The comparison of poverty rates across two countries, or across regions within a country, is a common occurrence in analysis produced at the World Bank and other development agencies, as well as in published academic papers. For any poverty comparison to have meaning, however, the analyst needs to norm the various observed states of the world to a known standard of living. In other words, any poverty comparison is meaningful only if it can be said to achieve welfare consistency.
Welfare consistent comparison across space requires local price data so that levels of living measured in dollars earned, or dollars consumed, do not get confounded with the differences in price levels across localities. After all, a poor area may be only nominally poor due to a low cost of living, but not any poorer in real terms. How would we know the difference without the right prices?
The problem is, though, that information on local prices is missing more often than not. Most Household Income and Expenditure Surveys (HIES) – the workhorse survey for poverty inference in much of the developing world – do not field a parallel local price survey. This forces local adjustments to rely on non-price information such as unit values or Engel curves. And unfortunately, these adjustments are, speaking euphemistically, second best and suffer from various biases.
It is also a cruel irony that national situations likely to produce the greatest differences in local prices, due to a lack of market integration across villages and regions suffering from the poor infrastructure found in the least developed countries, are also the least likely to have available and reliable information on local prices.
To address this gap, John Gibson and Trinh Le analyze a spatial cost of living survey conducted in Vietnam, done in conjunction with the 2010 VHLSS. This survey was fielded in over 1000 communes – the lowest official administrative unit in the country – and consequently covers almost one-eighth of all in Vietnam. A full-fledged price survey collected price information on 64 narrowly defined items, with photos of each item to facilitate standardization. In addition, in each commune a focus group was held with key informants, where the same photos were used to elicit expert opinion on the typical, highest, and lowest price for which the item currently sells. As the authors write, the experts were identified as follows:
The structured focus groups were based on three participants per commune, selected in a systematic way. One was a 30-35 year-old woman, expected to be the most informed about prices of children’s products and personal hygiene items, and to be busier than others and so less likely to haggle prices. The second was a 45-50 year-old man, expected to know more about the prices of alcohol, tobacco, construction items and durables. The third was a 40-50 year-old woman, expected to know more about food prices and to have more time to haggle over the prices. … The focus groups yielded 54,200 observations on local prices .. and so using local experts provided 99.3% of the price points obtained [through the market survey].
Since this study also has the benchmark price survey, Gibson and Le can compare the expert opinion-based prices to the gold standard as well as to the two main methods used for welfare comparisons in the absence of price data: the Engel curve spatial deflator and the unit value approach.
The Engel curve approach to spatial price deflation starts with the assumption that the same food share would arise at the same level of real income regardless of location. One early application used this assumption to impute bias In PPP deflators by exploiting the deviations from the base food share across countries.
In this case study of regional price variation in Vietnam circa 2010, the Engel curve deflator gives wildly different results than the benchmark price index. The correlation coefficient across the 12 regions of Vietnam is only 0.13. On the other hand, the price index based on expert informants is highly correlated with the benchmark price survey, with a correlation coefficient of 0.98. The sum of squared differences in regional price levels between the benchmark and expert opinion is 14, while the same statistic comparing the Engel curve with the benchmark price survey is over two orders of magnitude worse (!), at 5717.
The Engel curve deflator also makes inequality look substantially higher, with a Gini coefficient estimated at 0.48, while the Gini based on expert opinion in 0.41. The Gini based on the benchmark price survey is also 0.41. The simple expert opinion survey looks very promising.
The other common approach for understanding regional price differences in the absence of prices involves the analysis of unit value data. Unit value data is as ubiquitous as consumption data since the unit value of a good can be estimated directly from the HIES (by dividing the stated expenditure on the good by the stated quantity consumed). However, using the unit value as a price proxy often confuses the standard of living difference over space with the cost of living difference over space. Why? For several reasons but most importantly because, even within narrowly defined categories of goods, households choose both the quantity and the quality of the good consumed. The elasticity of quality with respect to income or price is often substantial – see this paper by John Gibson and Bonggeun Kim that estimates these quality elasticities with the same VHLSS data.
Gibson and Le generate a food price index with unit value information and compare it to both the benchmark price index and the expert opinion-based index. The correlation between the benchmark and the expert opinion price index is 0.98. The correlation between the unit value index and the benchmark is 0.70. In addition, the SSD for the unit value index is 10 times larger than the expert opinion SSD. Given the narrowly defined goods in the expert survey, it is apparent that quality effects do not confound the opinion price estimate as they do the unit value measure. The simple expert opinion survey looks to be even more promising.
A major constraint on valid poverty and inequality analysis is the lack of spatially disaggregated price data. It turns out that local experts, at least those in Vietnam’s communes, have accurate knowledge of prevailing prices. Hopefully this approach will soon be tested/replicated in diverse settings.
The price expert survey is also cost-effective: Gibson and Le estimate that the opinion survey cost about one-sixth the price of the benchmark market price survey. The use of mobile technologies may be able to further reduce the cost and increase the frequency of price collection without accuracy taking a hit. In fact, of course, I wouldn’t be surprised if several mobile technology-based efforts are already underway.