Factors associated with child overweight prevalence: HIES (2009/10)
The HIES 2009/10 analysis utilizes the subsample of households that reported a child under 5 years old with complete anthropometry data (1721 out of 4191 households had at least one child with complete anthropometric data and sufficient household consumption and expenditure data). Approximately 15% of households with children under 5 years old had at least one child that was overweight (8% of households) or obese (7% of households).
Descriptive statistics suggest that sugar-sweetened beverage consumption was significantly more prevalent among households with at least one overweight (overweight includes obese) or obese child (Table
6). However, there are no significant differences between households with and without an overweight or obese child in consumption of sugar-sweetened food, high-saturated fat food or other ultra-processed foods. Disaggregating households by poor (bottom 40% of expenditure distribution) and non-poor status, consumption of both sugar-sweetened food and beverage is higher in households with at least one overweight or obese child. Among poor households, consumption of sugar-sweetened beverages is significantly higher in households with at least one obese child (with weaker significance for overweight prevalence).
9Table 6
Share of food expenditure on ultra-processed goods of households with at least one overweight or obese child
Panel A: All households with children under 5 years old |
All ultra-processed food | 7.0% | 8.2% | | 6.9% | 8.0% | * |
Sugar-sweetened food | 0.5% | 0.7% | | 0.5% | 0.6% | |
Sugar-sweetened beverage | 2.5% | 3.4% | ** | 2.4% | 3.3% | *** |
High saturated fat food | 3.0% | 3.1% | | 3.0% | 2.9% | |
Other ultra-processed food | 1.0% | 1.1% | | 1.0% | 1.2% | |
Processed food | 5.4% | 6.6% | ** | 5.3% | 6.1% | * |
Processed culinary ingredients | 4.6% | 7.5% | *** | 4.5% | 6.2% | *** |
Minimally or unprocessed | 83.1% | 77.8% | *** | 83.3% | 79.7% | *** |
Sample size | 1601 | 121 | | 1459 | 263 | |
Panel B: Non-poor (top 60% expenditure distribution) |
All ultra-processed food | 8.2% | 8.9% | | 7.9% | 9.7% | * |
Sugar-sweetened food | 0.6% | 1.2% | *** | 0.6% | 0.9% | ** |
Sugar-sweetened beverage | 3.1% | 3.5% | | 2.9% | 4.3% | *** |
High saturated fat food | 3.6% | 3.2% | | 3.5% | 3.5% | |
Other ultra-processed food | 0.9% | 1.0% | | 0.9% | 0.9% | |
Processed food | 5.6% | 5.5% | | 5.5% | 6.0% | |
Processed culinary ingredients | 4.4% | 5.0% | | 4.4% | 5.0% | |
Minimally or unprocessed | 81.9% | 80.6% | | 82.2% | 79.3% | ** |
Sample size | 851 | 67 | | 785 | 133 | |
Panel C: Poor (bottom 40% expenditure distribution) |
All ultra-processed food | 5.7% | 7.6% | * | 5.7% | 6.5% | |
Sugar-sweetened food | 0.4% | 0.2% | | 0.4% | 0.3% | |
Sugar-sweetened beverage | 1.8% | 3.2% | *** | 1.8% | 2.5% | * |
High saturated fat food | 2.4% | 3.0% | | 2.4% | 2.4% | |
Other ultra-processed food | 1.1% | 1.2% | | 1.0% | 1.4% | |
Processed food | 5.2% | 7.5% | *** | 5.2% | 6.2% | |
Processed culinary ingredients | 4.8% | 9.5% | *** | 4.7% | 7.2% | *** |
Minimally or unprocessed | 84.4% | 75.3% | *** | 84.5% | 80.0% | *** |
Sample size | 750 | 54 | | 674 | 130 | |
Recognizing that other influencing factors may shape child anthropometric outcomes, we use a probit regression to better evaluate how the consumption of different food types is associated with the probability of a household with an overweight or obese child. Descriptive statistics of the model regressors are reported in
Appendix Table 2. We run two probit models using different samples from the HIES 2009/10. The first model uses the national sample that includes 1721 households from all regions of the country. The second model focuses on households in Momase region and includes 545 households. We focus on Momase region to provide a parallel analysis with the 2018 RSFS survey data where ¾ of the survey sample was located in different provinces of Momase region.
10
The marginal effects from the probit model are reported in Table
7 (Probit model estimates are reported in
Appendix Table 3). Results suggest that a greater expenditure share of sugar-sweetened beverages in total household food purchases increases the probability of a child being overweight (which includes obese), controlling for other potential household influencing factors.
11 Specifically, a 1% increase in the expenditure share of sugar-sweetened beverages increases the probability of a child being overweight by 0.5%. For Momase region, the estimated effect is stronger, suggesting that a 1% increase of sugar-sweetened beverages expenditure share increases the probability of a child being overweight by 1%. While greater consumption of sugar-sweetened beverages has a significant positive effect on child overweight prevalence, we find no significant association when focusing on correlates of obesity prevalence. This may be due to the fact that the survey data only collected anthropometry data for children under 5 years old, resulting in a small sample of obese children in this age category. In addition, effects of increased sugar-sweetened beverages on child obesity may require a longer time horizon than the first 5 years of life.
Table 7
Marginal effects of household characteristics associated with an obese/overweight child (under 5)
Share of consumption of sugar-sweetened food in total food (%) | 0.228 | 0.294 | 0.101 | 0.935 |
(0.497) | (0.495) | (0.648) | (0.754) |
Share of consumption of sugar-sweetened beverages in total food (%) | 0.173 | 0.684 | 0.508** | 1.091** |
(0.203) | (0.424) | (0.222) | (0.507) |
Share of consumption of high-saturated fat food in total food (%) | 0.063 | −0.404 | 0.074 | − 0.011 |
(0.170) | (0.312) | (0.242) | (0.324) |
Total HH expenditure, thousands (PGK/capita/year) | −0.002 | 0.002 | −0.010 | − 0.003 |
(0.005) | (0.004) | (0.008) | (0.007) |
Household in metro area (0/1) | 0.031 | 0.019 | 0.025 | 0.015 |
(0.035) | (0.034) | (0.038) | (0.049) |
Household in urban area (0/1) | −0.030 | − 0.031 | − 0.038 | − 0.074 |
(0.020) | (0.045) | (0.026) | (0.060) |
Number of children (0–15) | 0.001 | 0.006 | 0.010 | 0.028*** |
(0.007) | (0.008) | (0.009) | (0.010) |
Household size | 0.006 | 0.006 | 0.012** | 0.004 |
(0.005) | (0.005) | (0.005) | (0.008) |
Female Household head (0/1) | −0.017 | − 0.003 | 0.034 | −0.084 |
(0.028) | (0.046) | (0.053) | (0.070) |
Age of Household head | −0.001 | − 0.001 | − 0.002* | −0.000 |
(0.001) | (0.001) | (0.001) | (0.001) |
Household head completed primary education (0/1) | 0.007 | −0.021 | 0.027 | −0.018 |
(0.019) | (0.030) | (0.022) | (0.047) |
Highland region (base = Southern) | 0.089*** | | 0.198*** | |
(0.031) | | (0.038) | |
Momase region (base = Southern) | −0.004 | | −0.014 | |
(0.024) | | (0.028) | |
Island region (base = Southern) | 0.016 | | −0.015 | |
(0.034) | | (0.047) | |
Pseudo R2 | 0.045 | 0.082 | 0.086 | 0.089 |
N Observations | 1721 | 544 | 1721 | 544 |
Factors associated with increases in soft drink consumption: RSFS (2018)
Given that there are no updated, nationally representative consumption and expenditure data in PNG since 2009/10, we utilize the IFPRI Rural Household Survey on Food Systems (RSFS) to evaluate more current household consumption trends. The RSFS collected comprehensive data on household consumption and expenditure, and anthropometry for children under 5 years of age, however several challenges should be noted when interpreting the data. First, there are very few observations of overweight or obese children given the rural focus of the IFPRI survey. Thus, we are unable to evaluate food consumption correlates of childhood overweight or obesity prevalence. Second, given that the RSFS survey focused on rural household consumption trends, less detail of urban processed food consumption is available in the data. Thus, we use information on reported household consumption of soft drinks as a proxy indicator to identify household characteristics that are associated with greater consumption of sugar-sweetened beverages. Finally, the RSFS sample was located in Momase region, and included 3 geographically diverse provinces (East and West Sepik, and Madang) and Southern Bougainville.
