Participants and Procedure
Participants of this study were from a larger survey that investigated Chinese public’s intention/motivation of getting COVID-19 vaccine and its correlates. Student helpers (N=168) from various regions in China volunteered to spread an online survey link on their social networks via different social media platforms (e.g., WeChat). Within the survey period (from 23-July to 12-August, 2021), 3,273 participants provided consent and completed the survey. As a quality control, we excluded a number of participants because: (1) their age was out of our expected range (<12 or >90 years old; N=11); (2) they were living oversea (N=7); (3) they had duplicate IP address (N=195), or (4) they answered the survey too fast according to Leiner’s (2019) relative speed index (>2; N=75)[i](15). The remaining 2,985 participants constituted the final sample of the larger survey (Mage=22.07 years; 1,019 males).
Adolescents aged 12 to 24 from the said larger survey consisted of the sample of this study. The current sample size was 2,567 adolescents from 32 provinces/regions in China (850 males; Mage=19.87 years, SD=2.02). Among them, 157 and 28 participants reported that they had history of physical or psychiatric illness, respectively; and 44 participants reported that they were directly related to COVID-19 (e.g., confirmed/suspicious cases and/or relatives/friends of the confirmed cases, etc.).
This study was approved by the ethical committee of xx University. The survey was conducted online. Participants (or their guardians) provided consent by checking the box in the front page of the survey that fully explained the study. They reserved the right to withdraw from the survey at any point. Participation was voluntary and no incentive was provided. We did not collect identifiable personal particulars and confidentiality was stressed.
Positive Changes in Life Outcomes
We adapted the positive events subscale of the Epidemic-Pandemic Impacts Inventory to measure used in previous study to measure Chinese adolescents’ positive changes in life outcomes(11). The previous study used 21 items measuring whether participants had experienced positive changes during the initial lockdown across a number of life outcomes (e.g., relationships, physical activity, sleep, work) with binary response options (11). In this study, we modified this subscale in several aspects to suit the current participants better. First, we grouped some items that had similar meaning. For instance, in the original subscale, the items “more quality time with partner or spouse” and “more quality with children” were combined and modified as “more quality time with family members”. Second, social media information was filled with positive energy (e.g., sharing of gratitude, appreciation, social responsibility)(16) and patriotism (17) in China over the last two years when the government was combating COVID-19. In this regard, we added a few items that reflected this situation, such as “become more gratitude” and “become more patriotic”. Finally, we changed the binary response options to a five-point scale (from 1=strongly disagree to 5=strongly agree). A higher score indicates higher levels of positive changes in life outcomes. The final scale has 19 items. Participants were asked to rate how much they had experienced the listed positive changes in various life outcomes since the outbreak of COVID-19. The full set of items are illustrated in Table 2. The Cronbach’s a of this scale was 0.95 in this study.
We used the Brief Resilience Scale (13) to measure participants’ ability to bounce back. In this study, the Chinese version of BRS was used (12). This scale has 6 items rated on a five-point scale (from 1=strongly disagree to 5=strongly agree). A higher mean score indicates better ability to bounce back from stressors. Sample items are “It does not take me long to recover from a stressful event” and “I usually come through difficult time with little trouble”. The Cronbach’s a of this scale was 0.69 in this study.
We used GHQ-12 (18) to measure participants’ mental health. The Chinese version of this scale was used (19). This measure consists of 12 items which are either stated in negative or in positive wordings. Participants were asked to rate on these items according to their situation over the past month compared to their usual situation. All items are rated on a four-point scale (from 1=better than usual to 4=much worse than usual). A higher mean score indicates more mental health problems (i.e., less mental health). Sample items are “feel unhappy and depressed” and “lose confidence in self”. The Cronbach’s a of this scale was 0.90 in this study.
We measured several demographic variables as covariates, as they had been found to be associated with various life outcomes in Chinese public at the beginning period of COVID-19 (20). These covariates include biological sex (1=male, 2=female), age, their relationship with COVID-19 (1=directly related, such as confirmed/suspicious case and/or relatives/friends of the confirmed cases, etc., 2=not related), history of physical and psychiatric illness (1=yes, 2=no), and their current physical health condition (from 1=very poor to 5=very good).
We analyzed the data with SPSS 26.0 and Mplus 7.31 (21). The variables examined in this study had never been used in other studies. The research question, hypotheses, and the data analytic plan were pre-registered at aspredicted.org (protocol number: #73896). First, we conducted preliminary analyses, including means and standard deviation. Second, we carried out latent profile analysis (LPA) to examine the first research question. All the 19 items listed in Table 2 were used as the indicators of analysis. We first started with one-profile, and then increased the number of profiles systematically until we identified the best fitting model according to a number of indices, including Akaike Information Criteria (AIC)(22), Bayesian Information Criterion (BIC)(23), adjusted BIC (aBIC), Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (LMRT)(24), and Bootstrapped Likelihood-Ration Test (BLRT)(25). Smaller values of AIC, BIC, and aBIC indicated better model fit. The p-value values associated with LMRT and BLRT indicate whether the k-profile model (p <.05) or the k-1 profile model (p >.05) has a better fit. Besides, the value of entropy no less than 0.6 indicates good profile separation (26). In addition, we also considered theoretical meaningfulness of the profile (27) and the proportion of participants represented in the profiles (28). As a rule of thumb, no profile should have a group comprised of less than 5% of the participants (29). Multivariate analysis of variance (MANOVA) was conducted to examine whether the mean level of each indicator significantly differed across profiles. P-values and effect sizes (η2p) were used to judge the significance. Third, we performed a logistic regression model using a three-step procedure in Mplus (i.e., R3STEP auxiliary command)(26) to examine the second research question after identifying the best-fitting model. Logits from the model output were transformed into odd ratios for explanation purposes. Demographic variables measured above were included in the model as covariates. Fourth, analysis of variance (ANOVA) was performed to examine the third research question, with the average score of GHQ as the dependent variable and the profiles as the independent variables. Both p-values and effect size (η2p) were used to determine the significance. Based on ANOVA, we further conducted ANCOVA controlling for covariates. Finally, as a robust check, we replicated the analyses for research questions 2 and 3 with winsorized scores of resilience and GHQ[ii], respectively.
[i] Leiner’s (2019) relative speed index was calculated in the following steps. First, although the questionnaires administered to participants who had and who had not got vaccinated were largely the same, several questionnaires were different, thus resulting in different completion time between the two groups of participants. Hence, we split the two groups and gauged the relative speed index of each participant for each group. Second, we calculated the median completion time for each group. Third, we used each participant’s completion time to divide the median completion time of the group he/she was in, and a relative speed index could be obtained. For instance, if the median completion time of the non-vaccinated group was 600 seconds, a participant in this particular group had a completion time of 200 seconds. Then, this participant’s relative speed index was 600/200 = 3 and he/she should be excluded because his/her index was larger than the cut-off point (i.e., 2).
[ii] Winsoring approach (Tukey, 1962) was used to handle the outliers of the resilience and GHQ by replacing the outliers with the nearest number within the -3 to +3 SD range.