The 2015 revisions of the Labour Force Survey (LFS)
Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.
The Labour Force Survey (LFS) provides estimates of employment and unemployment, which are among the most timely and important measures of performance of the Canadian economy. With the release of the survey results only 10 days after the completion of data collection, the LFS estimates are the first of the major monthly economic data series to be released.
Statistics Canada has an established history of applying a standard revision to its LFS estimates following the release of final population estimates from each census. Along with this revision, other changes are introduced at the same time. The purpose of this document is to explain each of these changes. It should be noted that these changes do not involve modifications to the questionnaire nor the content of the survey.
In brief, starting with the release of January LFS data on the 6th of February 2015, estimates will be based on the 2011 Census population counts and sub-provincial estimates will be based on 2011 Census boundaries. Following is a short summary of each change:
- Population rebasing: Up until December 2014, labour force estimates had been based on population data from the 2006 Census. As of January 2015, the estimates have been adjusted to reflect population counts from the 2011 Census adjusted for net undercoverage. These revisions have had some minor impacts on the LFS estimates, while rates of unemployment, employment and labour force participation were essentially unchanged. Given the changes to the estimates were minimal, only revisions back to 2001 were necessary. See Section 2.0 and Appendix I for more details.
- Geographic boundaries: Census metropolitan areas (CMAs), Economic regions (ERs) and Census agglomerations are now based on 2011 Census boundaries rather than 2006 boundaries. No new regions were added, but boundaries were modified for CMAs and ERs. See Section 3.0 and Appendix II for more details.
- Methodological enhancements to imputation: The overall imputation strategy did not change, but the list of variables used to create the imputation groups for donor imputation was reviewed and updated to include industry. At the same time, codesets used for certain age groups and labour force status variables were modified, while the country of birth variable was removed. These changes were implemented historically, starting with January 2008 (See Section 4.0).
- Sample redesign: Starting in January 2015 and ending in June 2015, an updated sample design will be implemented. This redesign defines new strata based on the most recent 2011 Census information, whereas the previous design was based on the 2001 Census. As a result, the sample allocation will also change. See Section 5.0 and Appendix III for more details.
- Industry and occupation classification update: Often, the LFS moves to more recent classification structures for industry and occupation when data are rebased. These updates will take place in January 2016. At that time, the current North American Industry Classification System 2007 (NAICS 2007) will be updated to NAICS 2012 and the National Occupational Classification– Statistics 2006 (NOC-S 2006) will be updated to the NOC 2011.
2.0 Population rebasing
The LFS uses population estimates of its target population, which are derived independently from the survey, as benchmarks for producing survey estimates. These population estimates start with a Census base and are then updated using administrative data between censuses to reflect the current population of Canada. Using these population counts reduces the sampling variability and the risk of coverage bias of survey estimates. Proper population numbers are crucial in determining estimates from a sample survey like the LFS. In order to transform the results of the sample into estimates, each individual in the sample is assigned a weight indicating the number of persons in the population that he or she represents.
The Census base used for obtaining these estimates is updated 28 months after each new Census is conducted. Beginning with the release of the January 2015 survey data, population estimates used by the LFS will change from a 2006 Census base to a 2011 Census base
These updated population counts result in more accurate labour force estimates compared to using those with a 2006 Census base. As the population estimates move away from their original Census base over the years, imprecisions in the administrative data used to update the numbers tend to become more pronounced.
The LFS uses population counts that include an adjustment for net Census undercoverage. In any census, there is both overcoverage and undercoverage: some people are counted more than once or should not be counted, while others are not counted but should be included. The net result is usually undercoverage. In the 2011 Census, the undercoverage rate was 2.3%.
As an example, in December 2014, the 2006-based estimate of the target population was 0.3% higher (+73,000) than the 2011-based estimate. This means population estimates were overestimated for that survey month. The differences in the two sets of population estimates can be more pronounced for some age and sex groups, provincial and sub-provincial areas.
Since the difference between the old estimates (based on the 2006 Census) and the new estimates (based on the 2011 Census) was relatively small, an historical revision back to the start of the series was deemed unnecessary. The year 2001 was chosen to provide an historical series for sub-provincial estimates.
For more detailed information on the component changes for this population rebasing, please see Appendix I.
2.1 Impact of population rebasing on LFS estimates
Estimates of the LFS population have been revised from January 2001 to December 2014. In general, estimates have been revised downward and the magnitude of the revision varies over the period.
At the national level, for the working-age population (aged 15 and over), the differences are negligible between January 2001 and June 2006 (Chart 1). Over that period, the new estimates are lower by 24,000 or less (-0.1%) than the old population counts.
Beginning in July 2006 through to January 2010, differences move to their peak, increasing from -27,000 to -97,000 (or from -0.1% to -0.4%). The difference between the revised and unrevised population estimates diminishes from February 2010 to May 2013, to about -0.1% (-20,000).
From June 2013 onward, the difference starts to increase again, reaching -73,000 (-0.3%) in December 2014. This December 2014 difference is similar to the rebasing in 2011 when estimates were rebased from the 2001 to 2006 Census population counts. In December 2010, the difference in the estimates was -79,000 or -0.3%.
This population rebasing did produce a small break in the population estimates between December 2000 and January 2001 at the national level as well as for Manitoba and Saskatchewan. This can also be seen for certain age groups. These shifts, however, are minor for any of the labour force estimates or rates.
As a result of the revision to the population estimates, the levels of employment were little changed from 2001 to 2006. Starting in February 2007, the gap started to increase, reaching -0.6%, or -102,000 in December 2014.
For the unemployment estimates, positive and negative differences between the revised and unrevised estimates were observed throughout the 2001 to 2014 period, with larger differences between 2008 and 2014. Estimates for those not in the labour force were revised downward from 2001 to 2007 and slightly upward from 2008 to 2014.
When evaluating the data, it is important to keep in mind that estimates for some age groups for both men and women have been affected differently by the population revision.
Differences by gender and age groups
As can be seen from Chart 2, population estimates for both men and women of working age were revised downward, but more so for men than women.
Chart 3 shows that population estimates for youths and those aged 55 and over were affected differently by the revision than was the case for people aged 25 to 54. By December 2014, the new population count for youths was higher by 1.3%, while it was 0.2% higher for those aged 55 and over. For men aged 25 to 54, the population estimate was 1.5% lower and for women in the same age group, 0.5% lower.
Because these groups have different labour market characteristics (for example, a higher proportion of 25 to 54 year-olds are employed and working full time), the relationship between the new and the old estimates can be complex. See Section 6 for more details.
The impact of the population revision also differs by province (Charts 4, 5, and 6). Population estimates for Prince Edward Island, Ontario, British Columbia, and Manitoba have been revised downward. The magnitude of the revision varies over the period, similar to the national estimates.
Alternatively, for Newfoundland and Labrador, Quebec, Alberta, and Saskatchewan, estimates were either stable or revised slightly downward from 2006 to 2008 and were then revised upward. The magnitude of the revision generally increased over the period for Newfoundland and Labrador and Quebec, while it started to decline in 2012 for Alberta and Saskatchewan.
New Brunswick and Nova Scotia were the provinces with the smallest percent differences between the initial and rebased population estimates.
In January 2001, Manitoba was the province with the largest difference between the rebased and non-rebased estimates. Rebased population estimates for Manitoba and Saskatchewan show a slight level shift between December 2000 and January 2001. However, this level shift is minor for the labour force estimates or rates for these provinces.
3.0 Geographic boundary changes
With the change to the 2011 Standard Geographical Classification (SGC) from the 2006 SGC, boundaries were modified for some of the sub-provincial areas such as Census metropolitan areas, Economic regions and Census agglomerations.
New tables for all sub-provincial areas have been created based on the 2011 Census boundaries and rebased going back to 2001.
A CANSIM table concordance is available in Appendix II. Vector concordance tables are available on the following webpage: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvDocument&Item_Id=213135&SurvId=152713. For CANSIM vector concordance tables for 282-0109, 282-0110, 282-0118, 282-0119, contact Client services (toll-free 1-866-873-8788; 613-951-4090; firstname.lastname@example.org).
3.1 Census metropolitan areas (CMAs)
Although there were no new CMAs added, seven existing CMAs had boundary modifications between the 2006 SGC and the 2011 SGC. These are, in Quebec: Saguenay, Québec, Sherbrooke, Trois Rivières, Montréal, Ottawa-Gatineau, Quebec part; and in Ontario: Guelph. Among these, Guelph (+6.0%) and Saguenay (+3.4%) had the largest change in the revised and unrevised population estimates in March 2001, while the other CMAs had changes of 3% or less.
In addition, an improved method was used for population allocation in this rebasing. Ten CMAs were affected by this change: St. John’s, in Newfoundland and Labrador; Moncton and Saint John in New Brunswick; and Peterborough, Hamilton, Thunder Bay, St. Catharines-Niagara, London, Windsor, and Barrie in Ontario. Among these, Barrie had the largest change in population estimates (-7.6%) in March 2001 while the other CMAs changed by less than 2%.
3.2 Economic regions (ERs)
As a result of the move to the 2011 SGC, there were small boundary changes for three Economic Regions (ERs): Campbellton-Miramichi and Fredericton-Oromocto in New Brunswick; and North Coast and Nechako in British Columbia. The changes in population estimates in these ERs were less than 1% in March 2001.
To improve data quality of the LFS estimates, three small ERs were combined: South Coast-Burin Peninsula and Notre Dame-Central Bonavista Bay in Newfoundland and Labrador (1020, 1040); South Central and North Central in Manitoba (4620, 4640); and Banff-Jasper-Rocky Mountain House and Athabasca-Grande Prairie-Peace River (4840, 4870) in Alberta. These new combinations were done in consultation with the provinces involved. Excluding the Territories, the LFS now publishes estimates for 66 ERs, seven of which are combined for data quality purposes.
The combined ERs of Parklands and Northern (4670, 4680) in Manitoba had the largest percent change in the population estimates between the 2006 and 2011 rebasing, at -8.1% in March 2001. With this rebasing, the estimates of the population living on reserves was higher bringing the sampled population estimates for this region lower than in the previously published estimates. The ER of Northern (4680) has over 70% of Manitoba’s population living on reserve and was therefore impacted by this change more than other regions.
3.3 Census agglomerations (CAs)
The redesign of the LFS sample changed its provincial sample allocation (See Section 5.0). As a result, the data quality for some of the previously published CAs will become insufficient and will no longer be published. This is the case for the following CAs: Drummondville; Saint-Hyacinthe; Shawinigan; Rimouski; Baie-Comeau; and Sept-Îles in Quebec; Belleville, Kawartha Lakes, and Brockville in Ontario; Grande Prairie in Alberta; and Kamloops, Vernon, Courtenay, Duncan and Dawson Creek in British Columbia.
Furthermore, with the new sample allocation, the data quality for the three following CAs is now sufficient for publication: Saint-Georges in Quebec; Thompson in Manitoba; and Fort St. John in British Columbia. These CAs will be available on request in March 2015.
4.0 Improved imputation strategy
Imputation is the method used to replace missing data values in cases of both person and item non-response in a survey. Complete person non-response occurs when all questionnaire data for a person in a sampled household are missing, while item non-response means that some, but not all data, are missing. For detailed discussion on the imputation strategy used in the Labour Force Survey, see Chapter 5 of Methodology of the Canadian Labour Force Survey (2008).
For the 2015 rebasing, the overall imputation strategy was not changed. However, the list of variables used to create the imputation groups for the donor imputation was updated in order to reflect current response patterns and relationships between key variables. As a result of this update, one variable was added (industry), some were modified (age group and labour force status) and one was removed (country of birth: born in Canada; born in the United States; or born elsewhere). These changes were implemented historically, starting with January 2008.
See Sections 6 and 7 for the impact of these changes on the estimates.
5.0 Sample redesign
The Labour Force Survey (LFS) plays a central role in the national statistical system in several ways. First, the LFS provides monthly estimates of employment and unemployment which are among the most timely and important measures of the overall performance of the Canadian economy.
Secondly, the Employment Insurance Act has designated the LFS as the official source of monthly unemployment rates for the 62 Employment insurance regions (EIRs) used in the administration of the Employment Insurance program.
