Research design, target population, and sampling
This study used a cross-sectional quantitative survey design, with data collected through telephone surveys from June to August 2019. The target population consisted of older people aged 55 and older who were also legal citizens residing in Hong Kong at the time of the survey and able to communicate in Cantonese. To minimize sampling error, telephone numbers were randomly generated using known prefixes assigned to telecommunication services providers under the Numbering Plan of the Office of the Communications Authority (OFCA) in Hong Kong. A number of measures were taken to select the required amount of Random Digit Dialling (RDD) numbers and to optimize the operational efficiency. Firstly, all telephone numbers obtained from the Numbering Plan were pre-screened by a computer system to detect the tritone signal for invalid numbers. Secondly, computer screening results were combined with a previously obtained database of actual manual dialling records to identify and eliminate invalid numbers. Known business lines gathered from previous dialling records or publicly available sources were also excluded from the sampling frame. Finally, the estimate required amount of telephone numbers were randomly drawn from the database to produce the final sampling frame for the survey.
For second-level sampling, when telephone contact was successfully established with a target household, one qualified respondent was selected using the ‘next birthday rule’ to further eliminate potential bias. To ensure the representativeness of the final sample, the raw data collected were rim-weighted according to the latest gender-age distribution of the Hong Kong population aged 55 or older and educational attainment distributions [34]. The sample size calculation was based on the number of Hong Kong Chinese population older than 55, which was 246, 0300 by the end of 2018 [35]. A total of 1067 participants was needed to ensure the sampling error of the percentages obtained would be within no more than +/-3% based on the full sample at a 95% confidence level.
Data collection
Data collection was completed by a professional team of survey interviewers from the Hong Kong Public Opinion Program, using a structured questionnaire. The questionnaire was first developed by the research teams and finalized after consulting several older adults who meet our recruitment criteria. The questionnaire included about 50 questions, and each interview took about 15 minutes to complete. Each sampled telephone number was called up to five times at different times and on different days before being dropped as ‘non-contact’. All interviews were conducted anonymously. On-site supervision, voice recording, screen capturing, and real-time camera recording were used to monitor interviewers’ performance throughout the data collection period. Ethical approval was obtained from the Human Subjects Ethics Subcommittee of the authors’ home university.
Measures
Grandparent status was determined by asking the participants to self-identify as a current grandparent (having at least one grandchild), future grandparent-to-be (answering “yes” to the question “do you expect to be a grandparent in the future?”), or not expected to be a grandparent. Health and wellbeing variables included self-rated physical health, self-rated mental health, happiness, and resilience. Self-rated physical health was measured by a single item asking participants to rate their physical health along a 5-point scale (1=very poor, 5=very good), with a higher score indicating better physical health. Self-rated mental health was measured by a single item asking participants to rate their mental health along a 5-point scale (1=very poor, 5=very good), with a higher score indicating better mental health. Single-item self-rated health status measures have been found highly correlated with objective measures and are reliable instruments [36, 37]. Happiness was measured using the Chinese version of the Subjective Happiness Scale (SHS) [38], consisting of four items, each rated along a 7-point scale, with a higher score indicating a higher level of happiness. The scores for all items were summed to form an overall score ranging from 4 to 28. The Cronbach’s Alpha of the SHS in this study was 0.798, indicating a good internal consistency. Resilience was measured by the Chinese version of the Connor-Davidson Resilience Scale (CD-RISC) [39], a two-item scale with a score range of 0 to 4 for each item. Total scores range from 0 to 8 with a higher score indicating a higher level of resilience. The Cronbach’s Alpha of the CD-RISC2 in this study was 0.673, indicating an acceptable internal consistency.
Based upon previous research results concerning predictors of health and wellbeing among older people, the effects a few confounding variables were also tested. These included gender, age, marital status, education level, self-rated financial status, and level of physical activity. Gender was grouped as either male or female. Age referred to participants’ chronological age, while marital status was grouped as either married or non-married (single/widowed/divorced). Education level was grouped into eight levels ranging from no formal education (1) to post-graduate level or above (8). Educational status was further re-categorized into three groups: lower than middle school, middle school, and higher than middle school. Self-rated financial status was measured by a single item asking participants to rate their level of financial adequacy along a 5-point scale (1= very inadequate, 5=very adequate). Level of physical activity was measured by asking participant to rate their general level of physical activity, including going out walking, house chores, and exercises along a 5-point scale ranging from very low (1) to very high (5).
Data analysis
Single imputation method was adopted to fill in the missing value as the missing rate was minimal. Multiple imputation with five imputed datasets generated by Markov chain Monte Carlo (MCMC) method was also performed for sensitivity analysis. The results of the two methods were similar. Descriptive analysis was conducted to examine the demographic background of participants. ANOVA tests were performed to compare differences between the three groups of participants. If the ANOVA tests were significant, Tukey’s HSD tests were further run to test the differences between each two groups. Hierarchal regression analysis was conducted for each of the well-being variables as the dependent variable. Independent variables include gender, age, education, financial status, marriage, physical activity, and grandparent status. The controlling variables were first entered into the model, followed by the grandparent status. The increases in R2 of the regression models were examined. Interaction terms between grandparent status and five demographic characteristics, including age, gender, marital status, education, and finance were examined in the regression models, respectively, as well.