Participants and Procedure
Six hundred and thirty Chinese older adults (≥60 years) were surveyed in Shanghai during January 2019. A dozen of investigators with unified training completed this survey using iPads or paper questionnaires. These targeting older Chinese were reached in several communities using a diverse range of recruitment strategies, which included family doctors’ advice to their older targets, community workers’ introduction in the neighbourhood centre, investigators’ initiative recruitment outdoor within neighbourhoods and encouraging referrals from participants themselves. For instance, we surveyed parts of the autonomous participants under the help of family doctors and community workers; another part of participants was from snowball sampling; and some other participants were completed through the measure of household survey with calls ahead. Participation was voluntary, and participants were informed that their responses were anonymous and confidential before starting the survey. It took participants about 30 minutes to complete the survey and older people with severe mental or cognitive disorder were excluded. Finally, there was no missing data under a strict quality control, and of these all 630 participants comprised the final sample for statistical analysis. The current group ranged in age from 60 to 94 years, with a mean age of 74.19 years (Standard Deviation (SD)=8.53). Table 1 shows descriptive statistics for sociodemographic variables, attitudes to ageing and frailty status.
Assessments and Measures
Experiences of ageism (EA) was measured with 11 questions including three aspects: 1) perceived ageism; 2) encountered AS; and 3) witnessed AS. Perceived ageism was measured by three questions: “How often, in the last year, has anyone shown prejudice against you or treated you unfairly because of your age?”; “How often, if at all, in the last year have you felt that someone showed you a lack of respect because of your age, for instance by ignoring or patronizing you?”; and “How often in the last year has someone treated you badly because of your age, for example by insulting you, abusing you or refusing you services?” All response scales ranged from 0 never to 4 very often . Encountered AS and witnessed AS were both consist of 4 similar questions, which were derived from the general perceptions (“An elder might be too old to be or to do something”)  and stereotypes (such as incompetence and memory loss) [27-30] based on the older age. For example, we ask the participants: “How often, in the past year, has anyone told you ‘as an older people, you should be…rather than…’?” and “How often, in the past year, have you witnessed someone told an old person ‘as an older people, you should be…rather than…’?”; “How often, in the past year, has anyone told you ‘you are too old for that, it’s for young’?” and “How often, in the past year, have you witnessed someone told an old person ‘you are too old for that, it’s for young’?”; “How often in the past year have you encountered someone doubted your competence because of your older age? such as don’t believe you understand well or you are more likely make mistakes if comparing to young” and “How often, in the past year, have you witnessed someone doubted an old person’s competence because of his/er older age?”; and “How often, in the past year, has anyone told you that’s so-called a ‘senior moment’ when you forgot something?” and “How often, in the past year, have you witnessed anyone blurt out ‘I’m old or muddled or useless’ when s/he cannot remember something?”. All response scales also ranged from 0 never to 4 very often. Given the extremely skewed distribution of the responses to these measures, we recoded each item as a dichotomy with consulting previous study [23, 24]. Older people who scored 1 or above on each item indicated a positive result, which was regarded as having experiences of ageism. For testing the psychometrics of this EA scale, we performed exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) to examine the structure validity. The initial eigenvalues (number>1) from the EFA showed a model with 3 components, which exactly fitted the proposed aspects and explaining 73.6% of the total variance. All of the three varimax-rotated components showed over 20% of the total variance (22.4%, 25.4% and 25.8%, respectively), and we also inspected for items that had very salient loadings (>.70) (see Appendix). The modified indices of the CFA were as follows: c2=106, df=38 (c2/df=2.802); and the requisite fitness parameters were within acceptable standards (Comparative Fit Index (CFI)=.983, Tucker-Lewis Index (TLI)=.975, Goodness of Fit Index (GFI)=.970 and Root Mean Square Error of Approximation (RMSEA)=.054) (also see Table 3). The Cronbach’s alpha for the three components of EA scale were .884, .858 and .857 respectively in the current study.
