HANDBOOK OF THE INTERNATIONAL COMPARISON PROGRAMME

3. TASKS RELATED TO PRICE DATA


1. Outline of price requirements
2. Some principles of item selection
1. Temporal and ICP price sampling
2. Importance and identity
3. Number of items per basic heading
3. Obtaining national average annual prices
1. Annual average prices
2. National average prices
1. With outlet price variation and quantity weights
2. With outlet variation but without quantity weights

3. TASKS RELATED TO PRICE DATA

93. The quality of ICP comparisons depends heavily on the individual item prices and the price parities at the basic heading level. This chapter is devoted to the problems of country selection of representative items (specifications) and the development of national average prices. The remainder of this section outlines the general character of the sequence of item selection, price collection and review of prices submitted by participating countries and discusses the core commodities. Section B deals with some principles of selection of the price sample by countries, and section C deals with the principles and practice related to obtaining national annual averages for the price sample for the ICP. The remaining parts of the price discussion are taken up in chapter IV, which deals with special pricing problems as they occur for specific categories of expenditure, while chapter V outlines how the price as well as the expenditure data submitted by countries are then processed to generate the ICP outputs.
94. Selecting and pricing representative items constitutes the most difficult, and typically quantitatively largest, most costly and time-consuming part of ICP work for national statistical agencies and ICP organizers. The scenario may differ from comparison to comparison; however, the main stages of this work are the same in all comparisons. These are:


1. Development of a list of representative items to be priced by a country; this list will be based on important items that are common to the existing national price data archives in the region or country group and draw on the core commodities;
2. Collection of the price data not readily available from regular surveys;
3. Submission of the national average prices for the items selected to the ICP organizers appropriate to the country;
4. Checking the price ratios and parities at the basic heading level and correcting any unsatisfactory basic data.


95. The selection of the representative items to be priced should take place before the reference period so that arrangements can be made for the inclusion of additional price observations in national surveys, when necessary. Prices collected for national statistical purposes, such as those used in the calculation of consumer price indexes, can be used for ICP purposes. Indeed most of the ICP specifications incorporate items collected in national price surveys of participating countries. However, in most countries, some additional pricing will be necessary. For national purposes, the items selected have to represent goods and services that are comparable between time periods in the same country; for ICP purposes, the selected items have to represent goods and services that are comparable across countries. Thus, items that are important I/ in a given country but that do not exist in any of the partner countries are not useful for ICP purposes. On the other hand, items that are comparable across countries but that are not in the sample of national price observations may be very useful for ICP purposes. To sum up: when a country participates in the ICP, it will need to enhance its national collection of price data to obtain the necessary ICP price observations; and it will suggest for inclusion in the ICP regional and core specifications some of the items it collects for national purposes that are not currently in ICP core or regional lists. The incorporation of ICP specifications into national statistical programmes and the continual modification of the ICP specifications according to the pricing experience of participating countries are strongly encouraged in order to increase the overlap and up-to-date character of items priced for national and international purposes.
96. The way in which the ICP facilitates price collection is through its written specifications, which describe the items that are suggested for pricing. At the world level, there are specifications that have been found during the various benchmarks to be items suitable for comparison across a number of countries. In phase VI of the ICP a subgroup of these specifications has been designated core commodities and they will serve as a basis for overlap of price observations across the regional or group comparisons. Countries in each region will be asked to include in their pricing as many of the core commodities as possible that are available on their markets. Within each regional group or other grouping of ICP countries there will be a set of commodities that will include many items specific to the country group plus most of the core commodities.
97. The core commodity list is a list of goods and services that in the experience of ICP countries, have been found to be common across large numbers of countries in all regions of the world. It is necessary that some countries in all regions provide enough prices from the core commodity list to provide coverage of all the basic headings to permit linking of comparisons between each regional or other grouping of ICP countries.
98. As with items entering national time-to-time indexes, the core commodity list will undergo changes as goods and services appear and disappear from markets. One important role of the core commodity list is to provide a common basis for the list of items to be priced in each region. Interactions between country groups and the central ICP staff generate changes in the core commodity list. Thus, it should be looked upon as a list of goods and services that appear widely available in markets around the world, but a list that will change as the economies of the world change.
99. An example of a specification of a core commodity is a yellow onion:

Item code:		Onion, yellow
	Description:		Common yellow dry cooking onions, globe in shape, all varieties. About 5 cm in diameter.
	Exclude:		Bermuda, Spanish and green onions.
	Unit:		One kilogram.

