|
Printable version
Key Statistics - article, July 2004, p. 9-12
Consumers Price Index supermarket brand share analysis1
Introduction
For the Consumers Price Index (CPI), the prices of a wide range of food and non-food groceries are collected each month from supermarkets. To ensure that the CPI accurately measures changes in grocery prices, the mix of brands in the CPI sample should reflect current consumer behaviour.
The supermarket brand share analysis project involved comparing brand shares in the CPI sample with those derived from scanner data (sourced from AC Nielsen). This article reports on the approach taken.
Outline
Statistics New Zealand publishes several price indexes. The most well known is probably the Consumers Price Index (CPI), which provides a measure of the change in prices of goods and services bought by households.
Using a more precise definition, the CPI measures the changing cost over time of the goods and services purchased in New Zealand in a specified base period by private, New Zealand-resident households.
A CPI is usually calculated as a weighted average of the price changes across goods and services priced for the index. The weights are meant to reflect the relative importance of the goods and services, as measured by their shares in the total consumption expenditure of households. The weight attached to each good or service determines the impact that its price change will have on the overall index.
The goods and services covered by the CPI are classified into nine groups, 22 subgroups and 70 sections. This creates a building block process to move from the individual price of a representative commodity at a particular outlet through to the aggregate index.
As consumer spending habits change, what we use to measure the CPI must also change. As the purchasing habits of the New Zealand consumer change across time, the CPI must continue to measure a representative basket of goods in order to measure the price change experienced by consumers. Therefore, Statistics New Zealand periodically updates the basket of goods that is measured, as well as the way it is measured.
This is where the supermarket brand share analysis project fits in. The aim is to check, and if necessary change, the mix of brands in the basket of goods measured at supermarkets to gain a fair representation of the products actually sold at these supermarkets. Basically, the intent of this project was to address the need for an explicit weighting system for brand share across goods from supermarkets. Currently, there is not a system in place to directly control the weights assigned to manufacturers and brands.
Current Method of Weighting
The CPI uses prices for 672 distinct commodities across 15 regions to measure price change for goods and services covered by the index. The index is broken down into hierarchical indexes from the prices collected in the field to the all groups measure. The all groups measure is split up into nine groups covering groups of CPI commodities such as food or apparel.
These nine groups are then divided into 22 subgroups to further define the group (eg the food group contains subgroups: fruit and vegetables; meat, fish and poultry; and so on). These 22 subgroups are further split into 70 sections.
The weights for the groups, subgroups and sections are their shares in the total consumption expenditures of the reference population. These weights are derived mainly from the Household Economic Survey (HES). As the HES is a sample survey that is subject to reporting and non-response errors as well as sampling errors, the estimated shares for certain areas of the CPI are often modified or revised on the basis of supplementary or additional information from other sources.
Regional weights may typically be obtained from the HES or they may be estimated from retail sales or population data. Currently, Statistics New Zealand weights regions from the population data sourced from the census. However, it should be noted that this approach is under review by the CPI revision advisory committee.
Prices are collected from a variety of outlets and outlet types. Information about the sale or market share of the outlets may be used to form elementary aggregate weights that are specific to a given region and outlet type. Statistics New Zealand uses data from the HES to derive weights for outlet types for given commodities. Within a region, the outlet weights are in some cases derived from sales data from the retail trade survey. In the case of many food items within the New Zealand CPI, only supermarket chains are assigned weights based on retail trade sales, with other outlet types, such as dairies, equally weighted.
Example of Current Approach
For the CPI, Statistics New Zealand prices more than 200 distinct commodities at supermarkets. One of these is bread. Prices for different types of bread are collected across supermarket outlets throughout the country. Within each of the 15 regions of the CPI, several prices are recorded, which are weighted by the type of outlet they are priced at. Data on expenditure from the retail trade survey is used to weight the prices collected from supermarkets. From there, a weighted average is calculated to derive a price for bread types across supermarkets and convenience stores within that region.
National expenditure data from the HES, allocated in proportion to the population across a region, is used to create a weighted price aggregate for bread types for the 15 regions. These are then combined to produce a total bread price aggregate across New Zealand.
From there, the bread price aggregate for each type of bread is combined with other contributing price aggregates for the cereals and cereal products section, all weighted from HES expenditure weights at a national level. Across the CPI there are 70 sections, which are a combination of price aggregates.
The 70 sections are then combined to create 22 subgroups. Again, a weighted average is taken of the sections within a subgroup. The weights used for the calculation of the subgroups price aggregate are based on data on expenditure from the HES. In the case of bread’s section, cereal and cereal products, the subgroup is grocery food, soft drinks and confectionery.
The 22 subgroups are then combined as a weighted average to create nine groups: food, housing, household operation, apparel, transportation, tobacco and alcohol, personal and health care, recreation and education, and credit services. Again, the subgroup weights are derived from expenditure weights from the HES.
The final step in the calculation of the CPI involves combining the nine groups to create the CPI all groups measure using HES expenditure data.
Supermarket Brand Share Analysis
When items are surveyed at supermarkets, the choice of brand and manufacturer is usually at the discretion of the pricing officer, based on availability, shelf space and advice from supermarket staff. This doesn’t necessarily give a good representation across the CPI of the actual choices made by the New Zealand consumer. Take the hypothetical case of a product where brand A has 60 percent of sales, and brands B and C have 20 percent each. A pricing officer would normally select the most popular brand, given advice from supermarket staff as well as shelf share and so on. The problem that may result is that all prices collected may be only those of the most popular brand, brand A. This excludes brands B and C from the sample, despite the fact that between them they account for 40 percent of the market.
This method causes a few problems. By not directly controlling brand shares of items priced at supermarkets, manufacturer and brand shares may be out of line with their real shares of the New Zealand supermarket environment.
This is where scanner data helps. By comparing scanner sales data with the current implicit weights of items priced at supermarkets, we can check how accurately the brand or manufacturer shares within our basket of surveyed goods reflect the scanner sales data.
AC Nielsen supplies scanner data sourced from all transactions across the eight main supermarket chains. This covers more than 95 percent of the supermarket population. The data supplied gives a breakdown of more than 330 categories of items sold at supermarkets by manufacturer, brand and product sales. Sales detail is given in measures of both volume and expenditure; we are mainly interested in expenditure.
The main advantage AC Nielsen scanner data offers in the breakdown of sales data is the market share of each main manufacturer and/or brand. Across a given type of item, the relative shares of main manufacturers and brands for the market is given as a percentage share of expenditure.
The first part of the project was to select the CPI items to be looked at. The criterion for selection was based on items that we price by brand and/or manufacturer. This excluded items that are not sold as a branded product, such as delicatessen products or fresh fruit and vegetables. Products where the specification for pricing is based on the cheapest available, rather than brand, were also excluded. This left around 100 CPI items that could be compared to AC Nielsen scanner data.
From there, the implicit weights of brands and manufacturers that are currently surveyed by Statistics New Zealand were compiled. This was achieved by using the weights products implicitly have, based on the weights of the outlets and regions from which they are surveyed. For each product, weights within individual regions and across New Zealand were compiled and then compared with those of the AC Nielsen data.
Comparison tables of the current weights and AC Nielsen scanner data were created to evaluate any differences in the brand and manufacturer shares. Most of the differences in the current CPI sample are simply manufacturer or brand shares that are not in the same proportions as the AC Nielsen data. However, the AC Nielsen data did highlight some brands that are not priced when they should be, and brands that are priced when they do not hold a significant market share.
From the comparison tables, differences were categorised into two groups: items whose sample brand shares are broadly in line with AC Nielsen, and those with sample brand shares that are not broadly in line. Within the group of items that are not broadly in line, a ranked list was derived in order of the item’s weight within the CPI. This is to ensure items that have more effect on the overall CPI are addressed first, while lower-weighted items are treated in a less urgent fashion. For around one third of items, the brand shares of the CPI samples were broadly in line with the AC Nielsen scanner data. For the remaining two-thirds, the CPI brand shares were not broadly in line with the AC Nielsen scanner data. It should be noted that many of the differences in brand share would require only minor changes to align them with the scanner data.
From this ranked list of items, a strategy for improving the sample brand shares of those items can be derived. For most items this means a realignment of the sample to more closely match the brand shares in the AC Nielsen data. Random selection of supermarkets in which items should be priced was undertaken to match the brand share from the AC Nielsen data. In most cases, it was possible to get the brand shares of items within a percentage point of the AC Nielsen brand shares. Aligning national shares was the priority, but checks were also done to ensure that brand shares across regions were also a fair representation of the scanner data. The difficulty here is that within a region, the degree to which the brand shares can be matched is limited by the number and weights of outlets surveyed in that region. It should be noted that the checks across a region are mainly to prevent regions having only one brand selected in the sample.
From this suggested reselection of brands, a report was prepared on an approach for implementing the changes. Because many changes have been suggested, implementing all changes in one quarter might be impractical, as it would place a large burden on the pricing officers who would have to implement the changes in the field. Therefore, a gradual deployment of changes may be done in order of the importance of the item to the CPI. This would mean the sample of higher-weighted items would be realigned first, before lower-weighted items are realigned.
When a brand is to be changed at an outlet, an overlap period needs to be recorded to remove the effect of any quality changes between the old and new product being priced. This would mean that, for a single period, both the new and old products would be priced by the pricing officer in order to derive a link factor between the two, to avoid a jump or drop in price for reasons other than pure price change. As mentioned, gradual implementation of the reselection would reduce the extra workload for the pricing officer. Gradual implementation would also spread the workload in the editing and analysis areas of the CPI. Due to the need to produce the CPI on a timely basis, reducing the increase in prices to be analysed would not involve extra pressure on these areas.
Although this project is something of a first for Statistics New Zealand, the need for ongoing brand share analysis is necessary. An ideal approach would be to review items periodically to keep brand weights up to date with current market brand shares. Other than just realigning during a reweighting, which is done every three years, regularly checking brand share alignment could reduce the number of brand replacements needed at outlets. If more regular brand share alignment checks are done, it is less likely that brand shares of items within the CPI will drift away from current market brand shares.
Using AC Nielsen scanner data, it is now possible to obtain up-to-date brand shares across product lines. Now that we can check these brand shares, it is possible to improve the accuracy of our sample of goods and services priced at supermarkets. One of the main advantages of AC Nielsen scanner data is that it is possible to source data on a timely basis. This means brand shares can be checked against recent scanner data of transactions, avoiding the possible loss of accuracy and relevance from old transaction data.
Conclusion
Because the purchasing habits of the New Zealand consumer change over time, what Statistics New Zealand uses to measure the price changes experienced by consumers must also change. Many other Statistics New Zealand surveys are used to calculate accurate weights at various levels of the CPI. However, Statistics New Zealand does not exclusively source its weighting information internally. With scanner data, Statistics New Zealand can gain timely sales data which covers the majority of transactions processed by supermarkets. This data would be difficult to obtain through a standard Statistics New Zealand survey, and provides ample scope for improving the relevance of the CPI basket of representative goods and services that is used to measure price changes affecting households.
Footnote
SupmktBSA.pdf (459Kb)
The downloadable file is in Adobe Acrobat format. If you do not have the Adobe Acrobat Reader you may download the reader to view or print the contents of this file.
|