A cross-sectional survey using convenience sampling was conducted. Before the beginning of the study, ethical approval was obtained from an institutional review board of National Cheng Kung University Hospital (IRB number: A-ER-107–025).
Two data collection strategies were used: a face-to-face survey and an online survey. The data collection period was from July 2018 to March 2019.The face-to-face survey was conducted in parks, community care centres, and a community activity centre of a special municipality in Southern Taiwan. These places were where older adults often engage in activities. Three trained research assistants with backgrounds in gerontology conducted the data collection. They had received training about the purposes, procedures, and contents of the measurements. An online survey link was posted on several social media sites. The purposes and contents of the measurements were also introduced clearly at the outset in the questionnaires. Therefore, the participants could complete the questionnaires by themselves.
The two collection strategies were used to increase sample size and diversity. For example, the face-to-face survey made it possible to recruit older adults and participants who did not know how to use the internet. The online survey made it possible to recruit participants from different areas, which increased the diversity of the neighbourhood environments. In the online survey, the aims and inclusion criteria were explained at the start of the questionnaire, and responses were excluded if the respondents did not fit the inclusion criteria, if there were inconsistencies in data, or if data were missing.
The potential participants were healthy adults aged 45 years old and above and able to complete the standardised questionnaires. Those who had cognitive problems (e.g., dementia), severe physical diseases (e.g., heart disease or lung diseases), or dependence on others (e. g, severe disability and cannot take care of themselves) were excluded.
The dependent variable was the amount of physical activity, and the predictors included demographic characteristics, self-rated health, self-efficacy of physical activity, and social support and neighbourhood environment for physical activity.
Demographic characteristics included sex, age, and educational level. Self-rated health status was also collected by asking ‘In general, would you say that your health is excellent = 4, good = 3, fair = 2, or poor = 1?’.
The Physical Activity Scale for the Elderly (PASE) was used to assess the level of the respondents’ physical activity in leisure time30. The scale includes six items relating to the preceding seven days: (1) sitting or sedentary behaviour, (2) walking outside the home, (3) light sport/recreational activities (e.g. fishing, billiards), (4) moderate sport/recreational activities (e.g. dancing, golf without a cart), (5) strenuous sport/recreational activities (e.g. aerobic dance, running), and (6) muscle strength/endurance exercises (e.g. hand weights, push-ups). The weights of the six activities were 0, 20, 21, 23, 23, and 30, respectively; and the weights were used to calculate total score of PASE24. The frequency (never = 0, seldom [1–2 days] = 1, sometimes [3–4 days] = 2, often [5–7 days] = 3) and hours per day of activity (less than 1 h = 1, 1–2 h = 2, 2–4 h = 3, and more than 4 h = 4) were rated. Total PASE scores were computed by multiplying the frequency of each activity (hours per day over a seven-day period) by the respective weights and summing all activities27. A higher score indicated a greater extent of physical activity in leisure time. The scale was translated into a Taiwanese format with good validity31.
Self-efficacy of physical activity was measured using a modified version of a previous scale32. It included seven items about the respondents’ confidence in their ability to engage in physical activity in specific situations, such as being in a bad physical condition, feeling depressed or in a negative mood, lacking time due to workload or household chores, lacking a place to exercise, lacking the skills needed, and bad weather. A 4-point Likert scale was used (4 = high confidence, 1 = no confidence). A higher score indicated higher self-efficacy in engaging in physical activities. The Cronbach’s alpha of this study was 0.85.
A 5-item scale was used to assess social support from family and friends for physical activity, such as engaging in physical activity with the participant or encouraging them to take physical activity20,33. A 4-point Likert scale was used (4 = strongly agree, 1 = strongly disagree). A higher score indicated greater social support for physical activity. The Cronbach’s alpha of this study was 0.74.
The physical activity neighbourhood environment scale (PANES) was modified34 including eight items: shops within easy walking distance of house, free or low-cost recreation facilities in the neighbourhood, sidewalks on most of the streets, visibility of people being physically active in the neighbourhood, facilities for cycling, transit stop within a 10–15 min walk from home, crime rate in the neighbourhood and safety of walking at night (reverse coding), and traffic on the streets making it difficult to walk (reverse coding). A 4-point Likert scale was used (4 = strongly agree, 1 = strongly disagree). A higher score indicated greater social support for physical activity. The Cronbach’s alpha of this study was 0.71.
With regard to sample size, there were 12 predictive variables, and at least 144 participants were needed for each group (sample size > 12 * 8 + 50 = 144)35.
Descriptive statistics were used to present the characteristics of the participants. Hierarchical regression was used. The items sex, age, educational level, self-rated health, and the method of data collection (online versus face-to-face survey) were entered in Model 1, self-efficacy, social support, and neighbourhood environments were entered in Model 2, and interactions between predictors were entered in Model 3. The reasons of the three models were that the variables in Model 1 were controlled as covariates, and Model 2 could specifically test whether the variables about different levels predict physical activity significantly, and Model 3 could test whether the interactions had higher predictive ability than the variables.
Interactions comprised the product of two variables, with mean centring of two variables conducted prior to computing the product interaction term. For example, where the predictors were self-efficacy, neighbourhood environment, and interaction, [self-efficacy – mean of self-efficacy] × [neighbourhood environment – mean of neighbourhood environment]. The strategy of mean centring was used to avoid multicollinearity, and the variance inflation factor was used to test for multicollinearity. The significant moderating effect that the interaction between two variables are also presented in figures.
In addition to analysing all the participants, we analysed middle-aged adults (aged from 45 to 64 years old) and older adults (aged over 65 years old) separately. SPSS 21.0 version was used for the statistical analysis and a p value of less than 0.05 was taken as the significance level. The Strengthening of Reporting of Observational Studies in Epidemiology (STROBE) checklist was used for reporting this study36.
Ethics approval and consent to participate
Ethical approval was obtained from the institutional review board of National Cheng Kung University Hospital (IRB number: A-ER-107-025). The participants who were recruited for the face-to-face survey signed informed consent forms. There were no illiterate adults involved in the study All procedures were performed in accordance with declaration of Helsinki’ statement.