This specification is found in countries in all regions of the world, and so it provides one natural bridge across countries within a basic heading, vegetables. As will be discussed below, even this item may raise questions in some countries that require special treatment.

100. An example of a specification that would be important for East Asian countries in ESCAP is bean curd:

Item code:		Bean curds
	Description:		Curds from soy beans sold moist. If purchased as package, 300-500 g.
	Unit:		One kilogram.
	Specify:		(1) Sold in bulk (2) Packaged

Bean curds are a well-known item of consumption, and while they are found in most countries of the world, they have up to now not been widely enough consumed to be proposed as a core commodity.

101. These specifications illustrate some of the detail involved in the 2,000 or more specifications that are part of the ICP framework. In each of these specifications, the unit for which the price is to be reported is indicated. Often, the price quotation is for a different unit, as is the case for packaged bean curd where 300 to 500 g is the purchase size; in such cases, country statistical offices must convert the price to a kg basis. Note also that in fact the ICP would treat bean curds in bulk and in packages as two different items for building up parities at the detailed heading level. These specifications have been built up through the interaction of country price experts and regional and world organizers of the ICP. By making these specifications available, countries have the option of integrating the collection of ICP item prices into their regular price collection.
102. Typically, participating countries are sent a draft list, built up from core specifications as well as items thought appropriate for their region or country group, which they are asked to review to determine:


1. Which items in their existing database they would be able to price exactly as proposed;
2. Which items in their existing database they would be able to price if some modification of the definition were made;
3. Which items they would not price because they were not important in the consumption in the country;
4. Which items might be added to the draft list because they are important both in the country and probably also in other participating countries;
5. Which items are not in the existing database but are important in the country and could be priced as additional items.


103. In drawing up their list, countries will want to minimize the resources devoted to pricing additional items under subparagraph (e) above, unless they are items that might usefully be added as a substitute or totally new item to their existing price collection for national purposes. In drawing up their lists, countries will therefore want to take as many items from their existing database as possible, as in subparagraphs (a) and (b) above, with one qualification. It is sometimes the case that the existing database is not available in one place, as when time-to-time price relatives are computed in different centres and the records or price observations are scattered and not in machine-readable form. In such cases, some additional pricing may require less resources than recovering existing national prices. This qualification excepted, countries should use their existing database as much as possible, including urging that items under subparagraph (d) above be adopted for ICP purposes.
104. Some trade-offs do exist between items under subparagraphs (b) and (e) above. Suppose , for example, an ICP specification calls for the price per kg of tomato sauce in tins of 200 to 250 g and the country specification is for tins of 500 g. The relationship is usually not proportional, so one cannot simply give the price of a tin of 250 g as half the price of a 500 g tin. However, if both size tins are commonly purchased in a country, it is sufficient to modify the national price for 500 g to a price for 250 g by the ratio of the prices of these two sizes in a few outlets. Put another way, it would not be necessary to take a price survey of tins of tomato sauce of 250 g and treat this as a new item. Other specifications may be less clear-cut and countries should raise questions about them when they submit their proposed price list to the regional coordinators.
105. The next step may be a regional workshop at which price experts review the list with a regional coordinator to resolve questions of commodity and service definitions and to agree on a final list. Sometimes a price expert will visit a country to draw up the final list. In either case, it is often necessary to visit stores, examine samples or review catalogues in order to settle on an operating version of the list of specifications for which the country will supply prices.
106. After a final sample of items is agreed upon, national statistical offices then have to collect the data and determine the national average prices for the items selected for submission to the organizers. It is neither necessary nor desirable for multilateral comparisons that countries price all items on the list; nevertheless, to obtain reliable PPPs at the GDP level, a relatively large number of average prices (several hundred, at least) have to be provided. While submission of the average prices can take place only after the reference period is over, arrangements for the necessary additional price observations are most conveniently begun much earlier, preferably before the reference period.
107. The third stage consists of a review of the submitted prices by the organizers of the country groups to check for possible errors and to make requests for supplementary data as required. Often errors arise because prices are provided for a different unit than called for in the specification; this quickly becomes apparent when price ratios of items are examined for a number of countries. At the next stage, a preliminary set of basic heading' parities is reviewed by the organizers, often in consultation with the participating countries in a region. This review of basic heading parities makes it easier to discover possible errors in the item prices ranging from simple clerical mistakes to misinterpretation of the item descriptions or units of measurement.

108. The item specifications entering into the calculations of parities at the basic heading level serve the same function as the price representatives in the building up of national temporal price indexes, such as the consumer price index, from category price changes. Of course, the dimensions of the comparisons are different in the two cases. It is space (country-to-country) in the case of the PPPS, and time (period-to-period) in the case of the national price indexes.
109. In their time-to-time indexes, most countries compute the ratio of prices in two time periods and take an arithmetic average of the ratios across outlets (APR); or less frequently, countries compute the arithmetic average of prices of comparable items across outlets in two time periods and then take their ratio (RAP) as the measure of price change. 5/ When ratios of average prices (RAP) are the practice, it is clearly the aim to compare the price of identical products in each period at the same outlet, as in the ICP. Whether the time-to-time price relatives are computed at each outlet, and the relatives are weighted across outlets (APR), or prices for identical specifications are averaged across outlets and a price relative calculated (RAP), usually implicit or explicit weights are used at the outlet level. I/ When the price change for an item has been calculated, these are then combined across items using some weights.
110. However, an item in time-to-time indexes is unlikely to have its own weight. Countries often have quantity weights at a level of detail, such as tinned vegetables, but they are unlikely to have total quantity within countries for a detailed specification, such as a 500 g can of peas. In time-to-time indexes, the weight assigned to the price change of tinned peas is likely to be the whole weight of tinned vegetables. The assumption that changes over time in the price of peas represent price movements of tinned vegetables is basic to price index construction.
111. The analogous assumption for place-to-place comparisons is that items chosen for a heading represent the price structure of the country for that heading. Thus far, the ICP has not developed a sampling frame to spell out exactly how this selection should be done. The aim is clearly to select items in all of the countries that appropriately represent the prices of goods consumed in each country in that basic heading. By using the term important to describe items, it is recognized that certain criteria for item selection exist, but that there is not yet a sampling design that allows guidance as to what are truly representative items. 1/
112. Where countries average ratios of price changes (APR), their task of matching specifications across outlets (space) is substantially reduced and, as will be discussed in chapter IV, the methods developed in these countries may provide some guidance for future ICP work. Because APR countries need not average prices across outlets, more supplementary price collection, or additional processing of existing surveys, will be necessary for these countries, because average prices of individual items are not an intermediate output of the time-to-time price estimation procedure.
113. Where countries follow the RAP approach, there will be a number of (national) average prices available from regular surveys that can be used directly or perhaps with some modification for size of purchase, or other feature, for the ICP. One reason that countries compute average prices and then take ratios is that the average prices are of interest in themselves. Countries participating in the ICP will as a part of their work generate a large sample of average prices that may prove useful for other purposes.
114. One major difference in the spatial comparisons is that the number of potential items in a time-to-time index remains relatively stable from year to year, which is not true across countries. Consider the fact that supermarkets in the United States carry from 10,000 to 20,000 items and brands, while in many countries the number would be under 1,000. This means that single specifications are likely to be a smaller percentage of the total expenditures in a detailed heading in more affluent countries and that sampling becomes a necessity. However, countries with relatively small numbers of items in their markets may still have a significant number of items that are not common to more affluent countries. In choosing what items to match across countries, the ICP has been guided by two basic principles of item selection, which are now discussed.

