Measuring employment and unemployment in Canada and the United States – a comparison
- 1. Introduction
- 2. Conceptual and Methodological Differences in Employment Measurements: Household and Business Statistical programs
- 3. Conceptual and Methodological Differences in Employment Measurements: Statistics Canada and Bureau of Labor Statistics
- 4. Concluding Remarks
Statistics related to employment are among the most watched indicators released by National Statistical Agencies. In Canada, Statistics Canada releases two sets of sample-based estimates of employment on a monthly basis: from the Labour Force Survey (LFS) of households and from the Survey of Employment, Payrolls and Hours (SEPH) of businesses.Note 1 Comparisons are naturally made to the two analogous surveys released by the Bureau of Labor Statistics in the United States: the Current Population Survey (CPS) of households and the Current Employment Statistics (CES) program survey of establishments (businesses). More specifically, employment, unemployment and participation rates from these surveys are often used to compare labour market trends in the two countries.
This report presents information to inform users who wish to compare estimates of employment and unemployment from the Canadian and American surveys. Considerations include concepts, methods, seasonal adjustments, timeliness, revisions, and principal uses. A brief description of the four surveys is presented in Appendix 1. The concepts used in these surveys are presented in a comparative fashion in Appendix 2. The main differences are summarized in Appendix 3.
This report compares these surveys from different perspectives. First, household and business employment surveys are examined in general in Section 2, whether they are conducted in Canada or in the United States. Within each type of survey (household or business), there are conceptual and methodological differences in how Canada and the United States measure employment and unemployment. These differences are documented in Section 3.
2. Conceptual and methodological differences in employment measurements: Household and business statistical programs
Statistics Canada and the Bureau of Labor Statistics use a number of program instruments to develop measures of employment, including household surveys, business surveys and administrative records.
The household and business statistical programs in both countries are in essence quite different. The household surveys collect information about the employment status and demographic characteristics of their members, while the business statistical programs receive information on jobs from businesses directly or from government payroll remittance reports. The nature of their outputs reflects the designs of the statistical programs, together producing a comprehensive portrait of the labour force.
At the most general level, both household and business survey programs are consistent in their measures of general employment trends. Taking into account conceptual differences such as those described in this and the next section, the two surveys in each country over time track each other fairly well (Charts 1 and 2), differing on average by less than 1%.Note 2 The smaller residual differences, more difficult to explain, may be due to further conceptual or methodological differences, including how quickly changes to employment status are reflected in government administrative files. Small differences are also observed to be more prevalent during periods of economic downturn or recovery.
Most users, however, seek more than a general sense of employment trends, and understanding their conceptual and methodological differences becomes important.
Section 3 highlights conceptual and methodological differences in employment measurements released by Statistics Canada and the Bureau of Labor Statistics. This is not straightforward, as both countries make use of household and business employment statistical programs in different ways.
To put the comparisons of Section 3 in context, the rest of this section highlights in a general sense differences in concepts and methods between a household labour force survey (like the Canadian LFS and American CPS) and a business employment survey (like the Canadian SEPH and American CES).
The main conceptual difference between household labour force surveys and the business employment surveys is that the household surveys produce estimates of the number of employed persons, while the business surveys estimate the number of jobs. While the LFS and CPS household surveys collect information on multiple jobs, they count persons only once according to their main job, regardless of how many jobs they have, while SEPH and CES include an individual more than once if they hold multiple jobs.
Another difference is that the LFS and CPS consider as employed certain persons absent from work without pay and people on maternity or paternity leave, while SEPH and CES only include persons receiving pay during the reference period.
The LFS and CPS also provide measures of both employment and unemployment including their rates, while SEPH and CES only provide measures of employment.
Beyond the employment and unemployment numbers, there are differences in the characteristics that are measured. Through interviews with household members, the household labour force surveys allow for more detail on their demographic, educational and family characteristics. The information covered through business employment surveys on the other hand provides for more detailed estimates by industry level and by size of enterprise, including average weekly earnings and hours paid.
