Productive activities
Drawing on previous studies [29, 30] and the new connotations of productive activities that have emerged in recent years based on Asian cultural contexts, this study includes paid work, formal volunteering, providing informal help, caregiving, social activities and leisure activities as the six indicators of productive activity.
Specifically, paid work includes (1) whether you do farm work for your household or other employers for pay and (2) whether you have a job other than farming. Participants who answered “yes” to either of these two questions were marked as having paid work. Formal volunteering is measured based on whether the participants have done voluntary or charity work (1 = yes, 0 = no). Providing informal help is measured by asking participants whether they provided help to family, friends, or neighbors who do not live with them (1 = yes, 0 = no). Caregiving is measured by asking participants (1) whether they care for grandchildren, (2) whether they help their parents with household work, cooking, doing laundry, shopping or financial management, and (3) whether they care for a sick or disabled adult who does not live with you. Caring for family members and other patients is qualitatively different, but the response distribution for caring for patients was uneven (35 individuals, 2.8%). When indicators have low variability (less than 5% of responses in one category), this may interfere with accurately distinguishing between latent classes [31]. Therefore, we combine the three activities of caring for family members and other patients to form a single indicator of caring status. Social activities were measured by asking older adults whether they (1) interacted with friends, (2) took part in a community-related organization, and (3) attended an educational or training course and other social activities. Leisure activities were measured by asking whether participants (1) had participated in playing Mahjong, played chess, played cards, or went to a community club, (2) went to a sport, social, or another kind of club and (3) used the Internet. The three dimensions of caregiving, social activities and leisure activities were marked as 1 if any question was answered yes and 0 otherwise.
Health
This study will evaluate the health of older adults through the three dimensions of self-rated health, functional health, and depression. Self-rated health was assessed with reference to previous research by asking participants how they think their health is (very good = 1, good = 2, fair = 3, poor = 4, very poor = 5) [32]. The higher the score, the worse the self-rated health.
Functional health is evaluated by the degree of difficulty in five activities of daily living(five ADL). These include dressing (includes taking clothes out from a closet, putting them on, buttoning up, and fastening a belt), bathing, eating (such as cutting up your food), getting into or out of bed, and using the toilet (including getting up and down). The participants were asked whether they had difficulty with the above five aspects due to health and memory. No difficulty = 1, I have difficult but can still do it = 2, I have difficulty and need help = 3, I can not do it = 4. The score ranges from 5 to 20, and the higher the score, the poorer the functional health of the older adult. The ADL assessment tool has been shown to have good reliability and validity in Chinese populations [33, 34]. The Cronbach’s alpha coefficient for this scale was 0.856.
The Center for Epidemiological Studies Depression Scale (CES-D 10) was used to evaluate the participants’ depression. The CES-D 10 consists of 10 items that ask participants about their feelings and behaviors in the past week. Each item is scored from 1 to 4 according to the degree of depression reflected by it, with a higher score indicating a more serious depression. The reliability and validity of the CES-D 10 scale has been extensively reviewed and is considered appropriate for use with the Chinese older population [35]. The Cronbach’s alpha coefficient of this scale is 0.816.
Control variables
This study controls for a set of variables that may be related to the health of older adults: age, gender, income, marital status (married, non-married), and education level (1 = No formal schooling, 2 = Elementary school, 3 = Middle school, 4 = High school or above). In order to avoid possible multicollinearity among multiple explanatory variables in the cross-sectional data from affecting the regression results, the tolerance (Tol) and variance inflation factor (VIF) were tested separately. In this study, Tol values were much greater than 0.1, and VIF values were less than 5, indicating no multicollinearity problem among the selected variables.
Statistical analysis
This study used SPSS 27.0 and Mplus 8.3 to analyze the data. Latent class analysis (LCA) was used to analyze the situation of older urban adults in terms of their productive activities (paid work, volunteering, informal help, caregiving, social activities and leisure activities). LCA classifies groups similarly to conventional cluster analysis. This employs an objective approach to determine the optimal number of groups using a model-based stochastic analysis method [36]. LCA evaluates a variety of model-fit statistics by incrementally adding categories to determine the best-fitting model. These statistics include the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted BIC (a-BIC), entropy values, and the Lo-Mendell-Rubin Likelihood Ratio test (LMR-LRT) and bootstrapped likelihood ratio test (BLRT) values. Lower AIC, BIC, and a-BIC values indicate a good model fit; entropy values close to 1 indicate more accurate classification. Significant p-values for BLRT and LMR indicate the K classes model is better than the K-1 classes [37].
Association between different latent classes and self-rated health, functional health, and depression were analyzed using linear regression. Two models with different predictor variables were tested: one that included only latent classes (unadjusted model) and one that included both latent classes and control variables (adjusted model). The association between different productive activity patterns and health was tested by estimating standardized regression coefficients(β).
As part of the sensitivity analyses, we employ an automated BCH method to estimate pairwise comparisons of health across different activity patterns(BCH χ2). This method is currently recommended as the best method for predicting continuous outcomes from latent class membership [38]. The BCH approach accounts for possible class classification error by using a weighted multiple group analysis to evaluate means across classes for a continuous auxiliary variable [39]. This method is more accurate than the traditional linear regression method applied in this paper, but problems such as negative variance due to small sample sizes occur [40]. Therefore, we will validate further the main report analyses using the results of the BCH.
link