The Physical Activity Transition among Adults in China: 1991–2011

Abstract

Previous studies have linked work, home production, travel activities, and inactivity with weight and health outcomes. However, these focused on average physical activity over time rather than changes in physical activity and associated socio-demographic and economic factors and urbanicity. Using the 1991–2011 China Health and Nutrition Survey data, we estimated the metabolic equivalent of task hours per week for individuals in occupational, domestic, travel, and active leisure domains and sedentary hours per week. We present the distributions among adult men and women (aged 18–60) and use quantile regression models to explore factors associated with these trends. Trend analyses on the distribution of physical activity show declines along the whole distribution of occupational physical activity for men and women and domestic physical activity for women in China. These patterns remain consistent after adjusting for individual- and household-level factors. Controlling for urbanicity mitigated the decrease in occupational physical activity, particularly for men, but not the decrease in domestic physical activity. Given China's rapid urbanization and its association with occupational physical activity declines and the strong time trend in domestic physical activity, there is a need to invest in interventions and policies that promote physical activity during leisure and travel times.

I. INTRODUCTION

International surveillance data and a number of studies have shown that physical activity (PA) levels appear to be declining globally,^1-3 and physical inactivity was the fourth highest risk factor for death in the world in 2004.^{4, 5} Indeed not only does PA bring about clear health and functional benefits^{6, 7} that extend to all segments of the population,⁶ but being inactive or sedentary has been shown to be a distinct risk factor for numerous noncommunicable diseases (NCDs) independent of PA.^{8, 9}

While it is clear that there are significant health consequences associated with PA and inactivity, measuring and monitoring the levels of activity at the population level across the broad spectrum of daily living domains have been limited. Monitoring and recommendations have primarily focused on leisure time activities, including walking, biking, jogging, and sports;^10-12 sedentariness, particularly television viewing and related behaviors (e.g., snacking while watching television);^{13, 14} or total PA levels. Consequently the key domains of occupational and domestic work have largely been ignored with few exceptions.^15-17

Among studies that have looked at changes in domain-specific activities, the focus has been only on changes in these PA domains at the average or mean along with factors that are associated with those changes at these average PA levels.^{16, 18-21} What has been needed is a study of the distribution of these domain-specific activities over time and estimations of what factors (individual, household, and environmental) might be associated with the distributional trends. This investigation can allow us to determine if factors associated with changes in PA are primarily occurring among those who are already relatively inactive or among those who are relatively active.

China is the world's most populous nation and second largest economy, and the health of its population can have significant social and economic implications. In fact a recent study found that physical inactivity contributes to 12–19% of the risks associated with the five major NCDs in China, namely, coronary heart disease, stroke, hypertension, cancer, and type 2 diabetes, and is responsible for at least 15% of the medical and nonmedical yearly costs of these NCDs in the country.⁷ China has been experiencing significant social and economic changes since the late 1980s. We investigated whether and how those might have affected PA distributions across various domains of daily living for adult men and women over a 20-year period. The goal is to begin understanding the key economic, environmental, and sociodemographic factors associated with these noted changes in PA distribution. In particular we are interested in understanding if the noted changes are purely due to secular trends (implying societal attitudes) rather than responses to these factors.

II. DATA AND METHODS

Study Population

We used data from the China Health and Nutrition Survey (CHNS), a prospective household-based study that includes multiple ages and cohorts across nine rounds of surveys between 1989 and 2011 in nine diverse provinces and three megacities (Beijing, Shanghai, and Chongqing were added in 2011).²² A multistage, stratified sampling design was used to ensure that the CHNS provided representation of rural, urban, and suburban areas varying substantially in geography, economic development, public resources, and health indicators.⁶ It is the only large-scale, longitudinal study of its kind in China. Our study was approved by the institutional review committees of the University of North Carolina at Chapel Hill and the National Institute of Nutrition and Food Safety, China Center for Disease Control and Prevention. Participants provided their written, informed consent. Further information on survey procedures and the rationale of the CHNS is in the cohort profile.⁶

Our sample included the eight rounds of survey data collected in 1991, 1993, 1997, 2000, 2004, 2006, 2009, and 2011. All 18- to 60-year-old adults were considered eligible subjects (N = 22,716). We did not use data from 1989, because the sample at that wave only included adults 20 to 45 years old. We excluded participants who had fewer than two waves of data collection (N = 9,646), as we would be unable to assess temporal trends for these individuals. This excluded all observations from the three megacities, as data were collected in those areas in 2011 only. Additionally observations were excluded from the analysis if a participant was currently pregnant, disabled, or retired at the time of data collection. This resulted in the exclusion of an additional 3 subjects who were pregnant at all measurements, 6 subjects who were disabled throughout the entire course of their participation in the study, and 281 subjects who were retired. Our final analysis sample included 12,777 participants with an average of approximately four observations collected on each subject (53,135 observations).

