Contents - Previous - Next

This is the old United Nations University website. Visit the new site at http://unu.edu



Results

Income cut-off

When the sample is broken at 50% (Y4,000), the Chow test fails to indicate any difference in the parameters for all three nutrients (table 3). The null hypothesis is accepted. The two sets of regression coefficients were almost the same for all three nutrient intakes. When the sample is broken at 75% (Y5,500) and 80% (Y6,000), the results are mixed. For calories and % fat, the Chow test shows statistical significance. There are differences in calorie intake and % fat intake between the low- and high-income groups. When the sample is broken at 85%, the test shows significant difference (the null hypothesis is rejected) at the 1% significance level for all three nutrient intakes. This shows a statistically significant structural difference in the regression equations for all three nutrients.

It is not surprising that there are different cut-off points for different nutrients, because the impact of income is different. Thus, the cut-off should be selected based on the purpose of the study. Because fat intake is closely associated with socioeconomic development and is important to our concern in researching how diets in developing countries are becoming modern, its significance should be specified. Thus, only when the Chow tests show all three nutrients significant do we define the consumption behaviour as different. Thus, Y7,000 in 1991 terms was selected as the switching point. Marked differences are seen in income and other socio-economic status factors for families below and above Y7,000. Household income is Y3,293 for the lower-income and Y10,709 for the higher-income group; the former is also less educated, more rural, and more northern than the latter.

TABLE 3. Chow statistics for different income cut off points

Cut-off point (yuan RMB) Fat Calories %
4,000 0.84 1.21 1.7
5,500 1.14 1.88* 1.82*
6,000 1.65 2.04* 1.93*
7,000 2.72** 5.33** 3.75**

*Significant at 5% level.
**Significant at 1% level.

Effect of Income on nutrient Intake

Tables 4-6 present the results for fat, calories, and percentage of calories from fat respectively, estimated by three methods. For each table, the first two columns are the results estimated from the first two methods, in which only one regimen was specified. When one regimen is specified, most of the coefficients are statistically significant for the three nutrient intakes. Estimation by equation 1, using income as a continuous variable, shows statistically significant effects for most explanatory variables, which suggests a positive effect on the fat, calories, and % fat consumed. When the effect of these variables is tested in equation 2, which includes interaction terms between income and all explanatory variables, we find that most of the interaction terms have no significance and should be dropped from the final model. This result appears to be contrary to the theoretical hypothesis that the impact of income differs at different levels.

However, if the whole population has two different consumption structures switching at Y7,000- with 85% of the population governed by a similar preference structure and impact of income-the results are quite different. The diet of higher-income persons is not affected by most of the explanatory variables. Only a few factors significantly affect the diet of the rich. This is not surprising, since they do not face income or other constraints in their consumption decisions.

TABLE 4. Estimates of three methods modelling fat intake (g)

Explanatory variable One regimen One regimen with interaction Two regimens
Low-income
( < Y7,000)
High-income
(
Y7,000)
Constant 87.7949 (25.46**) 85.4264 (15.92**) 84.3025 (22.01**) 93 7475 (10.63**)
Age (yrs) -0.0710 (-1.2) -0.0843 (-094) -0.1095 (-1.70) -0.0649 (-0.45)
Male -6.4611 (-4.4**) -8.3815 (-374**) -7.0418 (-4.40**) -1.7787 (-0.50)
Education (yrs) 1.2020 (8.0**) 1.3241 (5.62**) 1.0124 (6.27**) 1.1474 (2.73**)
Smoking 6.3474 (4.17**) 5.7637 (2.43**) 6.6148 (4.00**) 5.1517 (1.39)
Alcohol use 2.0382 (1.6) 1.9932 (0.98) 2.7242 (1.97*) -1.3698 (-0.43)
Family size -4.3897 (-12.3**) -3.0402 (-5.32**) -4.9022 (-11.93**) -4.2194 (-5.75**)
Urban 1.0197 (0.77) 0.1563 (-0.07) -0.4842 (-0.33) 3.7135 (1.21)
Northern -13.0186 (-12.23**) -14.9239 (-8.25**) -13.3567 (-11.76**) -11.9550 (-4.07**)
Moderate activity 4.2950 (2.82**) 3.0808 (1.18) 5.0023 (2.96**) 0.8339 (0.25)
Heavy activity 6.3921 (-4.2**) -12.4558 (-5.13**) -5.6709 (-337**) 1.8469 (0.51)
Income 0.0008 (6.14**) 0.0017 (1.88) 0.0034 (980**) -0.0002 (-0.77)
Income x age     -0.0000 (-0.18)        
Income x male     0.0005 (1.23)        
Income x education     -0.0000 (-1.01)        
Income x smoking     0.0002 (0.38)        
Income x alcohol use     0.0000 (0.03)        
Income x size     -0.0003 (-305**)        
Income x urban     0.0003 (0.75)        
Income x northern     0.0005 (1.32)        
Income x moderate activity     0.0002 (0.49)        
Income x heavy activity     0.0014 (3.43**)        
F value 58.71   33.38   60.16   5.54  

Numbers in parentheses are t statistics.

* Significant at the 5% level.
** Significant at the 1% level.

Difference in effects on high- and low-income families

Tables 7 and 8 summarize the results of significance tests for a two-regimen model. The sign and statistical significance of coefficients produced for the one-regimen model is almost identical to that produced from the low-income regimen. This may be explained by 85% of the sample falling below Y7,000.

Continue


Contents - Previous - Next