The Canadian Consumer Price Index Reference Paper
Chapter 5 – Sample Strategy and Price Collection

5.1 The number and variety of transactions that consumers engage in is immense. It would be neither practical nor affordable to collect prices for all transactions of products sold in all outlets^Note  to compile the Consumer Price Index (CPI). Therefore a sample strategy is necessary.

5.2 The CPI has always had a policy of adopting the most appropriate measurement methodologies for each of its elementary indices. This has led to sometimes very divergent sampling practices in different parts of the CPI. This chapter will cover the range of sampling practices currently used in the CPI, first focusing on the general sampling approach which covers more than 50% of the CPI by basket weight.^Note  The chapter will then discuss some of the more specific sampling approaches in other parts of the basket, including full-universe price coverage, cut-off sampling and price modelling.

5.3 Not all elementary price indices are estimated with field observed prices by interviewers.

• 5.3.1  For some elementary aggregates, particularly the “catch-all” classes with heterogeneous product mixes that typically represent more marginal consumer expenditures, price collection is not practical or necessary. In these cases, imputations are made whereby the price movement of the elementary aggregate is estimated via proxy, using the price movement of a donor class.
• 5.3.2  For some outlets, Statistics Canada receives transaction information as recorded at the cash registers of retailers. In that case, scanner data are used to calculate the transaction price of specific items. As the coverage of scanner data increase in terms of outlets and products, Statistics Canada switches away from traditional field collection to the use of scanner data. For some indices (passenger vehicles insurance premiums and homeowners’ insurance, mortgage interest cost), Statistics Canada uses administrative datasets to estimate the price paid by consumers. In other instances (air transportation, rental of passenger vehicles, traveler accommodation and package holidays), Statistics Canada uses Internet based collection to record the prices posted by the retailers.

General Sampling Approach

5.4 The general sampling approach for the CPI can be seen as a three-stage survey design. The first stage covers geography, the second is for outlets and the third stage is for the products. Even though the product component appears at the last stage, types of products have a major influence during the second stage when outlets are selected. Figure 5.1 depicts this general approach for selecting sampling units, starting from geography, then using the product type information for outlet selection and then assigning Representative Products (RPs) to be observed within each outlet. More details on each of these stages are provided in the following paragraphs.

5.5 The geographical sampling unit is primarily the Census Sub-Division (CSD) as defined by Statistics Canada’s Standard Geographical Classification (SGC). CSDs are similar to municipalities and are chosen for the CPI sample based on information such as population counts and economic activity.^Note  The CSDs are selected as the sampling unit mainly because they are stable over time and because every location in the Business Register (BR),^Note  the frame for outlet selection, is mapped to CSDs.

5.6 To facilitate outlet sampling and price collection management, CPI collection areas have been defined. For small to medium size CSDs, collection areas correspond to the CSD. In cases where the CSD is too large to represent one collection area it is broken down into smaller areas which are amalgamations of Census Tracts (CTs), which can be seen as equivalent to neighbourhoods. The number of collection areas within a CSD depends on the number and variety of retail and services locations as well as the size, in terms of square kilometres, of the CSD.^Note 

5.7 The CPI survey frame is used to select the outlets where price collection is to take place. The frame was built using existing Statistics Canada sources, mainly the BR, the quarterly Retail Commodity Survey (RCS) ^Note  and services industry surveys such as the monthly Food Services and Drinking Places Survey. The RCS and services industry surveys are used to link the CPI product classification to the industry classification in the BR, which is the North American Industrial Classification System (NAICS).

5.8 A major feature of the CPI survey frame is that it gives the revenue, according to the BR, of each outlet by RCS commodity class or service industry using the data reported in various Statistics Canada surveys. This characteristic of the frame helps in dealing with big retailers such as the department stores, which typically sell a range of products. It facilitates the selection of outlets which are among the most popular in each product class or service industry.