Data collection of the RSFS survey spanned 3 months (May–July, 2018) and collected weekly and monthly consumption and expenditure data of a detailed list of food items. Households were asked whether they had purchased soft drinks in the last month, and if so, the price and quantity of soft drinks consumed in the household during that month. In total, 1026 households were surveyed in 4 lowland provinces of PNG. This analysis utilizes 1023 observations that have complete information of the relevant variables used in the Heckman two-stage model (
Appendix Table 4).
Focusing on the first stage probit equation of the two-stage Heckman model, we select household characteristics that may affect participation (whether or not a household consumed a soft drink in the past month). We hypothesize that access to a market or trade shop that sells soft drinks is an important indicator of participation. The RSFS survey focused on rural households where walking was a primary means of mobility (the predominant mode of transportation of the Madang sample was via canoe across the Ramu river and associated tributaries), thus sex and age of the household head may affect access to and frequency of market interactions. During community focus groups, respondents identified extended travel times, insecure transportation routes and inadequate infrastructure as primary obstacles to market access. We also assume that total household expenditure, educational attainment of the household head, and the price of a soft drink would affect household consumption via household income and opportunity costs. Relatedly, brand marketing of soft drinks is aimed towards younger individuals, which may affect consumption demand differently depending on the age of the household head. Assuming that consumption is somewhat evenly divided within the household and among children, we include household size and number of children under 15 years of age as potential characteristics that may affect consumption decisions. Finally, we include a set of province fixed effects to control for differences in geographic and social characteristics and consumption preferences.
In the second stage equation (evaluating factors associated with the quantity per capita of soft-drink consumption), total household expenditure (proxy variable for household income) and unit price of a soft drink are the principal covariates we evaluate for this analysis. However, we also maintain household size in the second stage to account for potential factors influencing volume purchase per capita. Following the Heckman two-stage model specification, an exclusion restriction is required whereby at least one variable which appears in the first stage probit (participation) equation is absent in the second stage (level) equation. We assume that distance to a market is associated with whether a household purchases a soft drink, but once that decision is made, distance does not directly influence the amount purchased or consumed. Additional exclusions in the second stage equation include the characteristics of the household head (sex and age). We also assume that years of education of the household head and number of children within the household may be associated with whether to purchase a soft drink (weighing the trade-offs of purchasing a less healthy drink versus another beverage), but not on the amount purchased or consumed. We test these exclusion assumptions and results show that adding these covariates iteratively to the second stage regression has little effect on overall results (sensitivity results to the exclusion restriction are provided in
Appendix Table 5).
Table
8 reports the coefficients for the first stage probit equation and the second stage levels equation. We report both the coefficients for the levels equation estimated without correcting for selection bias (column B), and estimated by Heckman’s procedure (Column C). The coefficients in column B only evaluate households that reported purchasing and consuming at least one soft drink (412 households). Correcting for selection bias via the Heckman procedure (column C) suggests that the factors influencing households’ decision to consume a soft drink may differ from the factors that condition how much soft drink to consume, which is also reflected in the coefficient of the inverse Mills ratio. Conditional and unconditional marginal effects calculated at the mean are presented in columns D and E, respectively. The conditional marginal effect reports the effect of a covariate among consumers (those that reported purchasing a soft drink). The unconditional marginal effect considers the increased probability of purchasing a soft drink and potential sample selection bias using the Heckman two-stage estimator.
