Page 78 - IJPS-8-1

International Journal of
Population Studies Urbanization and body weight

index [BMI] ≥25) was significantly higher for residents of index) and obesity (BMI >30) among adults aged 18 years
communities in the top tercile. Thompson et al. (2015), and older. Gordon-Larsen et al. (2014) identified latent
who analyzed the adult sample from the 2009 wave of class trajectories of adult BMI in the 1991–2011 CHNS
the CHNS, found that living in a community in the top data and found that baseline urbanicity was not associated
tercile of an urbanicity index was associated with a greater with upward BMI trajectories for men or women. After
likelihood of having a high waist-to-height ratio (WHtR) dividing communities into terciles of urbanicity scores,
(>0.5) than living in a community in the bottom tercile they found a positive association between the 10-year
for men but not for women. The cross-sectional nature of change in community urbanicity scores and residents’
these studies means that the findings are based on between- rate of being overweight (BMI ≥25) in the least urbanized
community differences only. communities at the baseline, but a negative association in
The most common approach for longitudinal analyses the most urbanized communities at the baseline.
of the urbanization-weight association is to estimate a Despite using these different estimation strategies, the
hierarchical linear model (for continuous measures of previous research has paid little attention to explicitly
body weight) or logistic model (for dichotomous measures modeling the within-community process of urbanization
of body weight) that adjusts for the multilevel data separately from between-community differences in the
structure in CHNS, nesting measurement occasions (level baseline level or the rate of urbanization. The mixed
1) within individuals (level 2), who, in turn, are nested findings in the literature may be partially explained
within communities (level 3). Applying this approach to by different estimation strategies that make inferences
the adult sample (ages 18–59) in 1991–2009 CHNS, Jaacks about different components of the urbanization process
et al. (2013) found that a two standard deviation increase in that are related to body weight status in different ways.
community urbanization was related to a 0.23 unit (kg/m ) For example, the hierarchical modeling approach draws
2
increase in BMI and a 3% increase in the odds of being inference from pooling together within- and between-
overweight (BMI ≥25). The same approach has also community variability over time (Monda et al., 2007;
been applied to examining the diet and physical activity Jaacks et al., 2013; Ng et al., 2014), whereas the difference-
mediators through which urbanization affects body weight in-differences approach and its variant adopted by other
status. For example, using 1991–1997 CHNS data, Monda researchers (Jones-Smith & Popkin, 2010; Van de Poel et al.,
et al. (2007) found that a 1-unit change in urbanization 2012; Gordon-Larsen et al., 2014) adjusts for between-
score was related to a 7% and a 6% increase in the odds community difference in the baseline level of urbanization
of light or moderate (versus heavy) occupational activity but leveraged on between-community difference in the
for men and women ages 18–55, respectively. Using the rate of urbanization.
1991–2011 CHNS data, Ng et al. (2014) measured physical
activity for adults aged 18 – 60 at work and at home using The disaggregation analysis in this study will shed new
the metabolic equivalent of task hours per week. They light on the methodological issues (different modeling
found that higher urbanization scores were associated strategies that may or may not correctly specify different
with lower occupational physical activity for both men and processes of urbanization) that lead to inconsistent findings
women, and lower domestic physical activity for women, obtained from the same data source in the literature. More
but higher domestic physical activity for men. importantly, it will help clarify the mismatch between
the conceptual model of within-community process
However, alternative model specifications have led to (i.e., in situ urbanization) and the statistical model that
inconclusive findings. Focusing on a subset of women who conflates within- and between-community processes.
had their BMI measured in both the 1991 and 2004 waves The disaggregation of between- and within-community
of the CHNS, and who were not overweight or obese in components permits a precise and unambiguous test of the
1991, Jones-Smith and Popkin (2010) estimated a logistic longitudinal association between in situ urbanization and
model with robust standard errors and found significantly weight gain.
greater odds of being overweight (BMI ≥25) for women in
communities with greater increases in urbanization scores 3. Data and methods
over the study period than for women in communities 3.1. Sample
within the lowest quintile of urbanization in 1991 and that
experienced no increase over time. In contrast, applying a Subjects for this study were adults ages 18–65 in the CHNS.
difference-in-differences estimator with fixed-effects to the Although the CHNS data are not nationally representative,
1991–2004 CHNS data, Van de Poel et al. (2012) found no households were randomly selected from a diverse set of
significant association between urbanization (defined as nine provinces in northeast, central, and south China.
movement from below to above the median of an urbanicity Together, these nine provinces are home to more than

Volume 8 Issue 1 (2022) 72 https://doi.org/10.36922/ijps.v8i1.334