5.9 The outlet sampling process for the CPI is done in two phases. The first phase, pre-contact sampling, is designed to validate the information from the BR, such as activity status, industrial classification and contact information. Depending on the industrial classification, additional questions may be asked to determine whether specific products are sold. The second phase consists of selecting a subset of outlets from the pre-contact output (after all out-of-scope and refusal units are removed). Remaining in-scope outlets that were not selected for the sample are kept as a replacement list.

• 5.9.1  If the outlet selected is not one of the outlets for which Statistics Canada receives scanner data, the information is sent for field collection by specialized Statistics Canada employees, known as price interviewers.
• 5.9.2  If Statistics Canada has scanner data (transaction data) from the outlet selected, no field collection is needed. Instead, an employee from Statistics Canada head office will use the transaction data to calculate average unit prices.

5.10 After outlets are selected, a set of RPs is assigned to be observed in each outlet. Subject matter experts use external databases and market research to help define RPs in a way that strikes a balance between specificity and flexibility. Price interviewers or employees working with scanner data then select specific products that meet the RP definitions. This approach ensures a clear understanding about what kind of product to select (keeping intact the matched-model approach of the CPI) while at the same time providing leeway to choose products that are locally popular (upholding the representativeness of the estimated elementary indices).

5.11 The RP list may be different from one outlet to another according to the information obtained from the CPI survey frame and the pre-contact phase. For example, one pharmacy may be designated to have health care goods as well as some food items (such as milk and bread) collected if those products correspond to its main streams of revenue. However, this may not be the case for all pharmacies in the CPI sample.

5.12 The number of RPs assigned to each elementary aggregate depends on the weight and the complexity of measuring price change for the given product class. In a complex elementary aggregate, one which may have a lot of heterogeneous products included, several RPs will likely be assigned. In a simpler elementary aggregate, just one or two RPs might be enough to measure price change adequately.

5.13 RP assigned to an outlet is called a Target Product Offer (TPO). The TPO acts as the sample intention. The sampling allocation scheme allots a number of TPOs to each elementary aggregate, taking into account the basket weight, price variability and cost of collection for each elementary aggregate. The objective is to allocate the available sample optimally in order to estimate elementary indices of the best possible quality.^Note 

5.14 Every month the price interviewers observe Product Offers (POs) for TPOs. The monthly process begins with a detailed sample request from headquarters to the interviewers. Most of this request is the same as in the previous month, since the CPI follows the matched-model approach. The sample request is loaded into Computer-Assisted Personal Interview (CAPI) devices. Carrying these devices, the interviewers visit each of the outlets in their particular workload. In each outlet, the interviewer finds the required POs and enters their prices and characteristics into the devices.

5.15 When a PO is advertised as being “on special” interviewers record this in the CAPI device. If the item cannot be found because the shelf is empty, the interviewer consults with a store manager and determines whether it is temporarily out of stock or is no longer carried by the outlet. Either way, this information is recorded in the device and transmitted back to Statistics Canada headquarters.

5.16 If the PO being sought is determined to be no longer carried, a substitute PO is selected and this information is recorded in the CAPI device. For especially complex POs such as high-tech goods and items of clothing, the interviewers also fill out forms providing additional details about the characteristics of the substitute PO. This additional information assists the analysts at headquarters who assess the extent of quality change and estimate an appropriate adjustment.^Note 

5.17 When an interviewer must select a substitute PO, as just described, he or she is guided by the RP description, a set of detailed product specifications loaded in the CAPI devices. The interviewer is asked to select a substitute PO that fits the description.

5.18 For many RPs it is not necessary to observe POs every month, either because their prices tend to change less frequently or because they are only available at specific times of the year. For one example, tuition fees typically change only once a year, in a predictable month, so they are collected only in that month. For another, Internet access service fees are collected every second month because their prices change infrequently. In fact, every RP description specifies the month(s) of collection.

5.19 If a price change is known to take place outside an RP’s default pricing schedule Statistics Canada will conduct a special pricing.^Note  This ensures that elementary indices represent, as much as possible, price change in a timely manner.

5.20 As for the timing of collection within the month, most POs are observed in the first two weeks, with the first week being defined as the one containing at least three business days from the calendar month. Gasoline prices are an exception due to their typical intra-month volatility and thus are collected in four weeks.

5.21 When the POs have been collected they are sent to Statistics Canada headquarters by encrypted digital transmission, where they are compared against the original sample request to determine the sample’s completeness and conformity to requirements. Thereafter, the POs undergo further review and processing. Any unusual price movements are carefully checked to ascertain their validity. Corrections are made if necessary. Where POs have been substituted, the degree of quality difference is assessed and an appropriate quality adjustment is made if necessary. Steps are also taken to ensure the POs include applicable taxes. Finally, with all checks and adjustments completed, the elementary price indices are calculated.

Specific Sampling Approaches

5.22 Exceptions are made to the general sampling approach. These exceptions are intended to capture price change for elementary aggregates where information on the universe of consumer transactions is available or where the attributes of the products within the elementary aggregate are complex and require different techniques.

5.23 Elementary aggregates where full information on the universe of consumer transactions is available include those goods or services that have only one market and/or seller. Examples include passport fees and drivers’ licences. For these there is no sample drawn because all prices are collected and used in the CPI.

5.24 For some elementary aggregates in which there is one seller in a particular geographical stratum,^Note  a sample of outlets and products is not required. However, in these cases the first stage of the general sampling approach, in which collection areas are selected, is still necessary. Usually, the collection of theses POs is done by employees at Statistics Canada headquarters rather than by interviewers in the field.

5.25 The prices used to calculate the rent price index come from the Labour Force Survey, which uses a probabilistic sample.^Note 

5.26 Some elementary aggregates in the CPI follow cut-off sampling.^Note  When there is information available on the outlet and/or product universe the goal is to maximize the coverage of both. Most times information on market composition comes from third party administrative databases, often available via the Internet. These databases are used to rank outlets and/or products so the sample covers a majority of the market. Once TPOs are assigned via cut-off sampling, POs are collected at Statistics Canada headquarters.

5.27 There are some elementary aggregates where the target population purchases bundles of services rather than individual products^Note  or where prices are based on a set of specific conditions.^Note  The CPI uses the profiles method to capture price change for these products.

5.28 Representative bundles or consumer profiles are selected using available market information. The intention is to cover a majority of the services and outlets that are available in a particular market. In these cases POs for the defined consumer profiles are observed at Statistics Canada headquarters from large industry databases.

5.29 The profiles method is another application of the matched-model framework. By observing POs of identical consumer profiles every month, this method ensures that the quantity and quality of bundled services are constant over time and that the CPI reflects pure price change. As with the entire CPI sample, it is important that consumer profiles are reviewed and updated regularly to ensure that representative consumption bundles are being priced. When profiles must be changed because the component products are determined to be new or of different quality, then quality adjustment techniques are applied.

5.30 Some elementary aggregates are characterized by products whose prices are determined not by their physical characteristics, but by their intellectual content and novelty. For these elementary aggregates the CPI uses the bestsellers method to estimate price change.^Note  In these cases POs are observed in the field by price interviewers as well as on the Internet.

5.31 There are three elementary aggregates which use modelling to estimate price change: the Mortgage interest cost and the Homeowners’ replacement cost indices (two components of the owner-occupied accommodation price index), and the rent index (the largest component of the rented accommodation price index). Unlike the treatment of other durable goods in the CPI, owner-occupied housing follows a user-cost approach, which aims to measure the implicit price of the flow of services coming from a fixed stock of owned dwellings. The two components of this price index require special measurement methods.^Note  The rent index estimation is based on data collected from the Labour Force Survey and a hedonic regression model.^Note