Annual Retail Trade Survey

Detailed information for 2014





Record number:


The purpose of this survey is to collect the financial and operating/production data needed to develop national and regional economic policies and programs.

Data release - July 18, 2016


The Annual Retail Trade Survey measures, on an annual basis, the operating and financial characteristics of Canadian retailers.

Data from this survey provide information on revenue, expenses and inventory. The data are used by all levels of government, government agencies, the retail industry and individuals in assessing trends within the industry, measuring performance, benchmarking and to study the evolving structure of the retail industry. The information is also a critical input into the measure of gross margins in the Canadian System of National Accounts (CSNA).

The Annual Retail Trade Survey estimates does not include companies that are classified as direct sellers, e.g., direct door-to-door selling; sales made through automatic vending machines; sales of newspapers or magazines sold directly by printers or publishers; and sales made by book and record clubs. These retail activities are collected and estimated as part of the Annual Non-store Retail Survey.

Except for the following conditions, internet retailing activities are included in the Annual Retail Trade Survey:

- where the store retailer financially accounts and reports for their store versus internet sales as two separate activities, store sales become part of the Annual Retail Trade Survey estimates and internet sales become part of the Annual Non-store Retail Survey estimates

- internet retail sales, when performed as an exclusive activity, are always included as part of the Annual Non-store Retail Survey

- internet purchases from a non-Canadian retail site are always excluded from the estimates.

Statistical activity

The survey is administered as part of the Integrated Business Statistics Program (IBSP). The IBSP program has been designed to integrate approximately 200 separate business surveys into a single master survey program. The IBSP aims at collecting industry and product detail at the provincial level while minimizing overlap between different survey questionnaires. The redesigned business survey questionnaires have a consistent look, structure and content. The integrated approach makes reporting easier for firms operating in different industries because they can provide similar information for each branch operation. This way they avoid having to respond to questionnaires that differ for each industry in terms of format, wording and even concepts. The combined results produce more coherent and accurate statistics on the economy.

Reference period: The calendar year, or the 12-month fiscal period for which the final day occurs on or between April 1st of the reference year and March 31st of the following year.

Collection period: April to October


  • Retail and wholesale
  • Retail sales by type of store

Data sources and methodology

Target population

The target population consists of all retail establishments operating in Canada for at least one day between January and December of a calendar year. Direct sellers and operators of vending machines are excluded from the target population of this survey.

The survey population is the collection of all retail establishments from which the survey can realistically obtain information. The survey population will differ from the target population due to difficulties in identifying all the units that belong to the target population because of a possible lack of detailed information for some units, particularly small businesses with low sales levels.

The survey population is comprised of all statistical establishments coded to NAICS 441 through 453 on Statistics Canada's Business Register, as well as those small unincorporated businesses which are classified to the retail industry.

Instrument design

The survey questionnaires comprise generic modules that have been designed to cover the retail industry. These modules include revenues and expenses. The questionnaires also include industry-specific modules designed to ask for financial and non-financial characteristics that pertain specifically to this industry.

In order to reduce respondent burden, smaller firms receive a characteristics questionnaire (shortened version) which only include the industry-specific modules. For smaller firms, revenue and expense data are extracted from administrative files.

The questionnaire was developed in consultation with potential respondents, data users and questionnaire design specialists.


This is a sample survey with a cross-sectional design.

The frame is the list of active enterprises and establishments that were selected for Statistics Canada's Business Activity, Expenditure and Output Survey. This frame provides basic information about each firm, including address, industry classification, and information from administrative data sources. This information, initially coming from Statistics Canada's Business Register, has also been updated and expanded through Statistics Canada's Business Activity, Expenditure and Output Survey.

Prior to the selection of a random sample, enterprises are classified into homogeneous groups (i.e., groups with the same NAICS codes and same geography) based on the characteristics of their establishments. Then, each group is divided into sub-groups (i.e. small, medium, large) called strata based on the annual revenue of the enterprise.

Following that, a sample, of a predetermined size, is allocated into each stratum, with the objective of optimizing the overall quality of the survey while respecting the available resources. The sample allocation can result in two kinds of strata: take-all strata where all units are sampled with certainty, and take-some strata where a sample of units are randomly selected.

The total sample size for this survey is approximately 4,300 enterprises.

Data sources

Data collection for this reference period: 2015-04-27 to 2015-10-23

Responding to this survey is mandatory.

Data are collected directly from survey respondents and extracted from administrative files.

Data are collected annually primarily through electronic questionnaire, while providing respondents with the option of receiving a paper questionnaire, replying by telephone interview or using other electronic filing methods. Follow-up for non-response and for data validation is conducted by telephone or fax.

View the Questionnaire(s) and reporting guide(s).