Last, but not least, the LFS infrastructure (sample, interviewers, processing systems) supports a wide range of other Statistics Canada household surveys that are conducted in response to the policy and information needs of government.
In order for the LFS to continue to uphold these three key roles, the sampling frame must be up-to-date, and the estimates must be sufficiently reliable to support the various uses of the data.
Every ten years, after the decennial population census, the LFS undergoes a sample redesign to reflect changes in population characteristics and new definitions of geographical boundaries. Up until December 2014, the LFS sample design was based on information from the 2001 Census. It reflected the population size, provincial distribution, and the sub-provincial boundaries as of 2001. Since that time, there has been significant population growth, change in population and labour market characteristics, as well as realignment of municipal and Census metropolitan area (CMA) boundaries. Therefore, the sample distribution needs to be adjusted to reflect the most current characteristics of the labour market, population and geography.
The LFS design strata, which are a way to divide the frame in order to make sampling more efficient, are set out to be homogeneous with respect to some key labour market variables. However, the strata become less efficient the further the design is from the source year (i.e. 2001) and as the population and labour market characteristics shift over time. This redesign defines new strata based on the most recent Census information (2011).
The sample is allocated to provinces and strata within provinces in the way that best meets the need for reliable estimates at various geographic levels. The following guidelines were used in the 2005 sample allocation for estimates of unemployment:
- A coefficient of variation (CV or standard error relative to the estimate) of less than 2% for Canada, and 4% to 7% for the provinces.
- CVs of less than 25% for Economic regions.
- CVs of less than 15% for 3 month moving average unemployment estimates of Employment insurance regions (EIRs).
- Since most CMAs are also EIRs, setting objectives for the EIRs also guarantees the quality of the estimates for the CMA.
These guidelines continue to be used in the 2015 sample allocation with the following exception. The sample allocation to the provinces and to Employment Insurance regions was modified by using quality targets at a level that provides a consistent standard error when estimated unemployment levels for a region have been below 5% of the labour force. This modification prevents the allocation algorithms from automatically increasing sample sizes in areas of low unemployment, which would be at the expense of the other regions as the overall sample size has been maintained at the same level as in the last design.
This revised sample allocation has resulted in some changes to the target sample size for most provinces. The table below compares the national and provincial sample sizes from the 2015 design to those from the 2005 design. As can be seen, the largest absolute change was a decrease of over 900 sampled households per month in Ontario, which was offset by increases in the sample in the three Prairie provinces and Quebec. For the four Atlantic provinces and British Columbia, the changes were all relatively small. These revisions also reflect demographic and labour market changes.
The new sample design for the provinces is being “phased-in” one rotation at a time as households come into the sample (i.e. on the birth rotation). This process will start in January 2015 and will be fully implemented by June 2015.
5.1 Other sample changes with the 2015 sample redesign
Under the 2005 design, two Economic regions (ERs) were combined at the design stage (480: Cote-Nord and 490: Nord-du-Quebec) and three other pairs were combined during estimation (670: Parklands and 680: North; 750: Prince Albert and 760: Northern; 960: Nechako and 970: North Coast). In the 2015 redesign, all collapsing has been implemented at the design phase and includes three new pairs of ERs (See Section 3.2 for more details).
No changes were made to the sample design for the territories since it was updated in January 2011.
For more changes on the sample redesign, see Appendix III.
6.0 Impact on labour market estimates since 2001
Charts 7, 8 and 9 present monthly, seasonally adjusted revised and unrevised estimates of labour force characteristics at the national level by age and sex, while Tables 1 and 2 include differences based on annual averages for selected years.
As previously mentioned, employment levels were revised downward with the new population counts (Chart 7). This occurred for the total working-age population, particularly for people aged 25 to 54, but more so for men than women (Chart 8).
For persons aged 15 to 24, the difference between the revised and unrevised employment estimates varied throughout the 2001 to 2014 period. From 2001 to 2006, there was little difference between the two series, but from 2007 to 2010, the revised estimates were lower than the unrevised. Since 2012, the revised employment estimates were higher than the unrevised.
