Labour Force Survey: Robust and reliable

February 13, 2015

by Chief Statistician Wayne R. Smith

In recent months, news coverage has contained repeated errors of fact concerning Statistics Canada's labour statistics program. I would like to take this opportunity to set the record straight and to reassure data users and all Canadians that the Labour Force Survey program is every bit as robust today as it has ever been. It provides an excellent guide to labour-market conditions when used with an appropriate appreciation of how the data are produced and with a focus on trends rather than month-to-month changes.

What is the Labour Force Survey and what is it designed to do?

The Labour Force Survey (LFS) was established in 1945 to provide a timely, affordable and comprehensive reading of the state of Canada's labour market. Today, the LFS is Canada's largest monthly sample survey, involving 56,000 households and about 110,000 Canadians every month. The survey is designed to provide key estimates of the population aged 15 and over, the labour force, the employed and the unemployed, and their characteristics at the national, provincial, census metropolitan area and economic region level. It is also specifically designed to provide the best possible measure of month-to-month change in key variables within approximately three weeks of data collection.

To manage costs and ensure timely data, Indian reserves and some remote areas are currently excluded from the survey. The size of the labour force in these areas makes it highly unlikely that any local change could significantly affect the estimates of employment and unemployment for Canada and provinces. Nonetheless, smaller geographic areas are important and Statistics Canada uses the National Household Survey to measure labour conditions in remote areas.

Myth: The Labour Force Survey has been adversely affected by the quality of the 2011 Census data.

Not true. The LFS estimation process depends only on the demographic data generated from the census of population conducted every five years. The 2011 Census of Population was conducted on a mandatory basis and achieved response rates equivalent to those achieved in previous censuses. Data from the 2011 National Household Survey play no role in the estimation process. The LFS data are entirely as robust as in the past.

Myth: Statistics Canada has reduced funding for the operation of the Labour Force Survey program.

Not true. Statistics Canada has made no reduction to the funding of the LFS program (or any of the other programs producing key macroeconomic statistics). On the contrary, the agency is undertaking a major investment in the survey and its infrastructure.

Myth: Labour Force Survey employment estimates are more volatile than in the past.

Not true. Every month, about 600,000 people enter employment and a similar number leave. In any given month, the size of the real labour force is subject to random upward and downward fluctuations as businesses open, close and retool and as Canadians enter the labour force or retire, are laid off, or are recalled to work. There is no reason to expect that real-world movements would fluctuate smoothly from month to month.

Despite its extraordinary size, the LFS is a sample survey and introduces additional statistical randomness, on top of real-world random fluctuation, when measuring change in the number of employed people. The two random movements—real and statistical—are not correlated and can partially offset each other. The total employed labour force in Canada is 17.9 million, so even a month-to-month change of 40,000 is around 0.25%, far less than 1% of total employment. A very large expansion in the LFS sample would be needed to significantly reduce the statistical component of this month-to-month randomness in the movement of the level of employment, but would do nothing to remove real-world random fluctuations.

Some recent commentary has suggested that the month-to-month changes in the employed as measured by the LFS have become more volatile. Statistics Canada responded to this claim with an easily replicable study, still available on the agency website, tracking the history of month-to-month changes. The study clearly demonstrated that there has been no increased volatility in the size of month-to-month movements. The study also presented techniques that can be used to reduce the random noise, should analysts wish to do so.

Myth: Statistics Canada made a large correction to employment estimates in January 2015.

Not true. To maintain the accuracy of the LFS, Statistics Canada rebases the LFS estimates after each census using updated annual population estimates. This is standard statistical practice. In January 2015, the revised estimates were published and resulted in a decrease in the estimate of total national employment from 17,953,600, as originally reported in December 2014, to 17,851,500, a downward revision of 0.6%. An adjustment of this magnitude over a period of five years is, in the world of sample surveys, very small.

As a result of the revisions, measured growth in employment between December 2013 and December 2014 fell from 186,000 to 121,000. Some analysts argued this was a large revision. In fact, this revision was similar in magnitude to previous LFS revisions with the exception of 2005 that showed little change (+72,000 in 2000 and −70,000 in 2011).

Myth: U.S. estimates are better than the Labour Force Survey estimates.

Not true. The United States produces labour-market information from two different surveys. The survey producing unemployment estimates is a survey of households designed very much like Canada's LFS. The published employment estimates are from a survey of employers that covers only some classes of workers and most, but not all, industries. By contrast, Canada's LFS covers all industries and classes of workers.

The U.S. employer survey, measuring employment, generates an initial estimate every month based on partial response and revises this estimate every month for three months as additional data become available. In addition, each March, data from the employer survey are revised again to reflect new benchmark data for the sampling frame and new estimates of seasonal patterns. These revisions can be substantial.

While Canada also has an employer survey, its approach is to use one set of internally coherent data from the LFS to measure both employment and unemployment. The LFS data are typically based on a 90% response rate and are not subject to revision in subsequent months. The LFS data is revised annually to incorporate new seasonal factors and periodically to reflect new population estimates (as was done recently). This approach to revisions provides users with historically comparable data.

None of the above is to suggest that either approach is preferred. Each country has adapted its program to its particular needs and priorities.

Myth: Statistics Canada cannot produce a survey equivalent to or as timely as the U.S. current employment survey.

Not true. As mentioned above, Statistics Canada already produces an employer-based survey of employment. This survey is produced with a lag relative to the LFS. This lag is due to the use of significant amounts of administrative data, rather than survey data, to reduce production costs and reduce the reporting burden on Canadian businesses.

If requested, Statistics Canada could redesign the survey to be as timely as the U.S. version, but it would involve a significantly greater respondent burden and cost. Statistics Canada has felt this is unwarranted. Even though the LFS and the employer survey both measure month-to-month changes in paid employment, these changes are not always of the same size or even direction; however, the short- and long-term trends are similar. It is debatable whether the additional information obtained would justify the cost and imposition on businesses.


Canada's Labour Force Survey is very healthy and robust. Real employment is subject to random changes from month to month, and it is pointless to attempt to guess the size of movement from any one month to the next, given random movement in real-world employment. Users should focus their attention on short-, medium- and long-term trends.

This blog is an expanded version of a commentary by the Chief Statistician of Canada published in The Globe and Mail on February 6, 2015.

Please note that comments are moderated. It may take some time for your comments to appear online. For more information, consult our rules of engagement.

User comments

How much does the Canadian LFS cost? How much more expensive would it be if we used the US employment survey method?

The Canadian LFS costs approximately $21 million per year. In Canada, we already have a survey, the Survey of Employment, Payroll and Hours (SEPH), that is similar to the U.S. employer-based survey. SEPH uses a significant amount of administrative data, which is only available a few weeks after the reference period. Therefore, the release of SEPH is not as timely as the LFS. We would have to completely change this approach and replace the use of administrative data with a very large survey, which would be costly and increase the burden on businesses. It is difficult to estimate the cost, but it would be significant.

Hmmm, pretty interesting! How does the Canada's LFS compare to the labour stats methods gathered by other developed countries, such as England?

The concepts and definitions of employment and unemployment adopted by the Canadian survey are based on those endorsed by the International Labour Organisation (ILO), as is the case in other developed countries. Our methods are also similar to those of other developed countries. However, each country might modify some aspect of their Labour Force Survey. For example, the UK uses ILO concepts, but releases its LFS data on a quarterly basis.


Date modified: