In this section, we compare the estimates of the incidence of poverty (P0) derived from our application of the small-area estimation method to the estimated incidence of pov- erty produced by the Ministry of Labor, Invalids, and Social Affairs (MOLISA). As described earlier, there are a number of dif- ferences in the definition of poverty and the
SPATIAL PATTERNS IN POVERTY AND INEQUALITY 35
Figure 3.19 Gini coefficient of inequality as a function of the share of the population in urban areas
data collection methods. Some of these dif- ferences are summarized below:
• Our definition of poverty uses as the welfare indicator the value of per capita consumption expenditure, including the value of subsistence food production and the imputed rental value of owner- occupied housing. In contrast, MOLISA uses per capita income as its welfare indicator.
• To adjust for regional differences in the cost of living, we use a set of regional and monthly price indexes calculated by the GSO for the 1997–98 VLSS analysis. These price indexes are based on the cost of a basic consumption bas- ket in the urban and rural areas of each region. In contrast, MOLISA adjusts for the local cost of living by express- ing per capita income in terms of the number of bags of rice it will buy at local prices.19
• Our poverty line is equal to the “overall poverty line,” defined as VND 1,789 million per person per year in real consumption expenditure. MOLISA defines the poverty line in terms of the number of bags of rice, although the number varies somewhat from one province to another.
• We define the poverty rate in terms of the percentage of people living in households whose per capita expendi- ture is below the poverty line. MOLISA defines the poverty rate in terms of the percentage of households below the poverty line.
• Our poverty estimate for each district is based on the characteristics of house- holds in that district in the 1999 Popu- lation and Housing Census, given the relationship between per capita expen- diture and those household characteris- tics in the 1997–98 Vietnam Living Standards Survey. The MOLISA esti- mates are based on assessments of
MOLISA field staff in each commune, applying national and provincial guide- lines to identify poor households (for a description of the field work, see Conway 2001).
Do these methodological differences re- sult in different estimates of the incidence of poverty (P0) at the district level? As shown in Figure 3.20, the MOLISA poverty esti- mates are generally lower than those gen- erated by the small-area estimation method used in this report. The median value of the MOLISA poverty rates is 15 percent, com- pared to 41 percent for our poverty estimates. This difference is not particularly surprising because the two estimates are based on quite different poverty lines. Given the wide range of views about how to construct a poverty line, there is little to be gained from debates over the “true” poverty rate. It is more im- portant to examine whether the spatial pat- terns in poverty are consistent between the two methods.
What is surprising is that there is very little correlation between the district-level poverty rate estimates produced by MOLISA and the poverty rates estimated by this study (the R2of a linear trendline is just 0.17). To illustrate the disagreement in the estimates, we consider two districts in which the con- trast between the two methods is the great- est. In the upper left corner of Figure 3.20 is a dot representing Bat Xat district, located in the northwest corner of Lao Cai. Accord- ing to our estimates, Bat Xat district has a poverty rate of almost 82 percent. By con- trast, the MOLISA poverty estimate for the district is less than 6 percent. Given that Bat Xat is in a remote portion of one of the poorest provinces in Vietnam, we would ex- pect the poverty rate to be relatively high. In the lower right corner of Figure 3.20 is the urban district of Nha Trang, on the central coast of Kanh Hoa province. The MOLISA poverty estimate for Nha Trang is 68 per-
cent, whereas the estimate produced in this report is just 15 percent. Because Nha Trang is a popular beach resort town, benefiting from both local and international tourism, a relatively low poverty rate would be expected.
Clearly, the choice of poverty estimates can make a large difference in terms of the
targeting of poverty alleviation programs. Further research is needed to resolve the discrepancies between these two poverty estimates. One approach would be to select districts where the two estimates vary widely (such as the two cited above) and collect pri- mary or secondary data to determine which estimate conforms more closely to reality.
SPATIAL PATTERNS IN POVERTY AND INEQUALITY 37
Figure 3.20 Comparison of poverty rates (P0) from MOLISA and from small-area estimation methods