Canadian Labour Force Survey
Note: In 2018 Statistics Canada celebrated 100 years of its establishment. Over the years StatCan has been anointed the best statistical agency in the world because of its quality statistical information and its excellence in statistical methodology. The authors would like to thank Dr. Ivan P. Fellegi, Chief Statistician of Canada Emeritus for his inputs on the early history of the Canadian Labour Force Survey. They also thank the many SSC colleagues who have contributed to the success of LFS over the years. Please note a version of this article was published in the Statistics Canada internal publication @STATCAN.
The Canadian Labour Force Survey (LFS) is a Statistics Canada flagship, mission-essential survey first begun in November 1945. For several decades it brought excellence and leadership to the household surveys program in Canada. The LFS provides timely estimates of labour market conditions in Canada, providing the same quality and relevance today as it did more than 70 years ago.
The LFS began as a quarterly survey and was designed to meet the urgent need for reliable estimates of labour market conditions in the post-WWII transition from a wartime to peacetime economy. To meet timely data demands, it became a monthly survey in 1952, and in 1960 it was designated the official measure of Canada’s unemployment and employment condition. The LFS laid the foundation of probability-based surveys for StatCan and inspired the long history of sound statistical methodology and principles in all of the agency’s surveys.
The labour market information provided by the LFS is among the most timely and important measures of the Canadian economy’s overall performance. The monthly official estimates published today include the unemployment rates and employment totals at national, provincial and sub-provincial levels. The LFS plays a central role in the national statistical system: it is the largest monthly household survey conducted by StatCan and its survey frame, sample and processing systems support a wide range of other household surveys.
Excellence in survey methodology
In its 70-plus year history, major innovations have been introduced to the LFS in the areas of survey methodology, questionnaire content, collection methods, processing techniques and computer systems, to name a few. Methodologists in the LFS pursue advancements in every step of the survey design process, particularly in stratification, sample selection, collection and processing, editing and imputation, weighting estimation, and variance calculation. Over the years many innovations have been successfully implemented, in some cases becoming internationally accepted practices.
When the LFS was launched it constituted a major change in the production of official statistics in Canada. Led by Nathan Keyfitz, it was the first complete and thorough implementation of survey sampling. Regional offices were also established in order to maintain a local staff to conduct the survey. As a result, production was much faster and estimates were much timelier. Since that time, through redesigns, the LFS has remained a vehicle of choice for the implementation of solid and innovative methods.
Nathan Keyfitz worked for the Dominion Bureau of Statistics from 1936 to 1959
Major sample redesigns
Major methodological enhancements to the LFS have been introduced via sample redesigns, which occur following each decennial census. Census information is used to create a new survey design, and a new sample is introduced three-four years later, after extensive development and consultations with stakeholders.
The main objective of the 1955 LFS sample redesign was to extend the survey’s coverage to more remote and rural areas of the provinces. In the 1964 redesign, methodologists led by Dr. Ivan Fellegi worked on extensive stratification and a more uniform distribution of the sample. These efforts improved the reliability of the estimates significantly and enabled the release of separate estimates for all provinces for the first time.
Dr. Ivan P. Fellegi, Chief Statistician Emeritus of Canada
The 1975 LFS redesign, led by Richard Platek, implemented a substantial increase in sample size allocation — from about 35,000 households in the 1960s and 70s to 55,000 households. This increase enabled the official release of more detailed estimates at national and provincial levels, and expanded the number of reliable sub-provincial estimates.
M.P. Singh led the 1984 redesign, which endeavoured to be more cost effective. Innovations to the design included the elimination of one of the three sampling stages in the rural areas, more statistically rigorous procedures for stratification and delineation of sample units, the introduction of the integrated weighting method and the use of external population benchmarks in producing more reliable estimates for sub-provincial levels, and increased use of the telephone for conducting follow-up interviews. With this redesign came a major addition to the administration of the Employment Insurance program’s sample size.
In 1995 M.P. Singh and Jack Gambino led another LFS redesign, which saw the introduction of income strata for the first time and enabled special surveys to better target their samples. The new design included less clustering and more sampling in urban areas. The elimination of the place names design in remote areas, and the elimination of the special area design in general, further simplified the LFS. A new method of compensating for household non-response was introduced, taking into account patterns of non-response that varied according to the number of months a household had been part of the survey.
In the 2005 redesign, methodologists led by Jack Gambino and Edward Chen (with early ground work conducted by Eric Rancourt), a more cost-effective design was introduced. Innovations included using the Address Register to reduce field listings, contacting first-month respondents via telephone, reducing samples in expensive collection areas, and introducing computer mapping for all areas of Canada. Additional provincially funded samples in British Columbia and Alberta—and later Manitoba — were introduced to meet special requirements for these provinces.
The current LFS design, launched in January 2015, was further simplified. We introduced a one-stage design in Prince Edward Island and eliminated three-stage designs in every other area. Additional steps were taken to identify remote and hard-to-access areas that could be excluded from the sample frame with minimal impact on estimates. Field listings are now coordinated with the census listing program to further reduce cost. Furthermore, the listing applications used by interviewers for LFS and the census were unified into one application.
The adoption of a single sample selection method in all areas led to a major simplification in sampling methodology, which has allowed for the Division responsible for geography to directly maintain cluster boundaries on an on-going basis as street networks are updated. A web-based application was made available to regional offices in order to reduce the need for paper maps and a method was introduced to reserve and select independent clusters for surveys that use the LFS frame.
Besides the regular introduction of new sample designs, in the last 30 years there have been significant improvements to the LFS based on research and development. Some examples include the introduction of integrated weighting and Jackknife variance estimation in the late 1980s, composite estimation in the late 1990s and the use of the “bootstrap method” for the variance calculation starting in January 2015.
The Telephone First Contact program reduced the number of personal visits to households and the web-based response option, using an electronic questionnaire, was implemented after extensive studies. The use of the new Household Survey Frame services and products from the Address Register, with its standardized listing of addresses, has improved the efficiency of costly initial listing while the Small Area Estimation program has provided users with reliable small area estimates.
In the early 1990s, the LFS’ coverage was extended to the northern territories in consultation with territorial representatives. The LFS was in use when Nunavut was created and it provided territorial statistics soon after its birth. The territorial design update in 2011 led to better stratification and reduction of under-coverage in the northern territories.
The seasonal adjustment technique was introduced to the data series in the early 1970s in order to meet users' changing requirements. Indeed, computer systems have evolved over time to keep pace with changing technology; from manual methods of sampling to processing in mainframe, and later UNIX applications, there has been much evolution of the different systems needed to process and publish the LFS results efficiently.
The authors, Edward J. Chen and Eric Rancourt. Both were chiefs of methodology for the Labour Force Survey Methods at different points in time. Eric is now the Director General of Methodology Branch at Statistics Canada and Edward is a chief in Statistical Integration Methods Division.
As such, interviewers have moved from manual methods to the use of computer applications such as Computer Assisted Personal Interviews and Computer Assisted Telephone Interviews with several major iterations of hardware and software enhancements. In particular, the introduction of Computer Assisted Interviews beginning in November 1993 was a key innovation in data collection.
Excellence in all other areas
Major innovations and enhancements to the LFS, led by subject matter experts, were introduced in many other areas over time, including two major questionnaire redesigns in 1976 and 1997, which improved data quality and ensured that relevant labour market information is collected and published. The IT systems have been continually redesigned to use the corporate systems and the dissemination of LFS information has improved dramatically over the years to meet users’ demands.
The historical series of statistics from the LFS is recalculated every five years (after each census) in a process called “rebasing,” in order to better align outputs with the most recent geopolitical boundaries. Industry and occupational classifications have evolved over time as well, and the LFS follows suit to meet users’ ever-changing requests.
Special and supplemental surveys
The LFS infrastructure has provided a convenient and cost-effective starting point for a number of other household surveys over the years. Large current surveys using the LFS frame include the Canadian Community Health Survey and the Survey on Household Spending. Combined, these surveys use more of the frame sample every year than the LFS does. In addition, LFS infrastructure and interviews provide the opportunity for supplemental surveys or Fast Track Options/Disaster Catastrophe Evaluation so that information can be collected quickly based on household characteristics.
A tradition of excellence
The continued success of the LFS is the result of a large number of contributors, starting with the analysts, methodologists, IT specialists, geographers and data collection specialists working at StatCan. Without a doubt, the long history of the LFS also owes its success to the regional offices, interviewers, data users, researchers in universities and the Canadian public for their input and participation.
Over the years, Statistics Canada has been anointed the best statistical agency in the world and the LFS has remained innovative throughout the decades. It has a fine tradition of pursuing excellence, innovation and cost efficiency while meeting changing demands and data requirements. Today, the LFS continues to uphold its strong reputation and it continues to evolve with Statistics Canada’s modernization efforts toward more client-centric services, data integration through modern methods and approaches, and increased collaboration to better inform Canadians.