The environmental factors scanned in this study, among total of 13 dimensionalities and 45 provincial variables, are listed in Table 1.
Dimensionality Level Macro Environmental Factor Micro Environmental Factor Economic status Provincial GDP Per capita income Per capita GDP Demographical data Provincial sex ratio Urban population density Dependent children ratio Urban-rural ratio Parents' educational level Illiteracy and semi-literacy rate School policies Physical education policy Health education policy Nutrition education policy Food industry development Annual production of dairy products Annual output of soft drinks Transportation capability Provincial cargo logistics transfer Passenger transport volume Media and marketing TV coverage Per capita consumption of food in province Grains, meat, vegetables, oils, sugar Grains, meat, poultry, vegetables, oils, sugar Dietary behaviors Dining out Healthcare services Average number of doctors Sanitary conditions of living environments Popularity of tap water Modified water source utilization ratio Flush toilet utilization ratio Sports facilities Per capita public green space area Sports equipment per 100 households Per capita paved road area Sedentary behaviors Public transportation (bus, taxi) Private cars, washing machines, tractors Screen viewing (mobile phone, TV, computer) Electricity owned per 100 households
Table 1. Environmental Factors Scanned in the Study
We checked correlation among independent variables and formed combinations of variables with relatively small internal dependencies by dimensions. We used combinations of or single independent variables supported by expert opinions or relevant literature by dimensionalities, and the dependent variable (obesity prevalence in the 7-17 age group in the 12 CHNS provinces) as a training sample to test for linearity, independence, normality, and equal variance, and to perform a single factor linear regression. We obtained 12 independent variables that were eligible for simple linear regressions with statistical significance (P < 0.1) or that were supported by expert interviews or relevant literatures[2, 7, 9-11]. Table 2 shows the dependent variable and the 12 selected independent variables of 11 dimensionalities.
Variable Label No. Provinces Min Median Max M SD Provincial GDP (100 million yuan)1 gdp11* 31 507.46 10368.60 46013.06 14098.13 11401.35 Dependent children (%)2 chilrati11* 31 10.41 22.70 38.10 22.75 7.06 Modified water source use (%)3 mdwaterr11** 30 78.50 97.15 100.00 94.50 6.13 Healthcare (no. doctors/1, 000 persons)4 hlthcare11 31 1.04 1.73 5.24 1.99 0.82 Illiteracy rate of mothers (%)5 m_illi11 31 2.75 7.09 38.13 8.71 6.67 School policy score (points)6 schplcy11 31 0.00 3.00 9.00 2.61 2.43 Food industry level (10 thousand tons)7 foodindu11*** 31 0.46 31.00 345.36 69.61 85.09 Transportation capability (100 million FTK)8 ft_kilo11 31 38.50 2840.00 18918.20 4210.03 4143.69 Media (TV coverage %)9 media11* 31 91.40 97.70 100.00 97.29 2.13 Obesity prevalence (%)10 pctn11_4 12 0.00 11.25 24.66 10.65 6.57 Edible oils per capita (g)11 oilrpd11*** 31 8.03 18.11 29.42 17.98 6.25 Vegetables per capita (g)12 vegerpd11 31 48.11 221.64 411.86 240.63 86.45 Washing machines (no. units/100 households)13 washmr11* 31 8.47 65.75 101.47 62.42 25.74 Note. 1Gross domestic product by province in units of 100 million yuan; 2Ratio of people aged 0-14 to people aged 15-64 by province; 3Percentage of improved water source use by province; 4Average number of doctors per thousand people by province; 5Percentage of illiterate and semi-literate mothers among all those aged 15 and over; 6Comprehensive point score for school policies on physical education, health education, and nutrition education; 7Dairy products were used to represent the developmental level of the food industry by province; 8Cargo transportation by province in units of 100 million freight ton-kilometers; 9TV coverage was used to represent the developmental level of the marketing industry by province; 10Prevalence of obesity in children/adolescents aged 7-17 in 2011 by province; 11Average daily oil consumption in grams per person by province; 12Average daily consumption of vegetables in grams per person by province; 13Number of washing machines owned per 100 households by province. The total average obesity prevalence of the 1, 416 participants in the 12 surveyed provinces was 9.11%. In simple linear regression: *P < 0.1, **P < 0.05, ***P < 0.01.