Age stereotypes (AS) is usually measured by rating “old people” on some personal characteristics or domains in their lives . In current study, AS was assessed with the similar set of statements, which were used for measurement of encountered or witnessed AS in the four aspects. Participants had to rate “dis/agree” instead of rating “frequency”. We asked participants: “What extent do you disagree or agree with the following statements: ‘as an older people, it should be…rather than…’; ‘older people are too old for something, it’s for young’; ‘comparing to young, older people are more likely to make mistakes’; ‘the older a person is, the more likely to be forgetful or muddled’”. These four response scales ranged from 1 strongly disagree to 5 strongly agree. The initial eigenvalues (number > 1) from the EFA suggested a model with 1 component, explaining 67.3% of the total variance, and the varimax-rotated component was also inspected for items that had excellent loadings (>.70) (also see Appendix). The modified indices of the CFA for the AS were also adequate (see Table 3). The Cronbach’s alpha for the AS scale was .832 in the current study.
Attitudes to ageing (AA) was measured with the attitudes to ageing questionnaire (AAQ), which was developed and validated by Laidlaw K, et al. in a worldwide cross-culture populations . This questionnaire has been validated in multiple culture, also including a Chinese version . The 24-item questionnaire was evenly divided into three domains including psychological growth (PG), physical change (PC) and psychosocial loss (PL) with acceptable Cronbach’s alpha of .592, .760 and .790, respectively. The questionnaire uses a Likert response format for each item from 1 strongly disagree to 5 strongly agree. PG focuses on the wisdom and growth, which reflects both positive gains in relation to self and to others about ageing; PC emphasizes the positive beliefs on maintaining physical health and the experience of ageing itself; PL presents negative experiences involving psychological and social loss in old age . Higher summated scores in each dimension for PC and PG indicate a more positive perception of ageing, while PL is reverse.
Frailty was assessed by the FRAIL scale, which included 5 items (Fatigue, Resistance, Ambulation, Illness, and Loss of weight). The FRAIL scale was constructed based on consensus of a European, Canadian and American Geriatric Advisory Panel . It was showed similar predictive accuracy to both the Fried’s Frailty Phenotype and Rockwood and Mitnitski’s Frailty Index [48, 49]. The FRAIL scale was being increasing used in Asia-pacific region , and showed a favourable validity in community-based older Chinese with below and above 75 years old . The criteria defined frail as the presence of 3 or more of these 5 symptoms, the presence of 1 or 2 defined prefrail, and 0 corresponded to robust.
Age, gender, education, marital status, economic condition and residence status were chosen as the potential confounding variables. Educational level was generally categorized into 5 levels (illiteracy, primary school, junior high school, high school or equivalent, and college or above). Marital status was divided into married and unmarried (never married, widowed and divorced). Economic condition was assessed by the question: “How do you think of your current income and daily expenses?” and the responses were “income lower than expenditure, income equal expenditure, and income higher than expenditure”. Residence status was measured by a multiple-choice question: “Who are you currently living with? (alone, spouse, parents, child/ren, grandchild/ren, others)”; and it was divided into live alone, live with spouse (only spouse), and live with others.
To examine the hypothesized model in Figure 1, we used AMOS 21.0 for windows to appraise the SEM with latent variables. In the first step, the Chi-square test was performed to screen out the potential confounding variables by SPSS 22.0 for windows. Based on the results in Table 2, we selected all sociodemographic variables except for gender and economic condition as the covariates in the SEM. For analyses, we transformed the categorical variables into binary variables on the basis of merging the categories with similar percentages of frailty. For example, education was divided into primary school or below (0) and junior high school or above (1); residence status was divided into live alone (1) and not live alone (0). In addition, frailty was addressed into an ordered variable (robust=1, prefrail=2, frail=3) according to the criteria. When a mixture of binary, ordered categorical, and continuous variables are included in SEM, analyses are usually based on polychoric/polyserial correlations .
In the second step, we assembled the modified measurement models and the structural equations simultaneously to establish the proposed SEM, and the Maximum Likelihood method was used to estimate parameters. To improve model fit, we freed covariances between error terms based on their modification indices (M.I.) during the estimation process. There has been no universal rule as to which model fit indices should be chosen, therefore, the most common indices and acceptable reference values included the magnitude of χ2 divided by its degrees of freedom (c2/df<3), CFI (<.90), TLI (<.90), GFI (<.90) and RMSEA (<.08) were reported in our study. There was no necessary to use any imputation method because of no missing cases in the sample. Lastly, indirect effect between variables were used Sobel test, p<.05 was considered statistically significant for all analyses.