Error detection

Error detection is an integral part of both collection and data processing activities. Automated edits are applied to data records during collection to identify reporting and capture errors. These edits identify potential errors based on year-over-year changes in key variables, totals, and ratios that exceed tolerance thresholds, as well as identify problems in the consistency of collected data (e.g. a total variable does not equal the sum of its parts). During data processing, other edits are used to automatically detect errors or inconsistencies that remain in the data following collection. These edits include value edits (e.g. Value > 0, Value > -500, Value = 0), linear equality edits (e.g. Value1 + Value2 = Total Value), linear inequality edits (e.g. Value1 >= Value2), and equivalency edits (e.g. Value1 = Value2). When errors are found, they can be corrected using the failed edit follow up process during collection or via imputation. Extreme values are also flagged as outliers, using automated methods based on the distribution of the collected information. Following their detection, these values are reviewed in order to assess their reliability. Manual review of other units may lead to additional outliers identified. These outliers are excluded from use in the calculation of ratios and trends used for imputation, and during donor imputation. In general, every effort is made to minimize the non-sampling errors of omission, duplication, misclassification, reporting and processing.


When non-response occurs, when respondents do not completely answer the questionnaire, or when reported data are considered incorrect during the error detection steps, imputation is used to fill in the missing information and modify the incorrect information. Many methods of imputation may be used to complete a questionnaire, including manual changes made by an analyst. The automated, statistical techniques used to impute the missing data include: deterministic imputation, replacement using historical data (with a trend calculated, when appropriate), replacement using auxiliary information available from other sources, replacement based on known data relationships for the sample unit, and replacement using data from a similar unit in the sample (known as donor imputation). Usually, key variables are imputed first and are used as anchors in subsequent steps to impute other related variables.

Imputation generates a complete and coherent micro data file that covers all survey variables.


When some enterprises have reported data combining many units located in more than one province or territory, or in more than one industrial classification, data allocation is required. Factors based on information from sources such as tax files and Business Register profiles are used to allocate the data reported on the combined report among the various estimation units where this enterprise is in operation.

The sample used for estimation comes from a two phase sampling process. An initial sampling weight (the design weight) is calculated for each unit of the survey and is simply the multiplication of the inverse of the probability of selection from each phase. It is then adjusted to take into account units that might have been misclassified (large units found in a stratum of small units for example). In addition, the sampling weights derived are modified and adjusted using updated information from taxation data. Using a statistical technique called calibration, the final set of weights is adjusted in such a way that the sample represents as closely as possible the taxation data of the population of this industry.

The weight calculated for each sampling unit indicates how many other units it represents. The final weights are usually either one or greater than one. Sampling units which are "Take-all" have sampling weights of one and only represent themselves; units with larger than expected size are seen as misclassified and their weight is usually adjusted so that they only represent themselves.

The sampling unit being the enterprise, it can represent numerous locations which might contribute to different parts of the population (different sub-industries, province/territory, etc.). Each location is considered an estimation unit. The characteristics of the estimation units are used to derive the domains of estimation, including the industrial classification and the geography. Estimation for the survey portion is done by simple aggregation of the weighted values of all sampled locations that are found in the domain of estimation. Estimates are computed for several domains of estimation such as industrial groups and provinces/territories, based on the most recent classification information available for the location and the survey reference period. It should be noted that this classification information may differ from the original sampling classification because records may have changed in size, industry or location. Changes in classification are reflected immediately in the estimates.

In the case of the ineligible for sampling portion (also called take-none portion) of the target population defined in Statistics Canada's Business Activity, Expenditure and Output Survey, taxation data is simply aggregated to come up with an estimate. If an estimate is required and taxation data is not available, modeling using auxiliary taxation data is done in order to create data for all requested variables for each unit in the take-none portion. These are also simply aggregated to produce the estimate. The overall estimate includes the estimates from both the surveyed portion and the take-none portion.

Quality evaluation

Prior to the data release, combined survey results are analyzed for comparability; in general, this includes a detailed review of: individual responses (especially for the largest companies), general economic conditions, coherence with results from related economic indicators, historical trends, and information from other external sources (e.g. associations, trade publications, newspaper articles).

Revisions and seasonal adjustment

Revisions in the raw data are required to correct known non-sampling errors. These normally include replacing imputed data with reported data and respondent corrections to previously reported data.

Raw data are revised, on an annual basis, for the year immediately prior to the current reference year being published. That is, when data for the current year are being published for the first time, there will also be revisions, if necessary, to the raw data for the previous year.

Data accuracy

All surveys are subject to sampling and non-sampling errors. Sampling error occurs because population estimates are derived from a sample of the population rather than the entire population. Non-sampling error is not related to sampling and may occur for various reasons during the collection and processing of data. For example, non-response is an important source of non-sampling error. Under or over-coverage of the population, differences in the interpretations of questions and mistakes in recording, coding and processing data are other examples of non-sampling errors. To the maximum extent possible, these errors are minimized through careful design of the survey questionnaire, verification of the survey data, and follow-up with respondents when needed to maximize response rates.

Measures of sampling error are calculated for each estimate. Also, when non-response occurs, it is taken into account and the quality is reduced based on its importance to the estimate. Other indicators of quality are also provided such as the response rate.

Both the sampling error and the non-response rate are combined into one quality rating code. This code uses letters that ranges from A to F where A means the data is of excellent quality and F means it is unreliable. These quality rating codes can be requested and should always be taken into consideration.


Date modified: