Bidirectional relationships between physical exercise and depressive symptoms among Chinese older adults evidence from the China family panel studies

Data and sample

The data used in this study come from the China Family Panel Studies (CFPS) published by the China Social Science Research Centre of Peking University. This dataset aims to capture dynamic changes across individual, family, and community levels, reflecting trends in education, health, society, economy, and population in China. The CFPS sample spans 25 provinces, municipalities, and autonomous regions, with a target sample size of 16,000 households, encompassing all household members within the selected households. In terms of survey content, the CFPS questionnaire covers a lot of individual information about the elderly, including individual physical and mental health status represented by depressive symptoms, physical exercise, and self-assessed health, as well as demographic variables such as gender, age, marriage status, and education level. These contents provide data support for this study to investigate the causal relationship between physical exercise and depressive symptoms in the elderly. A total of six tracking surveys have been published to date, for 2010, 2012, 2014, 2016, 2018 and 2020. CFPS was approved by the Biomedical Ethics Committee, Peking University (IRB00001052-14010).

There are two reasons why the data from the three China Family Panel Studies from 2016 to 2020 were selected for this paper. Firstly, because of the inconsistency in the measurement of depressive symptoms, it is not appropriate to combine the data from 2016, 2018 and 2020 with the data from the previous surveys. Second, the cross-lagged model used in this paper requires only three periods of tracking survey data, and the data from the 2016–2020 period are the most current. Therefore, based on the advantages of CFPS data as well as the research needs, the three survey years of 2016, 2018 and 2020 data were selected for this study and processed as panel data to overcome the endogeneity problem caused by omitted variable bias and two-way causality problem. With reference to the criteria for defining the elderly in China, the elderly subjects aged 60 and above in the sample were retained²⁵. During the data cleaning process, participants with missing values on any of the questionnaire items were excluded from the analysis. This resulted in an initial valid sample of 7,782 participants in 2016. Of these, 5,757 participants completed the first two surveys, while 2,502 participants participated in all three surveys. A detailed flow diagram of the data cleaning process can be found in the supplementary file (Figure S1.pdf).

Given the importance of sample size in longitudinal tracking studies, the attrition rate was also analyzed in the present study. Specifically, the attrition rate for the 2018 data was 26.02%, while for the 2020 data, it was 56.54%. To assess potential bias, we compared demographic variables, physical activity, and depressive tendency scores between tracked and untracked participants using 2016 data. The normality test revealed that both groups exhibited some degree of skewness in several demographic variables, as well as in physical activity and depressive tendencies. Subsequent chi-square and Mann-Whitney U tests showed significant differences between untracked participants and those who completed all three survey waves. Detailed results can be found in the supplementary file (File S1.pdf). Specifically, differences were found in gender (χ²(1) = 8.76, p < 0.01), age (Z = 19.32, p < 0.001), hukou (Z = 193.02, p < 0.001), educational level (Z=−3.86, p < 0.001), marriage status (χ²(1) = 83.31, p < 0.001), self-assessed health (Z=−7.70, p < 0.001), and depressive symptoms (Z=−2.63, p < 0.01). These findings suggest potential structural attrition in the sample, which needs to be controlled for in subsequent analyses to mitigate any biases in the results.

Tools and instruments

Depressive symptoms

Referring to existing studies²⁶, this study measured depressive symptoms in older adults using the CES-D8 scale, the most widely used instrument for measuring depression levels. Specific questions for CES-D8 include ‘I feel depressed,’ ‘I felt that everything I did was an effort,’ ‘My sleep was restless,’ ‘I was happy,’ ‘I felt lonely,’ ‘I enjoyed life,’ ‘I felt sad,’ and ‘I felt that life was not worth living’. In processing the data, we reassigned the answer choices to 0, 1, 2 and 3, and reversed the scoring of the 2 reverse-question question scores. The final depression score is the sum of the respondents’ answers and ranges from 0 to 24, where higher scores represent more severe depressive symptoms. The Cronbach coefficients for this scale were 0.802, 0.789, and 0.747 at the three time points.

Physical exercise

This study chooses to measure it in two operationalized ways, continuous variable and categorical variable. Based on the reference of existing studies²⁷, we measured physical exercise of the elderly in two ways: firstly, as a continuous variable, the CFPS questionnaire interviewed the total number of hours of physical exercise per week of the respondents. In this study, we firstly excluded the samples who answered ‘don’t know’, and then calculated the average daily physical exercise time according to the formula (average daily physical exercise time = total weekly physical exercise time/7) and uniformly replaced the samples whose average daily physical exercise time was ≥ 360 min with 360 minutes²⁷. Since the overall data were positively skewed, this study performed a square root transformation to make them closer to a normal distribution. Secondly, as a categorical variable. Referring to the categorization of previous nationally representative studies that have been processed in the literatur^28,29, the respondents’ physical exercise levels were categorized into two groups based on frequency and duration: the samples with weekly physical exercise > 3 and exercise time > 30 were labelled as 1, indicating that they participated in physical exercise; the rest were labelled as 0, indicating that they did not participate in physical exercise.

Control variables

The present study referred to previous influences on depression and controlled for common variables that may confound the relationship between physical exercise and depression in the elderly^6,7, including gender (male = 1, otherwise 0), age, hukou (agricultural hukou = 1, otherwise 0), educational level (on a scale of 0–19 from ‘not in school’ to ‘postgraduate and above’), marriage status (in marriage = 1, otherwise 0), and self-rated health status(1–5 points, representing “unhealthy” to ‘very healthy”).

Statistical analysis

Using STATA 17.0 and Mplus 8.0 for data analysis, the analytical argument of this study is divided into three main steps, including cross-sectional data analysis, fixed effects analysis, and cross-lagged panel model analysis.

Step 1: The correlations in the results of the existing studies were tested by analyzing the cross-sectional data. This step is analyzed using OLS estimation method, the specific model is:

Step 2: Individual-level fixed effects analysis. This step effectively controls for the influence of individual invariant characteristics (e.g., lifestyle) on depressive symptoms. The effect of causality between physical exercise and depression levels can be assessed more accurately. The fixed effects model is:

Step 3: For possible bidirectional causal relationship, Cross-Lagged Panel Models (CLPM) were used for analysis. CLPM is a structural equation model based on longitudinal data that tests for bidirectional causality among variables by introducing time-lagged effects. Specifically, the model assumes that the value of variable X (e.g., physical exercise time) at time point t may affect the value of variable Y (e.g., depressive symptoms) at time point t + 1, and vice versa. By simultaneously estimating the following two paths (1) X(t) → Y(t + 1) and (2) Y(t) → X(t + 1) and controlling for the simultaneous correlation (i.e., covariance) between X and Y at the same time point in the model, the CLPM is able to differentiate between the direction and strength of the bi-directional causal relationship. In existing studies related to the health field, there has been a lot of literature on the use of cross-lagged panel models to test bidirectional causality. However, few scholars have analyzed the causal relationship between physical exercise and depressive symptoms in older adults using this method. In this study, we constructed a CLPM based on three waves of follow-up data: by controlling for covariates such as age, gender, and baseline health status, we reduced the interference of potential confounders on the results; finally, we used the Full Information Maximum Likelihood (FIML) estimation method to compute the relationship between the mean daily physical exercise time and depressive symptoms. The statistical significance level is p < 0.05.