Table 8
Heckman sample selection model for soft drink expenditure per capita
Log of total household expenditure (PGK/capita/year) | 0.676*** | 23.976*** | 12.477* | 24.499*** | 19.539*** |
(0.077) | (3.123) | (7.112) | (7.247) | (2.429) |
Household-level unit soft drink price (PGK/Liter) | −0.461*** | 4.036** | 8.784*** | 0.583 | −6.708*** |
(0.082) | (1.641) | (3.215) | (3.526) | (2.435) |
Household-level unit soft drink price squared (PGK/kg) | 0.024*** | −0.066 | −0.300* | 0.124 | 0.405*** |
(0.004) | (0.092) | (0.166) | (0.182) | (0.128) |
Euclidean distance to major market towna (km) | −0.014** | | | − 0.243** | −0.299* |
(0.007) | | | (0.121) | (0.167) |
Euclidean distance to major market town squared (km2) | 0.000 | | | 0.001 | 0.001 |
(0.000) | | | (0.001) | (0.001) |
Household size | 0.069** | −1.519* | −2.933** | −1.714 | 0.375 |
(0.033) | (0.907) | (1.228) | (1.358) | (0.801) |
Household head completed primary education (0/1) | 0.076 | | | 1.359 | 1.669 |
(0.124) | | | (2.208) | (2.742) |
Household head completed lower secondary (0/1) | 0.252* | | | 4.475* | 5.496 |
(0.140) | | | (2.490) | (3.356) |
Household head is female | −0.166 | | | −2.952 | −3.625 |
(0.150) | | | (2.669) | (3.402) |
Age of household head | −0.008* | 0.129 | 0.285 | 0.138 | −0.072 |
(0.004) | (0.162) | (0.190) | (0.205) | (0.115) |
Number of children (0–15 yrs) | 0.042 | | | 0.742 | 0.912 |
(0.041) | | | (0.722) | (0.916) |
East Sepik Province (base = Bougainville) | −0.745*** | −37.029*** | −30.890*** | −42.419*** | −36.927*** |
(0.211) | (4.615) | (6.003) | (6.882) | (6.012) |
Madang Province (base = Bougainville) | −1.136*** | −37.869*** | −15.326 | −34.059** | − 38.482*** |
(0.339) | (6.205) | (14.040) | (15.290) | (7.700) |
West Sepik Province (base = Bougainville) | −1.256*** | −36.041*** | −23.537*** | −44.594*** | −42.139*** |
(0.230) | (5.454) | (9.031) | (9.904) | (6.079) |
Inverse Mills (Lambda) | | | −25.516* | | |
| | (13.969) | | |
Constant | −2.372*** | − 133.379*** | −52.404 | | |
(0.765) | (26.419) | (52.280) | | |
N Observations | 1023 | 412 | 1023 | 1023 | 1023 |
Results from the probit equation reflect expected coefficient results for significant covariates (Table
8). Households with greater total expenditure are associated with a higher probability of purchasing a soft drink. Given that soft drinks are considered a luxury good, households with greater income would have greater flexibility in expenditure decisions. However, and as expected, the probit equation also shows that the probability of purchasing a soft drink decreases as the unit price of a soft drink increases. Results also suggest that as a household head ages, the probability of purchasing a soft drink decreases. This may be that marketing of soft drinks is largely targeted at younger individuals, however further investigation of how marketing campaigns in PNG affect consumer choice behavior is warranted.
The distance to the nearest major market town is negatively correlated with the probability of purchasing a soft drink. The probability of purchasing a soft drink decreases by almost 2% for each 1 km increase in distance to a major market town (and remains significant at the 10% level when we test for joint significance with the quadratic term). Given limited information on availability of soft drinks in each of the sample communities, we use a distance measure to larger market towns, however more precise information on soft drink availability and location could improve this estimate.
12
When evaluating the second stage Heckman regression results (column C), household expenditure on soft drinks increases until the unit soft drink price reaches about 15 kina per liter, at which point demand decreases. An increase in total household expenditure is also associated with an increase in per capita soft drink consumption (at the 10% significance level), while an increase in the household size is associated with a decrease in per capita consumption. This follows the assumption that consumption is relatively evenly distributed among household members.
Focusing on the unconditional marginal effects, total household expenditure has a strong and positive association with per capita consumption of soft drinks (this holds true for the conditional marginal effects of the censored sample of consumers only). While the price of a soft drink is negatively correlated with the quantity consumed (coefficient of − 6.708), analysis shows that total household income has a quantitatively larger (and positive) association with soft drink consumption. This suggests that policy aimed only at adjusting the price of unhealthy food items may not be enough to curb the increasing demand for ultra-processed food items. Finally, and as expected, results from both conditional and unconditional marginal effects demonstrate that greater distance from a major market is associated with a decrease in the per capita consumption of soft drinks.