Household labour force surveys provide a broader picture of employment—including the agricultural, fishing and trapping, and private household services sectors—and the self-employed; and can provide data on the unemployed. Business surveys, with a focus on payroll employment, do not include these sectors.
The reference period of the household labour force surveys is generally the calendar week that includes mid-month (15th in Canada, 12th in the United States).
The business statistical employment programs, on the other hand, are primarily based on a pay period, which can cover a week, two weeks, a semi-month or a month. In Canada, the reference period is the last week of the month, while in the United States it includes the 12th of the month. Month-to-month changes can be affected by which weeks are reported on. In many cases this is addressed through seasonal adjustment methods discussed later on. When a reported pay period is longer than one week, the data are adjusted by dividing the number of days of work reported on the respondent’s return by the number of working days covered by the pay period in question.
Sampling and non-sampling errors
Household and business surveys are subject to both sampling and non-sampling errors, while the administrative data component of business surveys is only subject to non-sampling error. The sub-section on quality provides an assessment of the extent of the sampling errors.
3. Conceptual and methodological differences in employment measurements: Statistics Canada and Bureau of Labor Statistics
The fundamental concepts and methods used by Statistics Canada and the US Bureau of Labor Statistics are quite similar. Both countries use internationally recognized methods and indicators to measure the performance of the labour market. In some cases they integrate information obtained from surveys with information from administrative data files on individuals and on businesses.
There are differences in the concepts and methods of household and business surveys that are common to both countries. These were highlighted in the previous section.
There are also differences in how these methods are applied in Canada and in the United States, in particular with respect to the thresholds within international standards that may influence comparisons of the more detailed results. These are highlighted below. Appendix 3 provides a summary of these differences.
Studies have examined to what extent conceptual differences affect comparisons between the two countries. For example, at the national level, on average between 2007 and 2013, adjusting the Canadian unemployment rate to American concepts lowered the Canadian rate by about one percentage point.Note 3,Note 4
3.1 Official national employment and unemployment numbers
In Canada, Statistics Canada releases the official monthly employment and unemployment figures from the LFS household labour force survey. It provides timely monthly estimates that are important indicators used to measure the performance of the Canadian economy.
In the United States, the Bureau of Labor Statistics releases the official monthly employment figures from the CES business employment survey and the official monthly unemployment figures from the CPS household labour force survey in a joint report. These estimates are the first main economic indicator of current economic trends in a particular month and are inputs to many gauges of the U.S. economy.
Said another way, the official national estimates of employment in Canada and the United States come from two different types of statistical programs: a household labour force survey and a business employment survey, respectively. The official national estimates of unemployment for these countries both come from their household labour force surveys. These distinctions should be kept in mind throughout this section.
Canada and the United States release their official national estimates of employment and unemployment at about the same time, generally within the first week following the reference month. Depending on the estimates, there are differences in the timing of the release of the more detailed data.
Employment details by industry at the national level
In Canada, SEPH employment estimates by industry are released about two months after the reference month, once payroll information provided by businesses has been reported and processed.
In the United States, the CES releases national employment estimates by industry within one week of the reference month, at the same time as the total national figures.
Employment details by industry at the regional level
In Canada, SEPH estimates of national, provincial and territorial employment by industry are released at the same time, two months after the reference month. Estimates by industry below this level are not part of the official release.
In the United States, in collaboration with state authorities, CES estimates by industry at the state level are released within three weeks of the national employment figures. They are produced independently of the national estimates, using only the state-specific portion of the national database. Summed state level CES estimates should not be compared to national CES estimates. County/metropolitan estimates are released within four weeks of the national employment figures. Modeling techniques are used for the employment estimates by industry for metropolitan areas and counties given the increased sampling error with the smaller sample sizes.
Regional employment and unemployment
In Canada, the LFS national and regional employment and unemployment estimates are released at the same time, one week following the reference month. Regional estimates are provided for provinces, territories, metropolitan areas and employment insurance economic regions.
In the United States, in collaboration with state authorities, the Bureau of Labor Statistics releases regional and state estimates of employment and unemployment within three weeks of their national release. Data for metropolitan areas produced by the Bureau of Labor Statistics and under the Local Area Unemployment Statistics (LAUS) Program are released one week later.
3.3 Household labour force surveys
The concepts used by Canada and the United States to measure employment and unemployment, are closely aligned with international recommendations defined by the United NationsNote 5 and the International Labour Organization (ILO).Note 6 Within the details of the recommended definitions, in some cases it is acknowledged that countries may set their own criteria based on legislation and national needs.
The lower age threshold for employment statistics differs in Canada (15 years) and the United States (16 years). While they use a minimum age of 15 in their published labour force tabulations, recommendations from the United Nations and from the ILO allow that the minimum age be set in accordance with the conditions of each country.
In Canada, there is no national law on minimum age for employment. The minimum age to work is determined by the provinces and territories. The LFS collects information for household members aged 15 years and over.
In the United States, to be eligible to participate in the CPS, individuals must be 15 years of age or over. However, employment data are only published for those aged 16 and over since those under 16 are limited in their labour market activities by compulsory schooling and child labour laws.
Employed and unemployed persons
In Canada and the United States, the LFS and CPS divide the working-age population into three mutually exclusive groups—employed, unemployed and not in the labour force. There are differences for some of the details within the definitions of employed and unemployed, also affecting the population considered not to be in the labour force.
Both Canada and the United States consider as employed persons those who did unpaid work in a family-owned enterprise operated by someone in their household. In Canada, the LFS does not set a minimum numbers of hours during the reference week for such work; in the United States, the CPS has a threshold of at least 15 hours. The ILO standard does not set a minimum threshold, though it recognizes that countries may prefer to have one.
In the LFS, the following three groups of people considered unemployed in Canada are deemed by the CPS to not be participating in the labour force in the United States: (1) people who were looking for work, but who looked only at job ads (passive search); (2) people who had not looked for work, but who reported that they had a job that would start in the next four weeks; and (3) people who had reported that they were not available to work because of personal or family responsibilities, or vacation.
Conversely, full-time students who reported that they were looking for full-time work are deemed by the LFS to not be participating in the Canadian labour force, but in the United States are considered unemployed in the CPS.
Regarding the Canadian inclusions above, the ILO standard for those considered unemployed also includes those who are waiting to start a new job. The ILO standard does not, however, include those searching for a job only by looking at ads. For the ILO standard on temporary absences, there are no guidelines on exceptions such as for personal or family responsibilities, or vacation. The American inclusion of full-time students who report that they are looking for full-time work is in line with the ILO standard.
In Canada, full-time employment is defined as usually working 30 hours or more per week, while in the United States the threshold is 35 hours per week. There is no international standard on what the threshold should be.
Mandatory and voluntary surveys
In Canada the LFS is mandatory under the authority of the Statistics Act, while in the United States the CPS is voluntary under the authority of the Office of Management and Budget (OMB).
In Canada, the sample size for the LFS is about 56,000 households, or about one in 240 Canadian households. The survey provides the official national employment and unemployment figures by demographic and other characteristics as well as regional figures for provinces, territories, metropolitan areas and employment insurance economic regions.
In the United States, the sample size for the CPS is about 60,000 eligible housing units, also referred to as households. With a national population ten times greater, this is a proportionately smaller survey, representing about one in 2,000 American housing units. The survey provides national estimates of unemployment and demographic, race, ethnicity and other characteristics of employment not available from the CES.
Estimation and weighting
Estimation and weighting are the processes for converting survey information that is collected from a sample of the population to estimates for the entire population.
These processes for deriving LFS and CPS estimates of employment and unemployment are similar in many respects. In their most basic form, household sample data are assigned a basic weight, generally equal to the inverse probability of selection. For example, in an area where 2 percent of the households are sampled, each household would be assigned a basic weight of 1/.02=50. As such, each response would be deemed to be representative of fifty households.
Three main adjustments to these weights are then applied that are common to the LFS and the CPS: (1) to account for additional sampling or sub-sampling due to unexpected population growth in an area and the maintenance of the overall sample size over time; (2) to account for differences in household non-response patterns for sub-groups of the population; and (3) to account for coverage errors by calibrating to independent estimates of the population, with additional composite calibration taking advantage of information from the overlapping sample of the previous month.
Regarding adjusting the weights for coverage, the process for the CPS is the more complex, introducing race and ethnicity characteristics.
For the composite calibration, the main difference is in the additional variables used. The LFS uses provincial level estimates from the previous month on employed by industry and class of worker (public sector, private sector), while the CPS uses national level employed and unemployed by age-sex-race and age-sex-ethnicity groups, and state level employed and unemployed aggregates.
Employment and unemployment indicators are often subject to seasonal variation attributable to such things as changes in weather, school schedules and reduced or expanded production.Note 7 When these seasonal events follow a regular pattern every year, their influence on statistical trends can be minimized by seasonal adjustment techniques.
Canada and the United States use similar methods and software (versions of ARIMA) to remove the influence of seasonal events on trends by adjusting the data from one month to another. The methodology generally allows seasonal adjustment for the more aggregate levels such as national and provincial/state levels by broad age group and sex, high-level industry aggregates or class of worker (employees or self-employed). Seasonal adjustments can also adjust for calendar effects such as moving holidays or variations in the number of weeks between the reference periods. In Canada, for smaller groupings where seasonal adjustment is not possible, comparisons are done on a year-to-year basis, i.e., September to September.
One minor difference in the seasonal adjustment processes used by the LFS and CPS for the national estimates is the level of detail of the age groups used for the adjustment. The LFS at the national level uses six broad age-sex groups (15-24, 25-54, 55+, crossed by sex) while the CPS uses four broad age-sex groups (16-19, 20+, crossed by sex). Seasonal adjustments at the provincial and state levels are done independently each using four broad age-sex groups.
A greater difference is in the additional raking step by the LFS to ensure consistency between national and provincial level estimates. In the United States, the detailed estimates by state are not necessarily consistent with national aggregates. Another difference is that in the United States the seasonal adjustment for states and a few sub-state areas account for sampling error.
The processes for revising the household survey estimates are similar in Canada and the United States in that they revise the seasonally adjusted series and may adjust for population controls based on the previous census and estimates of demographic change since then. These revisions generally do not affect overall trends but can impact month-to-month changes. There is no direct benchmarking for employment.
In Canada, the LFS revises its seasonally adjusted series every year. The revisions are applied back three years, which is a long enough period to provide historically comparable time series. Further revisions are made every five years following the quinquennial census of population. In 2015, historical data based on the 2011 Census were revised back to 2001.
The extent of the revisions on Canadian estimates of employment and unemployment at the national level tends to be very small. When rebasing following the Canadian 2011 Census, the monthly estimates of employment between 2001 and 2014 were in general lower, reaching a peak difference in January 2010 of -0.4%.Note 8 While the trends remained the same, the trough during the 2008-2009 economic recession of employment levels was revised one month from July 2009 to June 2009. The impact on the rates of employment and of unemployment was negligible.
In the United States, revisions are a part of the annual seasonal adjustment process. Seasonally adjusted estimates for the last five years are subject to revision to incorporate new seasonal factors.
Outside of seasonal adjustment, national estimates from the CPS are rarely revised. Each year the survey incorporates updated population controls from the Census Bureau. The impact of these population control adjustments is typically small, with negligible effects on employment-population ratios and unemployment rates. In 2015, for example, the impact on levels of employment and unemployment were about 0.2% to 0.3%, with no impact on the unemployment rate.Note 9 The population controls that include new information from the decennial census are typically larger. Some of the decennial census impacts have been large enough that data were revised back to the census reference year; this last occurred after the 1990 and 2000 Censuses but not after the 2010 Census. Research series are created for users that smooth out level shifts in the labour force and in employment estimates resulting from population control adjustments.
State-level estimates are revised annually through the Local Area Unemployment Statistics (LAUS) program that estimates employment and unemployment at this level using time-series models, drawing on several sources, including the CPS, the CES and state employment insurance (EI) programs. These annual revisions reflect updated population data from the U.S. Census Bureau, including the decennial census, revisions in the other data sources and model re-estimation.
