Literature search
Studies were primarily collected through searching three of the most important Chinese academic literature databases, Wanfang (the Chinese Wanfang data), CNKI (China National Knowledge Infrastructure) and Chongqing VIP (Chinese Journal of Science and Technology of VIP). These three databases cover nearly all of the Chinese journals of psychology (including all related disciplines) after 1985 and most of the doctoral dissertations and master’s theses since 1995. We searched for articles using the key words military, army, troops, soldiers, warriors, military officer, recruits, military students or armed police, intersected with anxiety. All relevant studies up to January 2012 were included in present analysis.
Inclusion criteria and moderators’ coding
The studies included in present analysis had to meet the following criteria: (a) participants were Chinese serviceman on active duty including recruits, military students or armed police; (b) the study included at least 10 male or 10 female participants; (c) participants were not clients at a counseling center or patients in hospital; (d) means were reported for unselected groups, not groups that were extremely high or low on anxious measure; (e) because the Chinese version of Spielberger’s State-Trait Anxiety Inventory (STAI) developed by Spielberger [21] (Chinese version is cited in Wang et al. [22]) is one of the most frequently used instruments assessing anxiety in Chinese military personnels, so we included original studies reporting anxious scores on this scale for our meta-analysis; (f) the study must report sufficient data including means, standard deviations and sample sizes. Some reports meeting these inclusion rules were excluded for the following reasons: If more than one paper using essentially the same or overlapping data were published, the report with the earliest, most complete, comprehensive results were included in the present meta-analysis. A flow diagram of the study inclusion and exclusion is shown in Figure 1. The final sample consisted of 45 separate studies with a total of 18,106 participants for state anxiety and 21,047 participants for trait anxiety. A reference list describing the 45 studies included in the meta-analysis was provided in supporting information (Additional file 1). We performed this cross-temporal meta-analysis mainly in accordance with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Additional file 2) [23].
According to previous studies [4, 16, 24], we used the following procedure to estimate the year of data collection: (a) if the year of data collection was mentioned in the article or by the author, we included it; (b) if the article reported the original date that the article was received, we included this year as the estimated year of data collection; (c) if the article reported only the final date that the article was accepted, we subtracted this year by 1, for publication time; (d) if no other additional information reported explicitly, the year of data collection was assumed as 2 years preceding publication, as in previous studies [16, 24]. For possible moderators, only publication type of studies and military categories of participants were coded. There were two types of publication studies in our meta-analysis, journal papers and doctoral dissertations or master’s theses. We code doctoral dissertations and master’s theses as one type of publication study due to a lower number of them. According to China’s military service system, military categories of participants include army, navy, air forces, second artillery force and military students. We included studies regarding military students for the following reasons. First, military students are members of Chinese military personnels in the terms of China’s military service system. Second, in China, most of military students are selected from combat troops to military college for continuing education. Other information was insufficiently reported in original studies due to military secret.
Data analytic strategy
Mean, standard deviation and sample size of each study were extracted for meta-analysis. Where a paper reported anxious scores on the normal group before and after treatment (e.g., psychological education), only the baseline (untreated if possible) data were used.
Changes of anxious scores in Chinese military personnels over time were examined by correlating mean scores with the year of data collection. According to previous cross-temporal meta-analyses, means were weighted by the sample size of each study. We performed our analyses using Stata software, version 12.1 (Stata Corp., College Station, Texas), and the βs reported were standardized to allow for easier interpretation. To calculate the magnitude of changes in anxious scores over time, we used the regression equations and the averaged standard deviation of the individual samples when they were available. To compute the mean scores for a specific year (e.g., 2004), we used the regression equation from the statistical output, which followed the algebraic formula y = Bx + C, where B = the unstandardized regression coefficient, x = the year of data collection, C = the constant or intercept, and y = the predicted mean anxious score. We computed the average standard deviation by averaging the within-sample standard deviations reported in the data sources. According to previous studies [25], this method avoids the ecological fallacy, also known as alerting correlations. To make the results to be easier understood, we converted magnitude of change into percentile scores and we used standard deviation to measure the magnitude of change of anxiety scores. We assume that anxiety scores of Chinese military personnels is a normal distribution. For example, we would refer the anxiety score in 1991 was the baseline norm (in the 50th percentile). If the anxiety score was increase one standard deviation from 1991 to 2011. According to the standard normal curve, the percentile score of anxiety score in 2011 was 84.13% (by checking the standard normal distribution table). Xin et al. [10] have utilized this method to make the magnitude of change to be informative. Finally, we converted the effect size into year-of-publication effect based on the formula r2 = d2 ÷ (d2 + 4) [26], which denoted the variance explained by year. We can examine the effect of sociocultural environment factors over time beyond genetics and their family environment through the variance explained by year [10, 26].
Sources for social indicators
Considering the social indicators used in Twenge’s study [4], the social indicators chosen were divorce rate, crime rate, percapita health expenditure, unemployment rate, military expenditure, promotion rate of senior school graduates. Divorce rate and crime rate come from the China Statistical Year book[27]; per capita health expenditure were obtained from the China Health Statistical Year book[28]; unemployment rate were obtained from the China Statistical Abstract[29]; military expenditure were obtained from the National Defense White Paper of China[30]; promotion rate of senior school graduates were obtained from the website of the Ministry of Education [31].