The data are an unbalanced panel, with some individuals being absent in certain waves and returning in future waves as the sample was replenished over time. However, relative to previous rounds of data collection, retention was between 80% and 88% across all surveys after 1991.⁶

As the trend analysis did not have any subjects from the megacities, we looked at all individuals in the megacities in 2011 for comparison purposes. After exclusion of disabled, currently pregnant, and retired individuals, the 2011 megacity population included 2,274 participants.

Outcomes

The outcomes of interest are the metabolic equivalent of task (MET) hours per week in each of four domains of PA, occupational, domestic, active leisure, and travel, and hours per week in sedentary leisure time. MET is defined as the ratio of a person's working metabolic rate relative to his or her resting (basal) metabolic rate.²³ Therefore the MET hours per week measurement accounts for both the average intensity of each activity (or subactivity) and the time spent in each activity. In the CHNS questions about occupational and domestic activities started in 1991, but questions about active leisure and travel were only included starting in 1997. Sedentary leisure time was calculated as hours per week due to the low MET values associated with sedentary activity. Measurements of sedentary leisure time among adults were only available starting in 2004. Details on how these values were calculated are described elsewhere.^{1, 18}

Covariates

Income measures were based on reported gross annual per capita household income, which was inflated to 2011 values and categorized into wave-specific tertiles. An individual’s education was based on the highest education level reported throughout the course of inclusion in the CHNS. Urbanicity was calculated at the community level using a multicomponent continuous scale. Communities could receive a maximum of 10 points for each of 12 components, including population density, economic activity, traditional markets, modern markets, transportation infrastructure, sanitation, communications, housing, education, diversity, health infrastructure, and social services.²⁴ We also included time as a continuous measure, defined as years since 1991.

Statistical Methods

We first present descriptive statistics on the outcomes and covariates of interest stratified by adult men and women due to differences in gender roles. We present the means, standard deviations, and percentages (when appropriate) for all outcomes and covariates of interest for select years and note when there were statistically significant differences in means (or percentages) between 1991 and 2011 when p < 0.05. For PA domains that were not collected in all years, these tests were conducted comparing the first year they were measured with 2011. We used chi-square tests for categorical variables and t-tests for continuous variables.

We also present the distribution of PA for men and women to describe the trends for each of the PA domains over time. We used nonparametric methods to calculate confidence intervals for quantiles.²⁵ We were interested in testing whether the noted changes in PA distribution over time were due to secular trends or other individual-level factors, such as age, education, income, or the degree of urbanicity of the community. As a way to handle the skewed distribution of the outcome variables and to describe the full distribution, we used quantile regression techniques that estimate quantiles conditional on the covariates. Using this method rather than traditional linear regression based on the mean is particularly helpful where the effect of covariates differs at different levels of the response variable. Traditional linear regression can also be very sensitive to outliers, and this method by comparison is much more robust.^26-29 We used longitudinal quantile regression techniques that allowed us to account for repeated measurements collected on the same individual over time by the inclusion of random effects.²⁸ For comparison, we also ran traditional linear mixed models on these outcome variables. In these models age was defined as age in 2011 when included as a covariate, since age changes linearly with time.

Based on our descriptive data and past studies,¹⁸ the vast majority of PA in China among adults is comprised of occupational and domestic PA, and the amount of active leisure and travel PA is exceedingly small. Moreover occupational and domestic components were the only ones for which we had complete data across all eight waves. Therefore we only ran quantile regressions for occupational and domestic PA as the outcomes. For these outcomes we considered three model specifications. Model 1 controlled for time only, so the coefficients from these models would tell us what change in MET hours per week of PA was associated with each additional year for someone who was at a particular percentile of the PA distribution. Model 2 controlled for time and individual and household characteristics (age, education, and income), and model 3 controlled for time, individual and household characteristics, and the community urbanization index. We used R version 2.15.1 to conduct quantile regressions and mean regressions, and for all other descriptive analysis we used SAS 9.3.