The quality of official employment and unemployment estimates produced in Canada and the United States is very high. Their internationally recognized survey methods, large sample sizes and use of census administrative sources generate accurate and detailed estimates of employment and unemployment.
Two indications of quality at the national level were discussed earlier in this report. The first is the close comparability of the independent household and business surveys. In Canada (LFS and SEPH) and in the United States (CPS and CES), these surveys track each other very well. Taking into account conceptual differences, the levels of employment at the national level differ on average by less than 1%. The second is the extent to which estimates are revised. For the household labour force surveys (LFS and CPS), recognizing there is no direct benchmarking for employment, revisions from updates to the population control totals have a negligible effect on the rates of employment and unemployment. An additional indication of data quality is the standard error, which is a measure of sampling error.
Sampling errors and non-sampling errors
Sampling errors occur because observations are obtained from a sample rather than from the entire population. Estimates based on a sample can differ from statistics that would have been obtained if a complete census had been taken using the same instructions, interviewers and processing techniques. This difference is called the sampling error of the estimate. At national levels with large sample sizes, sampling errors tend to be very small. The errors are generally larger for sub-groups, such as at a regional level or for specific industries. Statistically, these types of errors are to be expected and are quantifiable at all levels.
Non-sampling errors are not statistical in nature and are thus more difficult to measure. They can come from a number of sources, including from the use of proxy responses, the interpretation of questions by respondents, the interpretation of answers by interviewers, non-response, and coding of industry and occupation. While sound quality assurance practices in the planning, testing, training and operational phases of a survey minimize these types of errors, they do occur.
Sampling standard errors
If the process of selecting a sample from the population were repeated many times, approximately 68% of the intervals from one standard error below the estimate to one standard error above the estimate would include the true value. This level of confidence increases to 90% with intervals about +/-1.6 times the standard error, and to 95% with intervals twice the standard error. When the estimate is a measure of change such as employment, the larger the change compared to the standard error, the better the chance that the change is real, as opposed to a change due to sampling variability.
The standard error in both countries is small, with little variation from year to year.
In Canada, the LFS standard error associated with the monthly changes in employment using the LFS was about 28,700Note 10, or about 0.2%. Over time, it has remained relatively constant, at around 29,000, even if the employment base on which change is measured has increased over time.Note 11
In the United States, the CPS standard error of the monthly changes in employment is about 265,000Note 12, or 0.2 % when calculated based on the percent change in employment from month-to-month.
Volatility of official monthly estimates
The monthly estimates of employment from the household and business surveys over time give a robust reflection of labour market trends in Canada and in the United States. In the short time frame however, month-to-month changes in employment and unemployment can be much more volatile. The volatility of the estimates tends to increase with regional and industry sub-groups given their smaller sample sizes.
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 people enter the labour force or retire, are laid off, or are recalled to work. Month-to-month fluctuations in real-world movements are to be expected.
Sudden large or unexpected monthly movements may be due to a combination of three factors: actual movements in the labour market, sampling error and non-sampling error. At times these factors may cancel each other out; at other times they can amplify a real change.
Over time, the changes observed in employment are primarily reflective of real changes in the labour market. Due to their statistically random nature, sampling errors tend to diminish in importance over a longer period and non-sampling errors are generally smaller with much less effect on the estimates.
Measures can be taken to reduce the volatility of the month-to-month changes. Seasonal adjustments can address irregular changes due to calendar-based events. Users can also combine estimates for several consecutive months, for example using three- or six-month moving averages or using trend-cycle estimates.
In Canada, there has been increased scrutiny of the LFS results, particularly with regard to unexpected movements for a given month. However, the survey has exhibited much the same degree of month-to-month fluctuation recorded since the 1970s. From 1976 to 2013, using seasonally adjusted estimates, the standard deviation was on average 0.24% when calculated based on the percent change in employment from month-to-month.Note 11 There is no evidence that the month-to-month changes in employment have recently been more volatile.
In the United States, volatility can be seen by how often the monthly change in CPS employment numbers exceed a certain threshold. From 1994 to 2006, using originally published data not seasonally adjusted, the monthly change in CPS employment numbers exceeded 500,000 (about twice the average standard error) twenty-three times.Note 13In comparison, the larger CES survey showed a change of this magnitude only once in this period. This measure is not available in Canada.
3.4 Business employment surveys
Mandatory and voluntary surveys
In Canada, the component of SEPH used to estimate employment by industry under the authority of the Statistics Act is based on a federal government administrative source with mandatory coverage of all payroll businesses.
In the United States, the CES is conducted by the Bureau of Labor Statistics in collaboration with states, under the authority of the Office of Management and Budget (OMB). The survey is mandatory in four states.
In Canada, the component of SEPH used to estimate national employment by industry is a census of business and government agency administrative data. There is thus no sampling.
In the United States, the sample size for the CES is about 143,000 businesses and government agencies covering 588,000 establishments, about just over one in 15 establishments. This large sample allows for sufficient accuracy and industry detail required for the official national employment figures.
In Canada, the component of SEPH used to estimate national employment by industry is a census of administrative data. There is thus no estimation.
In the United States, the CES estimates of employment involve a two-part estimator. In its most basic form, the first part matches businesses that are part of two consecutive monthly samples. The change in their total number of employees between these two months is calculated as a ratio that is applied to the national estimate of the previous month. As this does not take into account new businesses or those that go out of business, the second part uses modelling techniques to estimate births and deaths of businesses. There is an unavoidable lag between an establishment opening for business and it appearing on the survey sample frame.
Employment indicators are often subject to seasonal variation attributable to such things as changes in weather and reduced or expanded production. When these seasonal events follow a regular pattern every year, their influence on statistical trends can be minimized by seasonal adjustment techniques. As for the household labour force surveys, Canada and the United States use similar methods and software (ARIMA) to remove the influence of seasonal events on trends by adjusting the data from one month to another.
Both SEPH and CES apply seasonal adjustment factors to employment estimates using industry data at the 3-digit level. The main difference is the additional raking step by SEPH to ensure consistency between national and provincial level estimates by industry. In the United States, the state estimates by industry are not necessarily consistent with national aggregates. As well, CES seasonally adjusts for the number of weeks between reference periods. SEPH does not do this adjustment—its reference week is always the last week of the month.
Initial revisions for the employment estimates are more extensive in the United States than in Canada, a reflection of their different methods and timelines.
In Canada, preliminary SEPH estimates are generally released two months after the reference month and are revised the following month. These revisions are relatively small, as by the initial release most of the government administrative information has been collected. There is a further annual historical revision (generally back four years) as a result of a detailed analysis and validation of the data. There is no benchmarking as SEPH is already based on an administrative census of businesses, which covers all payroll establishments in the country. The difference between the preliminary and revised monthly payroll employment estimates is quite stable. For the ten-year period 2001-2010, the absolute mean revision was less than 0.2%.Note 14
In the United States, preliminary CES estimates are generally released one week following the reference month. More timely than those from SEPH, these estimates undergo more extensive revisions over the next two months to take into account new information as it is received. The Bureau of Labor Statistics tracks collection rates for the CES sample on a monthly basis.Note 15 The average annual collection rates for 2014 increased from 78.1% for the first preliminary release to 95.3% for the second preliminary release and 96.9% for the third final sample-based release. The CES also benchmarks its estimates annually primarily to the Quarterly Census of Employment and Wages (QCEW) program based on Unemployment Insurance tax reports.Note 16 This benchmarking assumes that the error associated with the annual revision accumulates in a linear fashion. The differences between the third preliminary estimates and annual benchmarked estimates are quite stable, with a ten-year absolute mean revision of less than 0.3%.
Two indications of quality at the national level for business employment surveys were discussed earlier in this report.
The first is the close comparability of the independent household and business surveys. In Canada (LFS and SEPH) and in the United States (CPS and CES) these surveys track each other very well. Taking into account conceptual differences, the levels of employment at the national level differ on average by less than 1%.
The second is the extent to which the estimates are revised. In Canada, SEPH is a census of payroll businesses to complement the official estimates of employment with detailed industry breakdowns. The longer time lag before their release reduces the effect of delayed reporting by businesses. As a result, subsequent revisions at the national level are relatively small.
In the United States, CES is the official source of national employment figures requiring more timely estimates. Compared to SEPH they are revised more extensively over the next two months to reflect additional reported information. Subsequent revisions including benchmarking to administrative data at the national level are relatively small.
Sampling standard error
A third indication of data quality is the sampling standard error described earlier. This only applies to the CES as employment from SEPH is estimated from an administrative census. The CES standard error for the initial preliminary estimate of the month-to-month change in employment published in 2015Note 17 is about 65,000. With a much larger sample, this is about one quarter the standard error for CPS.
Employment estimates produced from business administrative data are also subject to non-sampling errors. This is the case for the Payroll Deduction (PD7) remittance forms as the foundation for SEPH and the Quarterly Census of Employment and Wages (QCEW) program used to benchmark the CES. Non-sampling errors include late reporting by businesses as they submit their monthly reports, or in calculating the number of employees for establishments with multiple remittances in a month. As well, the processing of forms can be quite complex and may also be a source of error. In both Canada and the United States, sound quality assurance practices in the planning, testing, training and operational phases of a survey program minimize these types of errors.
Volatility of monthly estimates
In Canada, with no sampling error, much of the observed change in the monthly SEPH estimates reflects real change in the labour market. The non-sampling errors just described may also introduce some variability in the observed month-to month changes.
In the United States, the CES is less volatile than the CPS primarily due to its larger sample. During the period from 1994 to 2006, as mentioned earlier the CPS monthly change in employment numbers exceeded 500,000 (about twice the average standard error), twenty-three times.Note 13 The CES, by contrast, showed a change of that magnitude only once during those years.
The economies of Canada and the United States are well served by accurate and timely official measures of employment and unemployment. The methods used are well recognized internationally for their rigor.
The application of the methods in each country best takes advantage of the statistical infrastructure within their respective governments, including the development of administrative data for statistical purposes and the conduct of national household and business surveys. In Canada for example, a large monthly household survey provides detailed measures of persons employed, unemployed and out of the work force. In the United States, a federal-state cooperative program provides timely data each month on employment from a large sample of non-farm establishments.
Studies have examined to what extent conceptual differences affect comparisons between the household labour force surveys of the two countries. For example, adjustment of the Canadian unemployment rate to American concepts has recently lowered it by about one percentage point. Factors contributing most to lowering the Canadian unemployment rate when adjusting to the American concepts include removing passive job seekers (people who were looking for work, but who looked only at job ads) and to a lesser extent persons waiting to start a new job within 4 weeks and persons unavailable to work because of personal or family responsibilities or vacation. This is partly offset when full-time students seeking full-time work are added.
When making comparisons at more detailed levels, one should consider the variables used and the nature of the analysis. For example, when comparing recent trends in national employment statistics by detailed industry, there is a time lag of about seven weeks between the earlier American and the Canadian releases. When doing analysis including concepts such as age and full-time work, the data should be standardized to take into account conceptual differences.
Bernard, André and Guy Gallatly. 2014. Interpreting Monthly Changes in Employment from the Labour Force Survey. Statistics Canada Catalogue no. 11-626-X. Ottawa, Ontario. Economic Insights Series, no. 033.
Bernard, André and Jeannine Usalcas. 2014. The Labour Market in Canada and the United States since the Last Recession. Statistics Canada Catalogue no. 11-626-X. Ottawa, Ontario. Economic Insights Series, no. 036.
Fortier, Susie and Guy Gellatly. 2015. Seasonally adjusted data – Frequently asked questions. Statistics Canada Catalogue no. 11-001-X. Ottawa, Ontario.
Gilmore, Jason and Dominique Pérusse. 2012. Employment, earnings and hours: Monthly data sources at Statistics Canada. Statistics Canada. Ottawa, Ontario. Internal document.
Statistics Canada. 2015. Guide to the Labour Force Survey. Statistics Canada Catalogue no. 71-543-G. Ottawa, Ontario.
Statistics Canada. 2015. Guide to the Survey of Employment, Payrolls and Hours. Statistics Canada Catalogue no. 72-203-G. Ottawa, Ontario.
Statistics Canada. 2015–. Improvements to the Labour Force Survey (LFS): The 2015 Revisions of the Labour Force Survey (LFS). Statistics Canada Catalogue no. 71F0031X. Occasional.
Statistics Canada. 2015–. Labour Force Information. Statistics Canada Catalogue no. 71-001-X. Monthly.
Statistics Canada. 2015.“Labour Force Survey, April 2015.” The Daily. May 8. Statistics Canada Catalogue no. 11-001-X.
Statistics Canada. 2008–. Methodology of the Labour Force Survey. Statistics Canada Catalogue no. 71-526-X. Occasional.
Smith, Wayne R., Chief Statistician. 2015. “Labour Force Survey: Robust and reliable.” StatCan Blog. February 13. Statistics Canada.
Smith, Wayne R., Chief Statistician. 2013. “The Labour Force Survey’s volatility – myth or reality?” StatCan Blog. November 20. Statistics Canada.
Bowler, Mary and Teresa L. Morisi. 2006. “Understanding the employment measures from the CPS and CES survey.” Monthly Labor Review. February. Bureau of Labor Statistics.
Bureau of Labor Statistics. 2015. “All Employee Standard Error Tables” (tables). Current Employment Statistics – CES (National). Washington. D.C.
Bureau of Labor Statistics. 2015. “CES Registry Receipts by Release” (table). Current Employment Statistics – CES (National). Washington, D.C.
Bureau of Labor Statistics. “Employment Situation Technical Note.” Economic News Release. Last updated January 8, 2016.
Bureau of Labor Statistics. “Labor force data derived from the Current Population Survey.” Handbook of Methods. Washington, D.C.
Bureau of Labor Statistics. “Monthly Employment Situation Report: Quick Guide to Methods and Measurement Issues (Payroll Survey).” BLS Information. Last updated February 6, 2015.
Bureau of Labor Statistics. 2015. “Population control adjustments to the CPS.” Labor Force Statistics from the Current Population Survey: Technical Documentation. Washington, D.C.
Bureau of Labor Statistics. 2015. “Seasonal adjustment of CPS estimates.” Labor Force Statistics from the Current Population Survey: Technical Documentation. Washington, D.C.
Bureau of Labor Statistics. “Technical Notes for the Current Employment Statistics Survey.” Current Employment Statistics – CES (National). Last updated June 5, 2015.
Bureau of Labor Statistics. “Updates and adjustments to Current Population Survey standard errors.” Labor Force Statistics from the Current Population Survey. Last updated January 18, 2007.
Sorrentino, Constance. 2000. “International unemployment rates: how comparable are they?” Monthly Labor Review. June. Bureau of Labor Statistics.
U.S. Census Bureau. 2006–. Design and Methodology: Current Population Survey. Technical paper no. 66. Washington, D.C. Occasional.
International Labour Organization. “Definitions of the topics.” LABORSTA. Last updated January 3, 2016.
United Nations Economic Commission for Europe. 2006. Conference of European Statisticians Recommendations for the 2010 Censuses of Population and Housing. New York, N.Y. and Geneva.
The author acknowledges with much appreciation the input and review by senior analysts, methodologists and managers of Statistics Canada and the Bureau of Labor Statistics. A particular note of thanks to Jason Gilmore, Alison Hale and Christel Le Petit of Labour Statistics Division, Susie Fortier and Jack Gambino of Methodology Branch and Isabelle Amano, André Bernard and Guy Gellatly of Analytical Studies Branch of Statistics Canada; and to Kirk Mueller, Current Employment Statistics Benchmark and Special Projects Branch, Dorinda Allard, Megan Dunn and Karen Kosanovich of the Current Population Survey and Ed Robison of Statistical Methods of the Bureau of Labor Statistics. Any errors or omissions are those of the author.