115. One principle of item selection that has been generally accepted, if not precisely defined, is that specifications priced by a given country should be sufficiently typical (characteristic) for the country. Pricing uncharacteristic items (i.e., goods or services which, though they exist in a given country, are not important in expenditure budgets and/or are not widely available in outlets for such items), is to be avoided. I/ Items not commonly consumed may have very high prices (and, just because they are so high, the given product is purchased in a very small quantity, thus, it is uncharacteristic) and can be judged as not important to price.
116. The principle of choosing important products often comes into conflict with the second principle of choosing identical products. This is perhaps the most important issue in price selection. Often a specification can be associated with a brand name so that identity of product can be assumed (in fact, even this need not be the case as firms frequently assign the same model number to items that are either technically different between countries or are produced under the same name in different locations with some modifications for different markets). However, if a brand name is found in a country but is not commonly consumed, then it may poorly represent that basic heading. This conflict between how identical and how important is a product will be less important in a homogeneous group of countries, such as the EC, where brand names can be part of the specification. However, when comparisons are made for core commodities throughout the world, it is often necessary to trade off identity to make sure that items are characteristic of the purchases of a country in a basic heading.
117. In the above discussion, identity of items was spoken of in terms of brand name, but there is more to comparing like with like than just trademarks. The principle of comparing identical items means that there should be no differences in either the quantity or the quality of the specifications selected among countries that significantly influence the use of the given good or service. In particular, identity means that:


1. The size of the good/service should be the same. As mentioned above, this requirement has a double meaning: the unit price reported not only relates to the same size (e.g., the price of 1 kg of potatoes), but the price originally observed should also relate to approximately the same purchased quantity in each country. It would not be appropriate, for instance, to observe the price of sugar in a 1 kg package in country A while observing it in a 10 kg package (and dividing it by 10) in country B, since the price of a larger size package covers relatively less distributive services (per kg) and is, typically, relatively lower;
2. The physical and functional properties should be the same (for instance, the thread count of fabrics, the capacity of machines, the life of electric bulbs). This relates to all such properties that may have significant influence on the price of the given product. Shape and colour may also be relevant for some products, although not for others;
3. Types of outlet should ideally be the same when matching items across countries (this is one of the more important conditions that are discussed further in the following section);
4. Delivery conditions (e.g., package, warranty, transportation cost included or excluded) should also be the same. This is particularly important for producer and consumer durables;
5. Other circumstantial factors (e.g., type of restaurants for consumption of specified dishes, access to repair services) should be the same in so far as they have a significant effect on the price. The set of characteristics that seem relevant are decided upon in the construction of item specifications. In principle, characteristics not listed in a specification are considered to have no influence on the price of an item.


118. How does a country try to balance these criteria? First, those items that a country has in its own database are, in ICP terms, important for the country. If these items match, or can be simply modified to meet ICP specifications, they should certainly be chosen as items for which the country will supply prices. When should a country substitute an identical product that is not a regular part of a country's national price collection for a comparable item for which prices are regularly collected? A necessary condition for substitution is that the identical item is well known and widely available and consumed in a country, where widely can mean all over a country, or in many different outlets even if concentrated in urban areas. If this condition is met, then a country should substitute the identical product when there is reasonable doubt about whether the items are in fact comparable.
119. A corollary of the above is that a country should not substitute a brand-named item that may in some cases be mentioned for illustrative purposes in some ICP specifications simply because it is known to be consumed in the country. Unless the item is widely available and widely consumed, it should not be substituted. What about the case where a brand-named item is consumed in a few places in the country but there is either no item priced in the country that meets the specification or there is considerable uncertainty about whether the specification in the country matches the ICP specification. In general, the brand-named item should not be priced. Secondly, if there are doubts about whether an item priced for national purposes matches the ICP specification, either the item should not be proposed for pricing or the question should be raised with the regional coordinators.

120. Identity and importance are principles with respect to specifications of individual items. In addition, it may take more items to represent one basic heading than another. In relatively homogeneous basic headings, a low number of specifications - even one specification - may sufficiently capture the price structure of the basic heading. For instance, in the basic heading Eggs (1.1.1.04.5), which in many regions consists overwhelmingly of chicken eggs, one single specification may sufficiently represent the heading, though weight or size will still need to be specified.
121. In a heterogeneous basic heading, such as Glassware, tableware and household utensils (1.4.4.01.0), consisting of thousands of different items and brands, a large number of specifications may be needed. The ICP has offered over 40 generic specifications for this basic heading, because it is so diverse. To be more precise, the homogeneity or heterogeneity of basic headings should be judged not so much in terms of number of items but in terms of their relation to the individual price ratios. If the dispersion of price ratios is small, only a few specifications may be necessary even when it appears that the basic heading is quite heterogeneous in terms of number of distinguishable items. Where price ratios have substantial dispersion, a larger number of specifications is needed.
122. In most country groups, the number of items suggested per specification differs from basic heading to basic heading. These differences have developed as experience has been gained about those headings where there is more dispersion in the price ratios. 9/

