and qq number (a popular instant messaging tool in China) so feedback on their mental health profile could be provided. Once serious suicide risk was detected (e.g., recent detailed suicidal plan) we would contact him or her to provide personalized treatment advice and help-seeking resources.
Measures
The survey instrument was a five -page questionnaire requiring about 30 minutes to complete. It included:
Socio-demographic characteristics of the participants included gender, age, city, ethnicity (Han nationality coded 1, other ethnic minority coded 0), household registration (rural area = 1, urban area = 0), perceived family finance (good = 1, mediocre = 2, bad = 3), father’s occupation, mother’s occupation, marital status, whether from single-parent family, whether an only child, and number of siblings if any.
Political belief was a dichotomous variable derived from two rounds of qualitative interviews and previous literature [22, 38]. First, 50 third-year college students were interviewed to obtain responses to the open-ended question, “What do you believe in politically?” Responses were coded by two independent research associates. Results showed that the responses varied, including socialism with Chinese characteristics (34%), Confucianism (26%), no belief (18%), individualism (10%), capitalism (4%), skepticism (4%), and others (4%). For example, one student responded: “I don’t believe in any specific political propaganda; however, I have my own personal core values, such as happiness.” That response was classified as individualism. Second, 40 college students were asked during regular class time to reflect on the question, “What are the most important elements that affect you politically and culturally?” The responses were again coded by two independent research associates based on the above classifications (i.e., socialism, communism, Confucianism, individualism, capitalism, and skepticism) and sorted by frequency. Results showed that the most frequent responses were socialism, communism and Confucianism.
On the basis of the two rounds of interviews and previous literature, the final survey item was determined as “Do you think you have an explicit political belief of socialism with Chinese characteristics?”, No = 0, Yes = 1. (Note: Socialism with Chinese characteristics is an integration of basic Marxism principles, Confucianism-based Chinese traditional culture, and contemporary Chinese conditions. It contains Deng Xiaoping Theorya, the Important Thought of Three Representsb and the Scientific Development Conceptc and other major strategic thoughts.) A total of 321 (27.5%) respondents replied "Yes" on this item.
Religious belief was a dichotomous variable with respondents reporting beliefs in any of the dominant religions (Christianity, Buddhism, Islamism, Taoism, or Catholicism) coded 1 and no religion coded 0. A total of 126 (10.8%) respondents were religious believers: Christianity, 20 (1.7%); Buddhism, 70 (6.0%); Islamism, 5 (0.4%); Taoism, 28 (2.4%); and Catholicism, 3 (0.3%).
Meaningfulness was assessed by the Purpose in Life scale (PIL), Chinese version [39]. The PIL consistsed of 20 statements rated on a seven-point scale with a high score (6–7) indicative of clear meaning and purpose, an intermediate score (3–5) representing indecision, and a low score (1–2) reflecting lack of clear meaning and purpose in life. A total score is calculated by summing the 20 ratings (total score ranges from 20 to 140). It has been validated in various samples [40]. Cronbach’s alpha was 0.89 for the current sample.
Psychopathology was assessed by the Chinese version of the Symptom Checklist-90-Revised (SCL-90-R). The SCL-90-R measures participants’ self-reported psychopathologic features on nine subscales including somatization, obsessive-compulsiveness, interpersonal sensitivity, depression, anxiety, hostility, phobia, paranoid ideation, and psychoticism [41]. Each question is rated on a five-point Likert scale (1 for no distress, 5 for extreme distress). This instrument has been used extensively in studies measuring a variety of mental disorders. The reliability and validity of the Chinese version of the SCL-90-R has been established in previous studies [42]. In the current study, Cronbach’s alpha was 0.96.
Suicidality was measured by the revised Suicidal Behaviors Questionnaire, Chinese version (SBQ-R) [43]. The SBQ-R has four items, each tapping a different dimension of suicidality: lifetime suicidal ideation (plan and attempt), prevalence of suicidal ideation over the past 12 months, threat of suicide attempt, and likelihood of suicide in the future. Specific items answered on a Likert-type scale are “Have you ever thought about or attempted to kill yourself?” (1–4), “How often have you thought about killing yourself in the past year?” (1–5), “Have you ever told someone that you were going to commit suicide, or that you might do it?” (1–3), and “How likely is it that you will attempt suicide someday?” (0–6). The total SBQ-R score ranges from 3 to 18, representing relative risks for suicidality on a continuum. Satisfactory psychometric properties have been reported in previous studies [44–46]. Cronbach’s alpha was 0.68 in the current sample.
Statistical analyses
Statistical analyses were performed using SPSS 15.0 for Windows and AMOS 7.0 (SPSS Inc., Chicago, IL). First, a t test or χ
2 test was used to identify statistically significant demographic differences associated with suicidality. Pearson product moment correlation was used to explore univariate associations between suicidality and independent variables including demographics, political belief, religious belief, meaningfulness, and psychopathology. Univariate analysis of variance was performed to test the interactive effect of political belief and religious belief on suicidality.
Multi-group analysis, a special case of SEM, was used to test the effects of gender differences on hypothesized models. Suicidality as a latent variable were assessed by lifetime suicidal ideation, plan and attempt, suicidal ideation over the past 12 months, suicide threat, and suicide possibility. Manifest variables included political belief, religious belief, political belief by religious belief and meaningfulness. Psychopathology was a latent variable with 10 indicators. Maximum likelihood estimation was employed as a global test of models. Model fit and comparison were ascertained using the following indices: root mean square error of approximation (RMSEA), comparative fit index (CFI), normed fit index (NFI), incremental fit index (IFI), Akaike information criterion (AIC), and expected cross-validation index (ECVI). Cutoff criteria for fit indices were determined following the recommendation of Hu and Bentler [47].
The tests of gender differences in the SEM framework start with estimating the hypothesized structure without constraining any parameter in both groups simultaneously (baseline model). An observed adequate fit of this model is required for further testing. All subsequent tests involve comparing a constrained model in which required parameters (i.e., factor loading, correlation paths, measurement residuals, and structural residuals) are equal for both groups with an unconstrained model that does not include the equality requirement. If the statistical fit of the constrained model reveals a significantly worse solution than the unconstrained one, this is interpreted as evidence for non-invariance. The fit of nested models can be compared by inspecting the significance of the change in χ
2 values. In other words, when we force certain parameters to be equal, the significantly decreased model fit indices suggests that at least one of the parameters is different across groups.