Table 2. Distribution of Observed Obesity Prevalence in 12 Provinces and Environmental Variables in 31 Provinces
We fitted an analytical model with PLSR according to Y = β0 + β1X1 + β2X2 +...+ βkXk + ε, in which the dependent variable Y was provincial obesity prevalence and the independent variables X1, X2, …, Xk were provincial combinations of or single independent variables that had statistical significance (P < 0.1) or that were supported by expert opinions or relevant literatures. We developed parameter estimates and a hierarchy of environmental obesity factors as in Table 3.
Parameter Original Estimate Standardized Estimate VIP Intercept -36.67779825 0.0000000000 gdp11 0.00006046 0.1058703859 1.03560 oilrpd11 0.16333126 0.1569353660 1.53510 vegerpd11 -0.00603443 -0.0692222505 0.67712 chilrati11 -0.07960009 -0.1043472277 1.02070 foodindu11 0.01212167 0.1412824089 1.38199 ft_kilo11 0.00010370 0.0853258715 0.83464 media11 0.31446806 0.0987461071 0.96591 washmr11 0.02556610 0.1012743674 0.99064 m_illi11 -0.12781732 -0.0845105340 0.82666 schplcy11 -0.09646236 -0.0415767208 0.40669 mdwaterr11 0.13622520 0.1158640006 1.13335 hlthcare11 0.36085791 0.0659804330 0.64540 Note. VIP: Variable importance in projection.
Table 3. Parameter Estimates of the Obesity Prevalence Analytical Model with PLSR Based on 2011 Data
We found 2 kinds of environmental factors, obesity-promoting and obesity-impeding. By descending order of VIP, the obesity-promoting factors were edible oils per capita (average daily consumption in grams per person by province), food industry level (production of dairy products in units of 10, 000 tons by province, representing the development level of the food industry), modified water source ratio (percentage of households with improved water sources by province), provincial gross domestic product (GDP), washing machines (number of units owned per 100 households by province), media (TV coverage by province, representing the development level of the marketing industry), transportation capability (cargo transportation in 100 million freight ton-kilometers by province), and health care service (number of doctors per 1, 000 people by province). By descending order of VIP, the obesity-impeding factors were dependent child ratio (ratio children aged 0-14 to people aged 15-64 by province), illiteracy rate of mothers (number of illiterate and semi-literate mothers relative to total number of women aged 15 and older by province), vegetables per capita (average daily consumption in grams per person by province), school policy score (comprehensive point score of physical, health, and nutrition education).
From the predictive model we obtained a posterior distribution model through Bayesian analysis using the GENMOD Procedure in SAS 9.4. We fitted the predictive model with a DIC of 61.96 and with statistically significant parameter estimates (P < 0.05); we have 95% confidence that the intervals contain the overall parameters. Results of the parameter estimates are shown in Table 4. The predictive model yielded a spatial inference of the distribution of provincial obesity prevalence in which three provinces had predicted values above 15%, 10 provinces had values of 10%-15%, nine provinces had values of 5%-10%, and nine provinces had values below 5%.
Parameter Estimate SD Wald 95% CI Intercept 646.1034 161.7405 329.0979 to 963.1089 logoilrpd11 16.7047 1.8363 13.1057 to 20.3037 loggdp11 4.9868 0.9893 3.0477 to 6.9259 logmedia11 -162.2840 37.0773 -234.9540 to -89.6134 logwashmr11 2.9964 1.0572 0.9243 to 5.0684
Table 4. Parameter Estimates of the Obesity Prevalence Predictive Model with Bayesian Analysis Based on 2011 Data
Using the corresponding proportions of children and adolescents in each of the 31 provinces as weights accounting for the total population of the 31 provinces of China (based on data from the Sixth National Population Census of the People's Republic of China), we calculated that the total inferred average obesity prevalence among people aged 7-17 in the 31 provinces was 9.69%, higher than the average prevalence of 9.11% found in the 12 CHNS provinces surveyed in 2011.
We created an inferred child and adolescent obesity prevalence map of the 31 Chinese provinces using a GIS and found that obesity was clustered in northern and eastern provinces of China (Figure 1).