Page 80 - IJPS-8-1

International Journal of
Population Studies Urbanization and body weight

where z is the observed sample mean urbanicity index points in time (i.e., unbalanced longitudinal data with
j
score for community j throughout its entire observation respect to time); α is the mean intercept, indicating the
0
period. Estimates of disaggregated within- and between- average urbanicity index across all communities and all
community differences can be obtained by regressing the time points; α is the mean slope, indicating the average
1
outcome variable on  z and z . Let y be a continuous rate of change in urbanization over time; δ represents
jt
j
ijt
0j
body weight measure for individual i living in community the deviation of community j’s intercept from the grand
j at time t, and X be a set of individual- and household- mean α ; δ represents the deviation of community j’s slope
1j
0
ijt
level control variables. A three-level random effects model from the mean slope α ; and the residual term r captures
jt
1
is fitted to incorporating the hierarchical data structure as the within-community variability of urbanization around
the following: the community-specific mean. Between-community
differences at any given point in time are determined by
y 0 X 1 z  z u u ijt (2) both between-community variability in the intercept (δ )
i
j
j
2
ijt
ijt
jt
0j
and between-community variability in the slope (δ ).
1j
where β is an intercept, u and u represent person- and Instead of centering at the observed community-specific
j
i
0
community-level random intercepts, respectively, and ε mean as in Equation (1), each community- and time-
ijt
denotes residues. The coefficient γ captures the relation specific z can be deviated from its model’s implied value
jt
1
between average levels of urbanicity index and average (Curran & Bauer, 2011):
levels of body weight measures pooling over communities.  ˆ  ˆ 
− δ
In contrast, the coefficient γ captures the mean relation jt = ˆ r jt − z  ˆ 0 +  α 0j  1 + ˆ α δ 1j  T jt (4)


2
between a given community’s time-specific deviation in
ˆ
ˆ
ˆ
ˆ
urbanization (from its overall level of urbanization) and where α , δ , ˆ α , δ , and r are the estimates of the
jt
0
1
0 j
1j

the time-specific body weight status among the residents coefficients defined in Equation (3). Between-community
living in that community. differences averaged over time can be represented by:
The validity of the standard method relies on the ˆ 0 j = δ j − (z 0 ) − ˆ α ( 1 + ˆ α δ ˆ 1j ) j (5)
T
assumption that the community-level urbanicity index is
unrelated to time (Curran & Bauer, 2011). In other words, where T is the mean value of time for community j and
j
ˆ
ˆ r
it assumes potential growth in the body weight outcome allows unbalanced time structure. The values of δ and jt
0 j
(i.e., y ), but no systematic change in urbanization itself, can then be substituted into Equation (2) for z and  z ,
ijt
aside from random variations, with the passage of time. respectively. j jt
This assumption is unlikely to hold because not only has
China been one of the fastest urbanizing countries in the Three multilevel random-intercept models, in which
world since the 1980s (United Nations, 2015) but also the repeated measures (level-1) were nested within individuals
sampled communities in CHNS have gained considerable (level-2) which, in turn, were nested within communities
growth in their urbanicity index scores between 1991 and (level-3), were fitted to each outcome variable. Model
2015 (see the results section). 1 assessed the overall association between body weight
status and the urbanicity index. Model 2 applied the
The second method, referred to as growth curve standard method to rescaling urbanicity index scores and
disaggregation, allows the community-level urbanicity disaggregating the overall association into between- and
index to change as a function of time and allows the rates within-community differences. Model 3 adopted the
of change to vary randomly over communities. These growth curve disaggregation method instead. Multilevel
assumptions represent a more realistic scenario, in which the linear models were estimated for continuous measures of
initial level of urbanicity and the pace of urbanization over body weight status, while multilevel logistic models were
time differ across communities. A growth curve model for estimated for binary measures of body weight status.
the relationship between the time-varying community-level
urbanicity index and time can be specified as the following: 4. Results
T
z 0 1 T j 0 1 j jt r jt 4.1. Descriptive statistics of body weight status
jt
jt
( ) ( )T r (3) The gender-stratified secular trends of the continuous and
0 0 j 1 1j jt jt dichotomous measures of body weight status presented
where T is the measure of time at time t for in Figures 1 and 2, indicate that both Chinese men
jt
community j, which permits the possibility that all the and women have grown heavier over two decades. For
sampled communities are not surveyed at all of the same example, Figure 1 shows that the average BMI increased
Volume 8 Issue 1 (2022) 74 https://doi.org/10.36922/ijps.v8i1.334