2.1. Sampling and procedure
The data for the current study were obtained from questionnaires distributed among residents of Beijing in 2017. The survey was conducted by the Questionnaire Star, a professional online survey company. The Principal Investigator signed a contract with the Questionnaire Star to specify the cost of the survey and how the responses would be kept confidential. The respondents were current residents of Beijing aged from 17 to 59. The quality control methods undertaken were as follows:
(1) the questionnaire was only distributed to users who meet selection requirements;
(2) screening items were set to verify that participants met the sample qualifications;
(3) online systems were used to monitor the process; these took account of the IP addresses, tracked which electronic devices were used, and administered trap items, time limits, and sampling procedures;
(4) after the completion of all questionnaires, a quality check was performed to assess the completeness, formatting, and effectiveness of each of the data records.
A total of 950 responses were investigated, and 12 were eliminated due to incompletion, giving a completion rate of 98.7%.
2.2. Ethical approval
The research protocol of the study was formally approved by the Institutional Review Board of the Ethics Committee at the University where the Principal Investigator is affiliated. All participants provided written informed consent prior to completing the study self-report instruments and did not provide identifying information on any of the questionnaires.
2.3. Measurements
2.3.1. Outcome (Depression)
The Center for Epidemiologic Studies Depression Scale (CES-D) was used to evaluate depressive symptoms (Radloff 1977). This self-report scale includes 20 items, each assessed on an 8-point Likert-type scale, with responses ranging from 0 to 7, representing how many days the participants have experienced depressive symptoms over the past week. Consistently with previous studies, this 8-point scale was re-coded into a 4-point scale, as follows: 0 = 0 days, 1 = 1–2 days, 2 = 3–4 days, and 3 = 5–7 days. The total points for all items (ranging from 0 to 60) was computed to indicate the level of depression. The four inverted items in the scale (items 4, 8, 15, and 20) were reversed before the tally. The good internal consistency of the scale for this sample was confirmed with Cronbach’s alpha (α = 0.9311).
2.3.2. Indicator (Relative Deprivation)
Relative Deprivation in Psychological Strain. A slightly modified version of the relative deprivation subscale of the PSS was used to evaluate the level of relative deprivation (Zhang, et al., 2014). For this, the respondents were asked to assess ten statements regarding situations they experienced in their lives, such as “Compared to other families in my community, my family is poor,” and “I believe I am good enough, but am dissatisfied with treatment from others.” Subjects rate their responses on the following five-point scale: 1 = never, 2 = rarely, 3 = maybe, 4 = often, and 5 = yes. The total score (ranging from 10 to 50) indicates the level of psychological strain due to relative deprivation, with higher scores associated with higher relative deprivation. The Cronbach’s alpha coefficient for the PSS subscale was α = 0.923.
Relative income deprivation. We designed a four-item instrument to assess respondents’ perceptions of relative lack of income compared to four groups as their references: friends, colleagues, occupation peers, and significant others. Each item was rated on a 5-point Likert-type scale: 1 = not at all satisfied, 2 = not very satisfied, 3 = generally satisfied, 4 = somewhat satisfied, and 5 = completely satisfied. The points were reverse coded, and the total score (ranging from 4 to 20) was used for analysis. Higher the scores indicated higher relative deprivation. The Cronbach’s alpha coefficient for this scale was α = 0.806.
2.3.3. Potential Moderator (Perceived Social Support)
The Multidimensional Scale of Perceived Social Support (MSPSS) was used to measure social support (Zimet, et al., 1988). The Chinese 12-item version of the scale has been tested and found to have sound validity and reliability among Chinese adolescents (Chou 2000). The answers ranged from 1 = strongly disagree to 7 = strongly agree. The total score (ranging from 12 to 84) was analyzed in the present study. High internal consistency was found (Cronbach’s α = 0.9425).
2.3.4. Control Variables
Sex was assessed as a binary variable, with 1 = male and 0 = female. Age was computed by subtracting the date of birth from July 2017. Marital status was coded as 0 = others (including cohabitation, separated but not divorced, divorced, and widowed), 1 = single, or 2 = married (including married and remarried). Education was coded as 0 = college degree and below (including no formal education, elementary school, middle school, vocational high school, high school, technical secondary school, technical school, and college) or 1 = bachelor’s degree and above (including university completion, graduate degree, and above). Occupation type was converted to match the International Socioeconomic Index (ISEI), a general measure that evaluates the conversion capacity of occupations in terms of the substitutability of human resources and the potential payoff, where higher ISEI scores indicate higher socioeconomic status (Ganzeboom, de Graaf, & Treiman, 1992). Monthly personal income was transformed to a logarithmic scale and controlled for in the model. Location of origin was a binary variable, with 1 indicating those from urban China and 0 indicating those from rural China.
2.4. Analysis
All data were analyzed with STATA 16.0. Major variables for study were compared across gender distributions. For all continuous variables, t-tests were performed, and chi-square tests were performed for non-continuous variables. Pearson’s correlation was computed to examine the relationship between variables.
Robust multiple linear regressions were performed using the iteratively reweighted least square (IRLS) method to examine the associations between the dependent variable and the explanatory variables. IRLS does not impose an assumption of a normal distribution in a sample but instead assigns an analysis weight that is yielded from an iterative algorithm for each observation to deal with heteroscedasticity in the sample, meaning that the estimation is more effective and robust than the ordinary least square (OLS) model.