We were unable to include the three megacities in the trend analyses using longitudinal quantile regressions since there was only one wave of data for these localities. However, we were curious about whether there were differences in what was observed in the three megacities in 2011 compared to the other locations in the same year. It is also possible that the PA levels found in these megacities may forewarn of trends we might see in the other locations. Therefore we present the mean and distributional statistics for each of the PA domains by these locations in 2011.

III. RESULTS

Trends in Average PA among Chinese Adults

Figures 1a and 1b illustrate the change in average PA by the occupational, domestic, active leisure, and travel domains among adult men and women in China from 1991 to 2011. For both genders, declines in overall PA were largely driven by occupational PA reductions. These fell from 382 MET hours per week in 1991 to 264 MET hours per week in 2011 among adult men and fell from 420 MET hours per week in 1991 to 243 MET hours per week in 2011 among adult women. Interestingly men engaged in fewer MET hours per week of occupational PA on average than women and also reported much less domestic PA than women on average. Nonetheless there was also a decline in domestic PA from 66 MET hours per week to 48 MET hours per week among women. For the most part active leisure and travel PA contributed very little to the overall PA of Chinese adults regardless of gender. Active leisure remained below seven MET hours per week on average among men and below three MET hours per week on average among women over this period, and travel PA stayed consistently low, between one to two MET hours per week on average for both genders.

Figure 1.

Average trends in occupational, domestic, travel, and active leisure PA (in MET hours per week) and sedentary leisure time among adults in China, 1991–2011

a. Males

b. Females

During this time of PA decline, other aspects of this population were changing as well, as seen in table 1. Over the course of the 20 years of this analysis, the average age of the population increased only a little over 10 years due to our restriction of the analytic sample. Additionally education levels were changing. Among subjects included in the sample in 1991, 62% of males and 42% of females had completed a secondary education or higher as compared to 77% of males and 58% of females in 2011. The study areas also became more urbanized, as shown by an increase in the urbanization index over time.

Table 1.

Descriptive data on analytic sample by covariates for adult men and women in years 1991, 1997, 2004, 2011

Variable		1991		1997		2004		2011
		Male	Female	Male	Female	Male	Female	Male	Female
N		3,560	3,288	3,826	3,480	3,263	3,294	2,489	2,518
Age		35.21 (10.85)	36.4 (10.26)	37.09 (11.08)	38.36 (10.49)	41.65 (10.57)	42.14 (10.08)	45.82¥ (9.26)	45.69¥ (8.98)
Income tertile (wave-specific)€	Low income	33.03% (0.79%)	32.85% (0.82%)	33.12% (0.76%)	32.84% (0.80%)	32.64% (0.82%)	33.33% (0.82%)	30.70% (0.92%)	35.27% (0.95%)
	Middle income	32.92% (0.79%)	33.15% (0.82%)	32.98% (0.76%)	32.99% (0.80%)	32.58% (0.82%)	33.39% (0.82%)	33.07% (0.94%)	32.88% (0.94%)
	High income	34.04% (0.79%)	34.00% (0.83%)	33.90% (0.77%)	34.17% (0.80%)	34.78% (0.83%)	33.27% (0.82%)	36.24% (0.96%)	31.85% (0.93%)
Highest education achieved	Less than primary	14.27% (0.59%)	36.71% (0.84%)	10.90% (0.50%)	30.92% (0.78%)	7.72% (0.47%)	24.68% (0.75%)	5.83%¥ (0.47%)	19.98%¥ (0.80%)
	Primary completed	23.88% (0.71%)	21.05% (0.71%)	22.92% (0.68%)	22.21% (0.70%)	21.33% (0.72%)	23.01% (0.73%)	16.99% (0.75%)	22.32% (0.83%)
	Secondary or greater completed	61.85% (0.81%)	42.24% (0.86%)	66.18% (0.76%)	46.87% (0.85%)	70.95% (0.79%)	52.31% (0.87%)	77.18% (0.84%)	57.70% (0.98%)
Urbanization index		44.12 (15.98)	44.85 (15.83)	50.35 (17.62)	51.04 (17.68)	60.28 (19.73)	59.76 (19.57)	68.43¥ (18.82)	66.84¥ (18.56)

^€Due to multiple individuals in each household, the wave-specific tertiles of household income are not exactly evenly distributed.

^¥Significantly different from 1991 values based on chi-square tests for categorical variables and paired t-tests for continuous variables.

Shifts in the Distribution of PA among Chinese Adults

Appendix table 1 presents the distribution (at the 10th, 25th, 50th, 75th, and 90th percentiles) and mean of each domain of PA and sedentary leisure time along with confidence intervals for the available waves of data for men and women. We also show graphically in figure 2 the change in the distribution of occupational PA over time among adult men and women for select years. We see that for both genders the distribution shifted to the left, with the distributions becoming narrower and a smaller proportion of the samples having higher occupational PA over time. Similarly figure 3 shows the distribution of domestic PA among adult men and women for select years. Like occupational PA, the distribution for domestic PA is also quite skewed, with only a small portion of the population participating in larger amounts of PA. Domestic PA for males appears consistently low over time, but domestic PA for females shows a shift to the left and a narrowing of the distribution over time.

Figure 2.

Distribution of occupational PA among adults in China for waves 1991, 1997, 2004, and 2011

a. Male Occupational Physical Activity

b. Female Occupational Physical Activity

Figure 3.

Distribution of domestic PA among adults in China for waves 1991, 1997, 2004, and 2011

a. Male Domestic Physical Activity

b. Female Domestic Physical Activity

Additionally the results in appendix table 1 show that the mean and distributions for travel PA have not changed much over time. On average there were some slight increases in active leisure PA, but the distribution shows that these only happened for those on the higher end of the distribution (90th percentile) while remaining nonexistent for the less active. Interestingly the mean and distribution of sedentary leisure time did not appear to have changed much from 2004 to 2011 either with the exception of women on the lower end of the distribution (10th and 25th percentiles), for whom sedentary leisure time rose by two to three hours a week.

Table 2 shows that quantile regressions can provide more details about how individual, household, or environmental factors might be associated with various levels of occupational PA (figure 4) and domestic PA (figure 5). In table 2 we also present the unadjusted MET hours per week from 1991 to show the PA levels at the 10th, 25th, 50th, 75th, and 90th percentiles and provide a reference for the estimated annual change in MET hours per week at these percentiles. The results from model 1 of the quantile regressions appear to suggest the same decline. Occupational PA for females in model 1 shows a yearly decline most noticeably at the higher percentiles, although for females the decline in occupational PA is seen even in the 25th percentile. In the lower percentiles the yearly change in occupational PA is not as pronounced. These patterns remain consistent even after adjustment for individual- and household-level factors, as shown in the results for model 2. However, in terms of occupational PA, once adjustment is made for urbanicity (model 3), the decline is noticeably attenuated if not eliminated for males. The mean regression results tend to be fairly similar to the results in the higher percentiles. Both suggest declines in PA, as seen in unadjusted models as well as in models adjusting for individual- and household-level factors, and mitigated declines after adjustment for urbanization.

Table 2.

Quantile regression results for 10th, 25th, 50th, 75th, and 90th percentiles vs. mean regression results for year coefficients

Year effects on	Quantile regression, coefficient (bootstrapped standard error)					Mean regression, coefficient (standard error)
	10th	25th	50th	75th	90th	Mean
A. Adult men, occupational
Unadjusted MET hrs/wk in 1991	100.00	200.00	368.00	552.00	678.00	382.05
Model 1	0.00 (0.00)	0.00 (1.05)	−5.33 (0.46)	−6.67 (0.20)	−7.14 (0.29)	−6.32 (0.16)
Model 2	0.00 (0.00)	−5.53 (2.15)	−5.07 (0.26)	−5.02 (0.30)	−6.02 (0.23)	−5.95 (0.16)
Model 3	1.15 (0.95)	0.35 (0.31)	0.57 (0.23)	0.56 (0.28)	0.80 (0.45)	−0.05 (0.18)
B. Adult women, occupational
Unadjusted MET hrs/wk in 1991	100.00	200.00	426.00	624.00	762.00	419.96
Model 1	−0.20 (0.51)	−7.69 (0.36)	−9.09 (0.21)	−10.26 (0.41)	−11.90 (0.74)	−10.69 (0.19)
Model 2	0.00 (2.52)	−7.63 (0.54)	−9.61 (0.30)	−7.50 (0.42)	−11.03 (0.91)	−10.47 (0.19)
Model 3	−4.17 (0.33)	−3.27 (0.38)	−2.45 (0.38)	−2.38 (0.35)	−2.06 (0.41)	−3.86 (0.21)
C. Adult men, domestic
Unadjusted MET hrs/wk in 1991	0.00	0.00	0.00	20.77	54.60	17.50
Model 1	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	−0.35 (0.12)	−0.16 (0.03)
Model 2	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	−0.25 (0.08)	−0.15 (0.03)
Model 3	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	−0.23 (0.06)	−0.26 (0.03)
D. Adult women, domestic
Unadjusted MET hrs/wk in 1991	7.53	27.50	54.60	88.36	138.49	66.34
Model 1	0.28 (0.04)	−0.03 (0.10)	−0.63 (0.04)	−1.22 (0.07)	−1.53 (0.10)	−0.91 (0.05)
Model 2	0.00 (0.09)	0.10 (0.04)	−0.50 (0.05)	−1.35 (0.06)	−1.66 (0.10)	−1.05 (0.05)
Model 3	0.00 (0.05)	0.16 (0.04)	−0.39 (0.05)	−1.20 (0.07)	−1.51 (0.13)	−0.91 (0.06)

Note: Model 1 includes year only in the model; model 2 includes all components of model 1 as well as age, education, and income; model 3 includes all components of model 2 as well as the urbanization index as a continuous variable.

Figure 4.

Estimated annual MET hours per week change in occupational PA among adults in China from quantile regression models 1–3

a. Male Occupational Physical Activity

b. Female Occupational Physical Activity

Figure 5.

Estimated annual MET hours per week change in domestic PA among adults in China from quantile regression models 1–3

a. Male Domestic Physical Activity

b. Female Domestic Physical Activity

Most males had very low levels of domestic PA, so it was not possible to see any associations (figure 5). Females in model 1 show a yearly decline most noticeably at the higher percentiles. These patterns remain consistent even after adjustment for individual- and household-level factors, as shown in the results for model 2, and even after also controlling for urbanicity (model 3).

Appendix tables 2 and 3 present the estimated coefficients and standard errors for the quantile and mean models for male and female occupational PA and domestic PA from model 3. Higher urbanization indexes are associated with lower occupational PA for both males and females (appendix table 2). Additionally for both genders, accounting for urbanicity, those with higher occupational PA (75th and 90th percentiles) had a greater absolute reduction of PA compared to those with lower occupational PA (10th and 25th percentiles). Part of this is due to “bottoming out” effects. Looking at the other covariates, we also found that in general lower education levels and lower income levels are associated with higher occupational PA.

Appendix table 3 shows that a higher urbanization index, higher education, and higher income are associated with lower domestic PA for females. For adult males, who do not engage in much domestic PA on average, there were only significant findings for those on the 75th and 90th percentiles, and urbanicity is actually positively related to domestic PA on the right side of the distribution.

We present the descriptive statistics for 2011 for each of the PA domains by locality in figure 6. We found that the average occupational PA appeared to be lower and sedentary leisure time appeared higher in the three megacities compared to the nine provinces for both men and women. However, the opposite was the case for the average domestic PA and active leisure PA. The average travel PA seemed similar regardless of locality. Appendix table 1 also provides a comparison of the distributions for each of these domains of activity. We note that the distribution of occupational PA for both genders moved to the right but with the 10th percentile engaging in no occupational PA and the 50th percentile expending only 100 MET hours per week or less at work. Sedentary leisure time was higher across the whole distribution in the three megacities compared to adults residing in the nine provinces. As for the distribution of domestic and active leisure PA, it appears that adults residing in the three megacities were more active on average, and this is driven by those on the higher end of the distribution.

Figure 6.

Average trends in occupational, domestic, travel, and active leisure PA (in MET hours per week) and sedentary leisure time (hours per week) among adults by locality in 2011

IV. CONCLUSION

We found that throughout this 20-year period, for both adult men and adult women in China, occupational and domestic PA were by far the largest contributors to PA, and active leisure and travel PA were very low. PA levels in both the occupational and the domestic domains decreased significantly from 1991 to 2011. More importantly the change in the distribution of PA shows that declines are occurring along the whole distribution of occupational PA for both adult men and adult women in China and in domestic PA for women. These trends remain consistent even after adjusting for individual- and household-level factors. However, controlling for urbanicity (assuming urbanicity is constant over time) mitigates the decrease in occupational PA, particularly for men. Controlling for urbanicity did not seem to affect the decrease in domestic PA among women.

Implications

China continues to urbanize rapidly in both urban and rural settings. The country's aging population means that the lack of PA we are finding among adults may translate to a very high prevalence of chronic NCDs, which will be a large burden on future generations and will have serious implications for the Chinese and world economies. Already physical inactivity is estimated to be responsible for at least 15% of the medical and nonmedical yearly costs of the five most prevalent NCDs in the country.⁷ This number will only rise as the population ages if something significant is not done soon to arrest or reverse this trend in PA.

Previous studies have found that, in the case of China, urbanization and urbanicity have resulted in net negative health outcomes.³⁰ However, urbanicity does not necessarily need to be deleterious, and many argue that the benefits of urbanicity can be maximized so long as it occurs with public health considerations in mind.³¹ Our comparisons of the findings from the nine provinces with the three megacities in 2011 indicate that the decline in occupational PA and the increase in sedentary leisure time are unlikely to reverse. Nonetheless it is possible that domestic PA and active leisure PA could increase, although the experiences of the three megacities indicate that those increases have been minimal and limited to the higher end of the distribution. Therefore there is a need to invest in interventions and policies that promote active leisure and travel PA among all adults in China. This will involve thoughtful city and town planning, incorporation of accessible public spaces, such as neighborhood parks, and improvement of public and traffic safety to encourage active commuting. Other policy efforts can include incentives to encourage exercise that are tied in with health insurance (e.g., reducing premiums or copayments or providing freebies). Many of these efforts can also complement other emerging needs, including environmental concerns like air or noise pollution and increasing green space.

Limitations

There are a number of limitations with this work. First, we were unable to use all observations in this work due to missing data on some measures that could not be imputed. Domestic and occupational PA were missing less than 1% of data at each wave, however, transportation and active leisure PA were missing for 15–25% of the sample in 1997 and 2000. The percentage of missing data did drop to less than 1% in all later waves. Based on discussions with the study team, we believe a large component of these high rates is interviewers skipping questions when individuals reported that they did not participate in those activities. Future research will attempt to distinguish the missing active leisure values from these zero values.

Second, over the 20 years (eight waves) of the CHNS we used, survey questions changed, and so not all questions were consistent across all waves. As a result we were unable to include certain measures, such as MET hours per week spent in housecleaning, as this question was only added in 1997. Therefore our definition of domestic PA does not include this activity. Even some questions that were asked throughout all eight waves changed. In general these changes were fairly minor, but they potentially could have resulted in differences in an individual's ability to recall PA. For example, in terms of occupational PA, the 2004 survey question about collective and household farming asked, “How many hours last week did you work?” The 2011 question asked in the past year “how many days in a week, on average” and “how many hours in a day, on average” did you work in this capacity?

Third, MET hours per week do not actually measure the energy cost of PA in individuals in ways that account for differences in body mass, adiposity, age, sex, and efficiency of movement or geographic and environmental conditions in which the activities are performed. However, these measurements are useful for providing population-level estimates on PA levels. Additionally the use of self-reported time allocation to derive these domain-specific PA measures may have resulted in bias, particularly when respondents might have been multitasking, for which we are unable to account. In particular individuals who reported participating in child care (one component of domestic PA) could have been sedentary during that time or could have participated in other activities that expend more energy. However, past studies show that the measures of domain-specific PA derived from these data are highly associated with weight and body mass index outcomes.^{16, 20, 21, 32}

Future Work

This paper focuses on adults, but it would be important to understand how PA distributions might have changed among children and adolescents over the same 20-year period of China's rapid changes in the socioeconomic conditions and urbanicity. It would also be useful to conduct comparable longitudinal quantile regression analyses for travel PA, active leisure PA, and sedentary leisure time for both adults and children when more years of data become available. Additionally it would be interesting to study how many of the PA changes observed are due to aging compared to cohort experiences and compared to period effects. The longitudinal nature of these data will allow us to explore these questions in new and novels ways.

Abbreviations

PA

physical activity

NCDs

noncommunicable diseases

CHNS

China Health and Nutrition Survey

MET

metabolic equivalent of task