On the other hand, for those aged 55 and over, the revised employment estimates were very similar to the unrevised estimates throughout the period of revision, with little to no impact on the rates of unemployment, participation and employment.
As with the old series, the new employment estimates show that the employment peak before the economic downturn was in October 2008. With the revised estimates, the employment trough occurred in June 2009 as opposed to July 2009 in the unrevised estimates.
From October 2008 to June 2009, the decline in employment was very similar in both revised (-2.5%, or -427,000) and unrevised estimates (-2.4%, or -418,000). Between June 2009 to December 2014, employment grew at the same rate (+7.1%) for both the new and old estimates.
Compared to the unrevised data, unemployment levels and rates were little changed from the revised estimates. The national unemployment rate in December 2014 was 6.7%, a difference of 0.1 percentage points from the unrevised rate.
Both the employment and participation rates were revised downward from February 2008 onwards, as revised employment grew at a slightly slower pace than the unrevised. The revised employment rate in December 2014 was 61.3%, a difference of 0.2 percentage points compared to the unrevised estimates. The revised participation rate was also 0.2 percentage points lower, at 65.7%, compared to the unrevised.
By December 2014, revised employment estimates were higher than the unrevised for five of the provinces. Newfoundland and Labrador had the largest gap, at 3.6% higher, while Quebec, Nova Scotia, New Brunswick and Saskatchewan were less than 1% higher than the unrevised estimates (Charts 10 and 11).
Revised employment estimates were lower for four provinces, with British Columbia having the largest gap between the revised and unrevised estimates (-2.7% in December 2014). Manitoba followed with a difference of -1.6%, then Prince Edward Island (-1.2%), and then Ontario (-1.0%). In Alberta, the differences were mainly in 2012 and part of 2013, with little change by December 2014 (-0.1%).
Both the population rebasing and the update to the imputation strategy had an impact on the employment levels by industry and class of worker (Charts 12, 13, and 14). Specifically, comparing the revised and unrevised data, there was a general shift downward in the number of workers employed in industries such as educational services, public administration, professional scientific and technical services, and finance, insurance, real estate and leasing. At the same time, there was a general shift upward in the number of workers employed in construction, accommodation and food services, and ‘other services’.
For class of worker estimates, there was a general shift downward in the number employed in the public sector. There was little change in the number of private sector employees and a general shift upward in the number of self-employed (Chart 15).
7.0 Labour market in 2014
Looking at the annual averages for 2014 (Table 1 and 2), revised employment levels are lower than the unrevised. This was due to the unrevised series being overestimated for those aged 25 to 54 years of age, especially men. Revised employment levels in 2014 were lower by 1.3% for core-age men and 0.3% for core-age women
On the other hand, unrevised employment estimates were underestimated for youth and older workers. Revised employment estimates in 2014 were higher than the unrevised ones by 1.0% for youth and 0.2% for those 55 years and over.
Full-time employment was revised downward in 2014 (-0.6%) while part-time employment was revised upward (+0.5%). The downward revision for full-time employment was mainly for core-age men (-1.2%), although there was also a slight downward revision for core-age women (-0.3%). Part-time employment was revised upward for youth aged 15 to 24 (+1.6%) and for those aged 55 and over (+1.5%).
Unemployment levels were revised slightly upward in 2014 (+0.3%). This was mainly due to youth unemployment, which was revised upward by 2.2%. Unemployment levels for those aged 55 and over and for core-aged workers were revised downward (-1.0% and -0.4% respectively).
National revised rates of employment, unemployment and participation were little changed from the unrevised rates in 2014. Provincially, rates of participation and employment were slightly higher than the unrevised rates in the provinces of Newfoundland and Labrador, Prince Edward Island and New Brunswick. For the central and western provinces, revised participation and employment rates were down slightly, by 0.1 to 0.2 percentage points compared to unrevised rates (Table 3 and 4).
Tables 5 and 6 show that the major areas of employment growth from 2001 to 2014 (starting in 2006 for immigrant status), as well as the share of employment in 2014 by age, sex and immigrant status, were little changed between the revised and unrevised estimates.