123. The ICP seeks national annual average prices for items because these are in principle the prices that should be embedded in the national accounts. National expenditures on items at the basic heading level are represented by the sum of the values of all the different transactions occurring throughout the year on the items over the whole country in a basic heading. The annual national average price when divided into the expenditure on a given item should be the unit value that gives the quantity of the item purchased in the country during a year. The annual average question will be treated first, and then the more interesting question of how to obtain national average prices will be discussed.

124. In many countries, it is normal to compute annual average prices for many food items, especially those with seasonal price variations, but this is not common for other items. In practice, where it is not a heavy computational burden, countries have averaged monthly observations of prices regularly collected for other price purposes. Same countries, that have monthly data, have reduced computations by using an average of, say, April and September prices to approximate an annual average for non-seasonal items. Countries experiencing high rates of inflation, or erratic patterns of price change during a year, may find other methods more appropriate.
125. An alternative that has been used in many participating countries is to choose a month for ICP prices, say October, and then move these prices to annual average prices. For example, if the ratio of October prices in the consumer price index to calendar year annual prices in 1993 for footwear is 1.02, then each item price for footwear referring to October would be divided by 1.02 to obtain the estimated annual price.
126. The above discussion takes the annual average to be for a calendar year, which is the norm for the ICP. As discussed in chapter II, a number of countries do not estimate their national accounts on a calendar year basis. Usually, these countries provide the ICP with adjusted calendar year expenditures. It is clearly important that both the national accounts and price observations refer to a calendar year.

127. There are three sets of prices used in PPP calculations; (a) national prices that are part of those regularly collected by the statistical agencies, (b) prices specially collected for the ICP, and (c) normalized prices. Because normalized prices are generated in a way that makes them national average prices, most of the discussion of this section deals with the first two sets of prices.
128. Normalized prices may be estimated by hedonic regression methods, as in the case of house rents, motor vehicles and other large, complex and expensive pieces of equipment; or normalized prices may be calculated on the basis of pre-defined cost estimates, as in the case of construction projects and buildings. A normalized price is usually constructed as a national average price. Some specially collected prices, such as postal rates, are, by definition, national prices, while many machinery prices are national because there are only a few distributors.
129. However, for a large number of consumer items, prices vary greatly by outlet and region and between rural and urban areas. In contrast to the measurement of time-to-time price changes, where the necessity for absolutely identical items to be priced nationwide is not pressing, the ICP makes more rigorous demands on comparability of specifications across observations within a country. In consumer price indexes, for example, "like with like'' comparisons over time for a well specified item are based on a process of averaging price relatives, and there is no need to ensure that the products priced in different shops or outlets, and in various parts of the country, are identical. Indeed, in order to better represent time-to-time price changes, the items included in an index in different regions should be more characteristic of local expenditure patterns, including point of purchase. But for ICP purposes, the items - and the qualities they reflect - should be as similar as possible for a proper comparison to be made.

130. For many purposes, the ICP has adopted the potato-is-a-potato principle that says if an identical product is sold in a variety of markets, the ICP will treat it as the same. This means that the national average price of potatoes will include prices from village markets as well as specialty grocers. The reasoning here is that the utility derived from consuming a potato is the same whether it is purchased from a roadside stall or a supermarket. To the extent that less service is involved in the sale of potatoes in less affluent countries, this principle leads to lower prices, other things being equal, for potatoes in poorer countries. This principle seems to make sense for many commodities, but it certainly is not acceptable for services, so that for say a cheese sandwich to be eaten away from home, the outlet has always been an essential part of the ICP specification. Again, the reasoning on the cheese sandwich is that the consumer's utility depends upon the whole eating out experience, so that whether the sandwich is purchased in a roadside stall, a diner or a full service restaurant clearly needs to be part of the specification.
131. At present, the state of the ICP country practice in calculating national average prices varies substantially. The example below illustrates a number of issues in the construction of national average prices and discusses some approximations that have been made by individual participating countries in obtaining estimates of their average item prices. The example also makes clear why the problem is part of a more general concern for how the core and other specifications of the ICP continue to evolve in each region and across regions.
132. In the example, suppose prices for "'Irish' potatoes, white or red skinned", are collected from each of a broad cross-section of different outlets in countries A and B. In addition to the average price for each type of outlet, the quantities, which are not normally known, are provided.

Outlet	Price in A	O in A	Price in B	O in B	Pb/Pa
	(1)	(2)	(3)	(4)	(5)
Urban speciality	1.00	10	60	5	60
Supermarket	0.80	60	50	10	62.5
Urban market	0.60	10	45	20	75
Hawker/vendor	0.50	10	45	10	90
Rural market	0.40	10	40	55	100
Quantity weight national average	0.73		43.5		59.6
Geometric mean unweighted	0.6258		47.547		75.974

133. In the example, the prices differ across outlets in good part because urban distributors must bear costs, such as transport and taxes, that are an intermediate product. Some of the price difference may be a service element that urban consumers purchase when they go to a supermarket or specialty shop. The average price in each country is computed in two ways, first as a quantity weighted average and second as an unweighted geometric mean. The weighted average price is 0.73 in A and 43.5 in B and the ratio of these averages is 59.6. In this particular example, the ratios of the prices at any of the particular outlets as given in column (5) are all greater than the ratio of the average prices. This may be interpreted to mean that in country A the potato is typically bought in supermarkets and in country B typically in rural markets so that a ratio of the average prices is close to 50 (= 40/0.80, the ratio of rural markets in B to supermarkets in A).
134. Total expenditures in national currencies in A are 73 and in B 4,350.If A is an affluent country and B a poor country, then this example illustrates why the potato-is-a-potato principle, because of the larger amount of intermediate product and customer amenities entering into commodity sales, tends to make some price comparisons lower in low-income countries. 10/ The comparison based on national average prices does correctly reflect the unit cost of resources involved in each country in getting a kg of potatoes to consumers. Another advantage of the national average price calculation is that it produces the quantity of potatoes consumed in each country when expenditures are divided by average prices. That is, the quantity in A is equal to that in B, namely, 100. If we divide the expenditure ratio of 4,350/73 = 59.6 by the ratio of average prices of 59.6, we of course get the same result, that each country consumes the same quantity of potatoes.
135. This last point can be illustrated in another way. Consider an arithmetic weighted average of the price ratios. Using country A's quantities, the average of the price in B to that in A is 66.25, and using country B's quantities, the average is 88.25. The Fisher or geometric mean is 76.5. If we now make a quantity comparison between A and B, B will be only 0.78 of the consumption of A. One can interpret this result as saying that, given the outlet, the resources devoted to potato production and distribution in B is only 78 per cent of that in A. Clearly, averaging ratios of outlet prices and comparing the ratios of prices averaged over outlets both have advantages and disadvantages. However, one point is clear, the closer two countries are with respect to the distribution of sales by type of outlet, the less difference it makes which approach is followed.
136. Another point mentioned earlier that is illustrated in the example is that, though it is not a theoretical advantage, there is a convenient feature involved in using the geometric mean if weights are not known. The last row in the example gives the geometric mean of the prices in each country. The practical advantage of the geometric mean is that the last entry, 75.974, is the geometric mean of the price ratios in each outlet, as well as the ratio of the geometric mean of the prices across outlets in countries A and B.

137. There frequently are large rural-urban and regional differences in food, services and some other prices. It is therefore particularly important that these differences be captured in the national average price. Often, this may be done on a sample basis, so that a rural-urban price factor can be developed by type of item. For example, usually the rural price for cereals is only slightly lower than the urban price, while for fruits, vegetables and meats, the difference may be 50 per cent or more. The EC countries have developed such factors that allow them to go from capital city to national prices without making special collections of ICP items throughout their countries. Many ICP countries have found that a small pilot survey was adequate to find out the appropriate factors to adjust urban prices to national prices for foods and other expenditure groups.
138. Reality differs from the example above in at least one important respect, namely, that quantities are not generally known. This means that there is uncertainty about how one calculates a national average price or how one averages price ratios. What is the implication of this for country procedure? There is some information on this question for several industrialized countries.
139. Recent price inquiries undertaken in major cities for the purposes of international salary determination clearly show that in some cases the variation in prices for the same product within a given city are even greater (when converted at official exchange rates) than the difference in the average prices for the same item between several of the cities involved in the international comparison. This tendency was also observed in the OECD special direct "three-city" pairwise comparisons carried out for both the United States of America and Canada in their 1980 phase IV survey. 11/ This implies that outlet-specific as well as product-specific pricing may need to be followed to obtain the most robust form of item price matching. Consider the following example, which differs from the potato example in that the quantities are not given, but two regions of each country are represented.
140. For purposes of this illustration, suppose the item is a pair of training sneakers of a major world manufacturer, such as Adidas or Nike, and the price data are as follows:

	Country A			Country B
Outlet	Price in region		P2/P1	Price in region		P2/P1	Pb/Pa
	(a)	(b)		(a)	(b)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Urban specialty	60	55	0.92	350	400	1.14	6.52
Department store	40	40	1.00	250	300	1.20	6.88
Urban market	35	35	1.00	220	250	1.14	6.71
Hawker/vendor	30	25	0.83	200	190	0.95	7.09
Rural market	35	30	0.86	230	220	0.96	6.92
Simple arithmetic average	40	37		250	272
Geometric average	38.82	35.67		245.15	262.82		6.821

In this example, the variation across outlets in each country within each region is substantial. Columns (3) and (6) give the ratio of the price in each region for a given outlet. Prices vary across outlets in each country and region by at least 75 per cent, which is much more than the variation across region in either country.

141. Another way of illustrating this point is provided in column (7), where the ratio of average price by outlet in each region is given between countries B and A. For example, the simple average price in the rural market in A is 32.5 and in B 225, and the ratio is 6.92 (= 225/32.5). The variation in column (7) is quite small, intentionally reflecting the type of empirical finding commented upon in paragraph 120 above. If the ratio of the geometric mean of the prices in each country for each type of outlet were taken, the corresponding entries would also display little variation across outlets. The last row of the example again illustrates an advantage of the geometric mean; the last entry, 6.8il, is the geometric average of the ratios of the geometric mean of the prices for each type of outlet as well as the ratio of the geometric average of the geometric mean of the prices across outlets in each region.
142. As a rough generalization, differences in prices between regions and between rural and urban prices within a region and a country can be treated as differences owing to intermediate product, mainly transportation and handling. This is graphically illustrated in the case of the United States, which in the phase I comparisons had the highest prices of any country for tomatoes and the lowest price for canned tomato sauce. Tomatoes were cheap where they were grown and canned, but very expensive to deliver to urban outlets. The potato-is-a-potato principle can be safely applied for rural-urban and region-to-region price differences.
143. However, outlet price variation within urban areas may be due to a significant element of difference in the final service to the consumer in purchasing an item across outlets. Consider the size of error that might occur if countries A and B priced the item in only one of the first three outlets. This might occur, for example, if there was no information on quantities sold by type of outlet. Three of the nine possibilities are given in the first three rows of column (7) and the other six have been calculated using the average price of each outlet type across the region. The combinations are:

Ratio of price in country B to price in country A

	Country B's price from
	Urban specialty	Department store	Urban market
	(1)	(2)	(3)
Country A's Price from
Urban specialty	6.52	4.78	4.09
Department store	9.38	6.88	5.88
Urban market	10.71	7.86	6.71

One point of this example is that, if there is not much information about quantities purchased by outlets, then the result will be subject to much less error if outlet is held constant. That is, the diagonal elements have much less variation than the off-diagonal elements.

144. This example also makes clear that, if weights are not known, it may be even more important to devote resources to holding outlet type constant than to determining regional differences in prices. This can be done, for example, by a small survey that generates estimates of the difference in price by outlet for several varieties of consumer purchases. 12/
145. The more likely situation is that countries do have some information about the importance of outlets and can say one outlet type is more likely to be the volume seller of a particular item. In this case, countries should be guided by the three considerations developed in this section: (a) ensuring that price quotations from important outlets receive more weight in any national average price; (b) treating rural-urban and regional differences in prices according to the potato-is-a-potato principle, averaging their prices by quantities sold; and (c) whenever outlet types within urban areas can be made part of a specification, this should be pursued in collaboration with other participating countries.