As mentioned in Section 6, both the population rebasing and the update to the imputation strategy had an impact on employment levels by industry and class of worker. While the revised employment growth from 2001 to 2014 for certain industries and by class of worker (especially public sector employees) are somewhat different (Table 5), the revised and unrevised share of employment in 2014 was little changed (Table 6).
Also, despite little change in the revised and unrevised share of employment by highest level of education in 2014, revised employment growth from 2001 to 2014 for those with a university education was slightly lower compared to unrevised (62.1% vs. 64.8%).
8.0 The Territories
Estimates for Yukon, the Northwest Territories and Nunavut were also adjusted to reflect population counts based on the 2011 Census of population. Estimates were revised back to 2001 for Yukon and the Northwest Territories and 2004 for Nunavut.
For both Yukon and the Northwest Territories, population estimates were revised slightly downward from 2001 to 2010, and were revised upward from 2011 to 2014. In Nunavut, rebased population estimates were higher than the previous estimates from 2008 to 2014.
The annual average revised employment estimates in 2014 for Yukon and Nunavut were unchanged from the unrevised. For the Northwest Territories, revised 2014 employment estimates were 0.5% higher than unrevised estimates (Table 7).
Unemployment levels were little changed between the new and old estimates for the three territories in 2014. However, the revised unemployment rate for Nunavut was 0.6 percentage points higher than the old rate, and 0.2 percentage points lower for the Northwest Territories.
At the same time, revised participation and employment rates were higher than the unrevised rates in the Northwest Territories, while they were lower in Nunavut and Yukon.
9.0 Estimates for the Aboriginal population
The weights applied to the Aboriginal working-age population living off-reserve were also updated to reflect population changes. As mentioned in the publication Aboriginal Peoples in Canada: First Nations People, Métis and Inuit released on May 8, 2013, the Aboriginal population increased at a pace nearly four times faster than the non-Aboriginal population from 2006 to 2011. The two main factors that account for this growth of the Aboriginal population are high fertility and more individuals identifying themselves as Aboriginal persons.
The revised LFS working-age population estimate for the Aboriginal population living off-reserve in 2014 was 22% higher (+155,000) than the old estimate. Revised population estimates were higher than the old estimates for the three Aboriginal groups: First Nations, Métis and Inuit. By province, revised Aboriginal population estimates were higher than unrevised in all provinces but mostly in Quebec, the Atlantic region, British Columbia and Ontario (Table 7 ).
As with most other population groups, there was little difference between the revised and unrevised unemployment rates for the Aboriginal population. However, the revised participation and employment rates were generally lower than the unrevised rates by Aboriginal identity and by province.
10.0 Update to actual hours seasonal adjustment methodology
The seasonal adjustment methodology for the actual hours series has been improved to better reflect hours lost due to holidays in the reference week for self-employed workers. All actual hours series have been revised back to the start of the series based on this new methodology.
All seasonally adjusted actual hours series are adjusted for the timing of the reference week as this varies from year to year. There is also an adjustment to reflect holiday occurrences during reference weeks. Holidays that sometimes fall into the reference week include Family Day, March break for some provinces, Easter Friday and Monday, Thanksgiving, and Remembrance Day.
This adjustment is derived from hours lost due to the holiday as reported by respondents of the Labour Force Survey, with the exception of self-employed workers. As hours lost due to holidays are not reported for the self-employed, a model is used to estimate and remove systematic fluctuations due to holiday occurrence in the reference weeks. This model is based on special time series regression in a manner similar to the calendar adjustment performed for reference week location.
To better reflect the actual hours from the self-employed workers, the seasonally adjusted total actual hours worked series will now be derived as the sum of the three seasonally adjusted classes of workers (public employees, private employees and self-employed). The provincial series will be slightly modified to match this improved seasonally adjusted actual hours total.
Historically, the total actual hours series was directly seasonally adjusted. For more information on direct and indirect seasonal adjustment, please consult Statistics Canada Quality Guidelines on Seasonal adjustment and trend-cycle estimation.
Report a problem on this page
Is something not working? Is there information outdated? Can't find what you're looking for?
Please contact us and let us know how we can help you.
- Date modified: