Canada

Building a stronger evidence/policy feedback loop through regular and frequent social data time series

Kari Wolanski

Acting Director, Centre for Social Data Insights and Innovation, Statistics Canada

The pace of change is accelerating

• Era of “poly crisis”: climate risks materializing, geo-political instability, low productivity, high inflation; lingering effects of the pandemic on society such as polarization, less confidence in institutions, increased crime, growing mental health issues

• Society is at a post-pandemic turning point: thirst for timely and granular data, but also information overload; there is a strong need to make meaning, distill critical insights about what is changing, social statistics need to put emerging issues into historical and relative context

• Demonstrating societal progress is key to confidence in government; departments are calling for better impact assessment tools

Social statistics should

help us to see the big picture; put things into

historical and relative perspective

Pandemic sparked innovation

• “Building back better”: Canada’s Quality of Life framework developed during early pandemic unity and optimism (though policy commitment made prior to pandemic)

• Canadian Social Survey (CSS): pandemic provided a strong impetus for timely data; quarterly omnibus survey introduced

• Window of opportunity to understand rapid social change; CSS leveraged to develop a quarterly time series for some Quality of Life indicators

Canada’s Quality of Life framework

• Multi-dimensional well-being: Five thematic outcome domains supported by 84 indicators

• Inclusion: Cross-cutting lens and concurrent investments in disaggregated data to understand distributional differences in well-being

• Sustainability: Cross-cutting lens to bring a long-term perspective, prevention focus

• Budget 2021 funding to: • Fill quality of life data gaps

• Develop a hub to bring together quality of life statistics

6

FOUR TIMES A YEAR…

… a group of civil servants go into the “lock up” procedure in a secure room… When all analysis has been done, the document is approved by a group of top management officials. The report is transmitted to the

adviser of the President of the United States… This procedure is followed every quarter…. The following morning the report is made public… The media report the results almost instantly, politicians

comment,…investment decisions are considered….”

-- Hoekstra, Rutger; Replacing GDP by 2030: Towards a Common Language for the Well-being and Sustainability Community

Some key indicators have a built-in lag time

Social statistics weren’t really built for monitoring real-time progress

Window of opportunity for quarterly time series

Socio-demographic characteristics

• Age • Gender • Immigration status • Visible minority group • Educational attainment • LGBTQ2+ • Urban / rural

Quality of life indicators • Life satisfaction • Sense of meaning and purpose • Future outlook • Loneliness • Someone to count on • Sense of belonging to local

community • Difficulty meeting financial needs • Perceived mental health • Confidence in institutions

Quarterly time series

Fairness and inclusion lens

Q2 2021 Q1 2024

More Canadians are finding it difficult to meet their financial needs

Source(s): Statistics Canada, Canadian Social Survey, Waves 1-11.

More than one third (37.4%) of Canadians found it difficult or very difficult to meet their financial needs by the end of 2023.

(Difficulty meeting financial needs in terms of transportation, housing, food, clothing, and other necessary expenses.)

0%

20%

40%

60%

Q3 2021 Q4 2021 Q1 2022 Q2 2022 Q3 2022 Q4 2022 Q2 2023 Q3 2023 Q4 2023 Q1 2024

Percent reporting financial difficulty, Canadians aged 15 or older, selected sociodemographic or geographic groups, 2021 to 2024

percentagepercentage

Those who experience financial hardship report lower life satisfaction

13

0%

20%

40%

60%

80%

Q3 2021 Q4 2021 Q1 2022 Q2 2022 Q3 2022 Q4 2022 Q2 2023 Q3 2023 Q4 2023 Q1 2024

Percent reporting high life satisfaction, Canadians aged 15 or older, by experiences of self-reported financial difficulty, 2021 to 2024

Did not experience financial difficulty

percentage

Experienced financial difficulty

14

Young adults reporting lower levels of life satisfaction, racialized people a greater decline

Author: H. Foran Source: CSS, multiple waves

0%

20%

40%

60%

80%

Q3 2021 Q4 2021 Q1 2022 Q2 2022 Q3 2022 Q4 2022 Q2 2023 Q3 2023 Q4 2023 Q1 2024

Percent reporting high life satisfaction, Canadians aged 15 or older, 2021 to 2024 - Racialized group

percentage

Racialized

percentage

Not racialized

Source: Canadian Social Survey, Q3 2021 - Q1 2024

0%

20%

40%

60%

80%

Q3 2021Q4 2021Q1 2022Q2 2022Q3 2022Q4 2022 Q2 2023Q3 2023Q4 2023Q1 2024

Percent reporting high life satisfaction, 2021 to 2024 – Selected age groups

percentage

Ages 65+

Age 25 to 34

15

Early signal re: improving future outlook?

Author: H. Foran Source: CSS, multiple waves

40.0%

50.0%

60.0%

70.0%

80.0%

Q3 2021 Q4 2021 Q1 2022 Q2 2022 Q3 2022 Q4 2022 Q2 2023 Q1 2024

Percent reporting a hopeful future outlook, Canadians aged 15 to 24, 45 to 54, 2SLGBTQ+, and racialized, 2021 to 2024

Selected age groups

Ages 45-54

Ages 15-24

40%

60%

80%

Q3 2021 Q4 2021 Q1 2022 Q2 2022 Q3 2022 Q4 2022 Q2 2023 Q1 2024

Racialized group

Not racialized

Racialized

percentage

Eastern Canada and rural communities generally report higher levels of life satisfaction

16

Source(s): Statistics Canada, Canadian Social Survey, Waves 1-9.

Life satisfaction National average:

51%

Demand for Quality of Life Hub

Concluding thoughts

• It’s fun to design frameworks and pick indicators… but shifting the outcome focus reduces incentives for progress

• Consistent reporting of the same indicators teaches people what to expect and what type of change in the data is meaningful; choreography ensures that data dissemination is linked to decision-making processes and can lead to policy action

• Repeated observations are foundational for impact assessment; decisions about course correction/scale

• When selecting your indicators, it’s equally important to plan your cadence and fund your long-term time series

Friends of the Chair of Social and Demographic Statistics

• Lack of an overarching conceptual framework • Difficulty of modeling social phenomena • Growing demand for data disaggregation and timeliness • Declining response rates and rising survey costs • Tight fiscal environments • Decentralized statistical responsibilities • Low public acceptability for administrative data linkages

Main challenges

57th

session UNSC

Strategic recommendations for strengthened

social and demographic

statistics

55th

session UNSC

Mar ’23- Feb ‘24

Year 2Year 1 Year 3

Mar ’24- Feb ‘25 Mar ’25- Feb ‘26

Building blocks for a conceptual framework for social and demographic statistics

OUTCOMES

TIMEPEOPLE

The fabric, object, and unit of measurement for social

and demographic statistics.

✓ Population stock population flows

✓ Characteristics of individuals (data disaggregation

and other intersectional considerations)

Interactions between individuals which

collectively build up a society.

✓ Family structures ✓ Social connections,

✓ confidence in institutions

Qualitative or compound measures to be assessed

both objectively and subjectively.

✓ Objective: education, health, time use (such as unpaid work, work-

life balance and leisure), employment, income, wealth and

housing, among others

✓ Subjective: well-being, including physical and mental health, learning,

sense of purpose, meaningful work

Geography as an element for analyzing social and demographic statistics

and as a link across themes and domains.

✓ Geographic disaggregation

✓ Geospatial analysis ✓ Small-area estimation

Time as an aid in understanding and

anticipating changes in population, distribution

outcomes and relationships.

✓ Time series to monitor progress

✓ Life perspectives ✓ Age-specific lenses

✓ Cohort-based analyses

RELATIONSHIPS PLACES

21

Thank you! Please stay connected.

%Statistics Canada Website The Daily (released at 8:30 am every weekday) StatsCAN Plus StatsCAN App

Surveys and Statistical Programs Data Service Centres My StatCan

Questions? Contact us: [email protected]

22

%

Statistics Canada – Your National Statistical Agency

Delivering insight through data for a better Canada

UNITED NATIONS ECONOMIC COMMISSION FOR EUROPE

CONFERENCE OF EUROPEAN STATISTICIANS

ModernStats World Workshop

21-22 October 2024, Geneva

The statistical production LEGO set: using standard models and tools

to build metadata-driven pipelines at StatCan

Speaker: Flavio Rizzolo, Statistics Canada

Author(s):

Abstract

The development of the ModernStats models, from GSBPM to GSIM, and their related reference architectures,

CSPA and CSDA, provide a solid foundation for better understanding statistical production, informing business

decisions and designing reusable software components. Bridging the chasm between these conceptual

viewpoints and their implementation standards, most notably DDI and SDMX, and figuring out how to use

these implementation standards in a complementary way, has recently received considerable attention in the

international community.

This presentation describes Statistics Canada’s recent attempt at bridging that chasm by deploying a number of

standard-based data/metadata management tools (e.g., Colectica, Aria, Fusion Metadata Registry, Data

Lifecycle Manager) using ModernStat models to guide the decision-making process. Making these tools

developed within very different communities interoperate at a semantic level requires mappings between the

standards together with some “connective tissue” in the form of customized micro-utilities that enable the

creation of data/metadata pipelines across phases of the GSBPM. We describe some key use cases, our progress

to date, challenges found, and the road ahead.

Presented by Vincent Hardy, Ph.D, Chief, Centre for Labour Market Information, Statistics Canada

Creating job-quality profiles: A how-to guide

Job quality profiles: A how-to guide

• In response to interest from members of the Steering Group on Quality of Employment, a subgroup was formed to work on the topic of job quality profiles.

• As part of the activities of the subgroup, a short document providing guidance on how to create job quality profiles was drafted by Canada

– Comments from UNECE, Finland, and Switzerland were incorporated

• This guide will be available on the Expert group’s Wiki page after the meeting

Purpose and requirements

• Premise: quality of employment is not a linear measure that can be effectively captured by a single index (see Eurofound, 2012, p. 15).

– E.g. jobs that are challenging physically can have good wages and a good social environment

• The goal of job quality profiles is to classify jobs into different clusters based on their common quality of employment features

• Requires a dedicated quality of employment survey OR a Labour Force Survey supplement covering several quality of employment dimensions.

Eurofound: Job Quality Indices

Selection of measures

• Initial treatment of variables

– Identify variables that represent different dimensions of quality of employment

• Eurofound job quality dimensions (Eurofound, 2012) is a common starting point.

– If multiple variables are available for one quality of employment dimension, consider constructing an index.

• E.g. combining career prospects with job security to create a “prospects” index

• Verify quality of the index using Cronbach’s alpha or equivalent measures of reliability.

• Select individual indicators or indices based on theory – Note: indicators should be at the job level, not the person level

Applying clustering algorithms • Select appropriate clustering method

• Latent Class Analysis (LCA) or Latent Profile Analysis (LPA)

– Similar techniques, LCA is for categorical variables and LPA is for continuous variables

– Preferable for sample surveys, can produce standard errors and fit statistics

– Possible that a solution fails to achieve the recommended level of fit

• Other clustering algorithms

– K-means (continuous, requires standardization)

– Hierarchical clustering (categorical)

– No clear way to measure the overall quality of the clustering, “forces” cases into clusters

Interpreting the solution

• LCA and LPA models can include covariates

– E.g. sex and age

• Covariates can predict the probability of having a particular job quality profile based on a person’s characteristics (see for example, Chen and Mehdi, 2019)

• Other clustering algorithms directly assign individual cases to clusters, and these can be cross-tabulated with demographic characteristics and other variables.

• Some researchers like to assign labels to job quality clusters.

Example

• Immonen and Sutela (2021)

– Based on the 2018 Quality of Work Life Survey in Finland

– Created quality of employment indices similar to Eurofound’s job quality indices: skills and discretion, social environment, flexibility of working time, physical environment, and work intensity

– A model with five job quality profiles was found to be the best fit for the data.

Distance of the factors describing the quality of working conditions and the average in different job quality profiles

Chart taken from Immonen and Sutela, 2021

Data sources

• European Working Conditions Survey or similar are excellent data sources

• However, many national statistical offices lack dedicated quality of employment surveys.

• Possible solutions: collect a Labour Force Survey (LFS) supplement with at least one representative measure of each quality of employment dimension.

– Use single indicators rather than indices

Alternative: Apply clustering to aggregate data

• If quality of employment data are available across multiple data sources (e.g. LFS supplements) one option may be to classify occupations

– Calculate average scores for different quality of employment indicators by occupation (e.g. ISCO)

– Create a database of occupational scores & apply clustering algorithms at the occupational level.

– Create an identifier variable for the clusters, and merge the cluster identifier back to a representative survey using the occupational code (e.g. LFS)

• Note: experimental approach, see Hardy (2024) for more detailed information

Resources • Chen, W. & T. Mehdi. (2019). Assessing Job Quality in Canada: A Multidimensional Approach. Analytical

Studies Branch Research Paper Series. Statistics Canada, Ottawa.

• Eurofound. (2012). Trends in job quality in Europe. Publications Office of the European Union, Luxembourg.

• Eurofound. (2017). Sixth European Working Conditions Survey – Overview report (2017 update). Publications Office of the European Union, Luxembourg.

• Hardy, V. (2024) “Using aggregate data to generate job quality profiles”. Presented at the Group of Experts on Quality of Employment. UNECE, Geneva.

• Immonen, J. & H. Sutela. (2021) “Job quality profiles reveal division: men clearly more often in “good jobs” than women”. Statistics Finland, Helsinki. Available at: https://www.stat.fi/tietotrendit/artikkelit/2021/job- quality-profiles-reveal-division-men-clearly-more-often-in-good-jobs-than-women/

• Magidson, J. & J. K. Vermunt. (2002) “Latent class models for clustering: A comparison with K-means”, Canadian Journal of Marketing Research, Vol. 20.

Stay connected!

StatsCAN app

Eh Sayers podcast

StatsCAN Plus

The Daily

Website

Surveys and statistical programs

Data service centres

My StatCan

Questions? Contact us: [email protected]

Statistics Canada— Your National Statistical Agency

Delivering insight through data for a better Canada

Presented by Vincent Hardy, Ph.D, Chief, Centre for Labour Market Information, Statistics Canada

Using aggregate data to generate job quality profiles

Presentation outline

1) Overview of the multi-dimensionality of quality of employment

2) Recent attempts to create job quality profiles using working conditions surveys

3) Data gaps and an alternative approach

4) Constructing indices via occupational classifications

5) Creating job quality profiles at the occupational level

6) An exploratory assessment of trends in occupational job quality in Canada

Multi-dimensionality of quality of employment

Quality of Employment Framework | UNECEEurofound: Job Quality Indices

Capturing the multifaceted nature of quality of employment

• Single indicator

– Pro: Clear meaning and policy implications

– Con: Difficult to evaluate the overall quality of a job

• Indices

– Pro: Captures a broader dimension of quality of employment

– Cons:

• Overall quality of employment index typically not recommended (Eurofound, 2012, p.15)

• If using an index representing a single dimension (e.g. working time) jobs could still score differently on other aspects

• Job quality profiles

– Pro: Multi-dimensional, captures overall quality of jobs

– Con: More challenging to measure change over time

Example 1: Single indicator

Example 2: Work dimension index

Source: Eurofound, 2017, p. 48

Work intensity index, by country, EU28

Job quality profiles (1) • Eurofound (2017)

– Based on data from the 6th European Working Conditions Survey (EWCS)

– Created quality of employment indices for skills and discretion, social environment, physical environment, work intensity, prospects and working time quality

– Used Latent Class Analysis (LCA) to identify jobs that were similar in terms of quality of employment.

– Results show that some jobs score highly on some quality of employment dimensions, but lower on others

Source: Eurofound, 2017, p. 128

Job quality profiles (2)

• Chen and Mehdi (2019)

– Used data from the Canadian General Social Survey (GSS): Canadians at work and home.

– Regrouped indicators to create indices similar to Eurofound

– Missing the “physical environment dimension”

• Identified four job quality profiles Source: Chen and Mehdi, 2019

Note: Partial table shown, model includes other covariates such as immigration status and province.

Data limitations: the Canadian case

• From 2017 to 2024, Statistics Canada did not have a survey covering all dimensions of quality of employment.

– The Labour Force Survey (LFS) includes some measures of QoE (e.g. wages, long hours), but does not provide a comprehensive picture

– In 2022, a program of LFS supplements was implemented to address data gaps

• Short sets of questions covering 1 or 2 topics are collected each month to create regular time series

– e.g. April 2022: scheduling and hours, November 2022: training, March 2024: career prospects

• Lacking a single data source to produce “job quality profiles”

Linking and aggregating

• No options to link the data sources at the micro level

– Different time periods, small samples

• What if there is a way to combine this information at an aggregate level?

– Occupational classifications

• Common to many surveys

• Occupational categories are based on the nature of tasks, duties and skills, and are likely to be associated with similar quality of employment outcomes.

Prior analysis using occupational characteristics

• Torrejón Pérez et al. (2023)

– Calculated average hourly earnings for industry by occupation cells at t=0.

– Jobs are classified into five quintiles of “quality” based on these average earnings

– Number of jobs in each cluster is calculated at t+n

Employment change, EU27, 2019Q4-2021Q4, by gender

Source: Torrejón Pérez et al., 2023, p.25

Data sources with occupational information relevant to Canada

• 2016 General Social Survey (GSS)

– Dedicated quality of employment modules covering most QoE dimensions

• Except physical environment

– Smaller sample size, older data

• LFS & LFS supplements

– Large sample size, high-quality sampling frame

– Recent estimates

– Proxy responses

• O*NET

– Based on data collected in the United States, provides occupation-level information on skill use, knowledge requirements, as well as work activities and the work environment

– Regularly updated based on surveys and expert knowledge

Experimental approach

• Use of the National Occupational Classification system (NOC).

– Classification specific to Canada (equivalent to ISCO)

• Selected most recent data, whenever possible

• Based on Eurofound and UNECE dimensions, identify closest corresponding measures to construct indices

Examples of measures

Physical environment Data source Vibrations O*NET - Wholebody vibration

Loud noise O*NET - Sounds, Noise Levels Are Distracting or Uncomfortable

Work intensity Data source Enough time to get job done GSS - How often can you complete your assigned workload

during your regular working hours?

GSS - How often do you consider your workload manageable?

Tight deadlines O*NET - Time pressure

Skills and discretion Data source

Training paid for or provided by the employer LFS supplement

Creating the database of occupational scores

• Converted occupational classifications to create a common denominator across data sources (e.g. SOC to NOC)

• Calculated an average for each occupational category

– Occupational scores do not always have a clear meaning – objective is to differentiate occupations on a particular dimension.

– E.g. permanent job =1, temporary job=0

• Occupational average A: 0.25 vs B: 0.50

• Occupational score indicates higher probability of having a permanent job in Occupation B.

• Level of detail limited by data quality

– E.g. Data quality too low for the most detailed occupational groups

Creating indices

• When possible, indicators were combined within each dataset.

• Otherwise, z-scores of occupational averages were combined at the aggregate level to create a composite score for each occupational category

– E.g. Prospects index = (Z-score (average job security) + Z-score(average career prospects) + Z-score(probability of job permanence))/3

• All indices expressed in Z-scores to ensure comparability.

Final indices

Physical environment

Work intensity

Working time

Social environment

Skills and discretion

Prospects

Benefits

Wages

Access to training

Performing the classification

• Latent Profile Analysis/Latent Class Analysis

– Supports inferences from a sample to a population

– Expresses results in terms of the probability that a case belongs to a specific class

– Models can fail to reach an appropriate level of fit

• K-means clustering

– No assumptions regarding the nature of the data

– Forces all cases to fall in a specific class

– A solution is always found (quality of fit is relative)

✓ K-means clustering more appropriate for this type of analysis

Selecting a solution

• K-means does not provide a single solution

– Range of options to evaluate quality

• Weighted sum of squares (minimize within cluster variation)

• “Silhouette method” (how well each case falls within a cluster)

• Gap statistic: compares within cluster variation with a distribution that has no clustering

• See UC Business Analytics (2017) for a discussion, and implementation in R.

Description of clusters

• 5 cluster solution identified as the best solution based on 2/3 of methods, and third-best option based on 1 method.

Source: Statistics Canada, Labour Force Survey, General Social Survey & O*NET, author’s calculations

Average standardized scores of quality of employment indices across occupational clusters Prospects Time Intensity Social Discretion Training Benefits Wage Physical

More challenging physical environment, with lower wages, good social -0.271 -0.423 0.140 0.256 -0.744 -0.809 -0.416 -0.543 -1.181

More flexible, better physical environment, below-average wages -0.039 0.413 -0.031 0.141 0.294 0.201 0.223 -0.193 0.510

High quality, higher intensity 0.745 1.020 -0.419 -0.024 1.057 0.811 1.009 1.373 0.949

Low quality, lower intensity -1.033 -0.569 0.563 -0.211 -0.964 -0.801 -1.643 -1.149 0.006

Good prospects, less flexible, more challenging physical 0.560 -1.224 -0.155 -0.420 -0.012 0.508 0.540 0.423 -0.806

Examples of occupations: Cluster 1

• More challenging physical environment, with lower wages, good social

• Ex:

– Cleaners

– Heavy equipment operators

– Labourers in processing, manufacturing and utilities

Examples of occupations: Cluster 2

• More flexible, better physical environment, below-average wages

• Ex:

– Managers in food service and accommodation

– Financial, insurance and related administrative support workers

– Insurance, real estate and financial sales occupations

– Contractors and supervisors, maintenance trades and heavy equipment and transport operators

– Supervisors, processing and manufacturing occupations

Examples of occupations: Cluster 3

• High quality, higher intensity

• Ex:

– Managers in financial and business services

– Managers in public administration

– Physical science professionals

– Computer and information systems professionals

Examples of occupations: Cluster 4

• Low-quality, lower intensity

• Ex:

– Tourism and amusement services occupations

– Cashiers

– Machine operators and related workers in textile, fabric, fur and leather products processing and manufacturing

Examples of occupations: Cluster 5

• Good prospects, less flexible, more challenging physical environment

• Ex:

– Professional occupations in nursing

– Secondary and elementary school teachers and educational counsellors

– Machinery and transportation equipment mechanics (except motor vehicle)

– Contractors and supervisors, mining, oil and gas

Mapping trends in occupational clusters • Occupational classification merged back to Canadian LFS to map trends in occupational clusters

over time and by demographics.

0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

35.0%

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023

Distribution of occupational clusters among employees, 2003 to 2023

High quality, higher intensity Flexible, better physical environment

More challenging physical environment, with lower wages, good social Good prospects, less flexible, more challenging physical

Low quality, lower intensity

Source: Statistics Canada Labour Force Survey, General Social Survey & O*NET, author’s calculations

Distribution by age and sex

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

Total 15 to 24 years 25 to 54 years 55 years and over Men Women

Distribution of occupational clusters by sex and major age group, 2023

High quality, higher intensity Flexible, better physical environment

More challenging physical environment, with lower wages, good social Good prospects, less flexible, more challenging physical

Low quality, lower intensity

Source: Statistics Canada Labour Force Survey, General Social Survey & O*NET, author’s calculations

Limitations

• Ignores all sampling error and does not take into account how well the sample or O*NET measures accurately reflect the mean occupational scores in the population.

• Averages mask variations within the occupational group

– Can be reduced somewhat by doing gender, age analysis using the clusters.

• Scales have an ambiguous meaning – reflects the situation of the “average worker in a given occupation”, relative to the “mean of average occupational scores”

• Mixes data sources, O*NET scores may not reflect situation in countries that are not the U.S.A.

Implications and possibles next steps

• Value of compiling quality of employment information by occupation

– Can inform how changes in occupational structure are associated with shifts in quality of employment

• Possibility of building cross-country database of quality of employment information by occupation

• Validate O*NET scores against survey results across other countries.

• Perform similar analysis in different countries.

Bibliography

• Chen, W. & T. Mehdi. (2019). Assessing Job Quality in Canada: A Multidimensional Approach. Analytical Studies Branch Research Paper Series. Statistics Canada, Ottawa.

• Eurofound. (2012). Trends in job quality in Europe. Publications Office of the European Union, Luxembourg.

• Eurofound. (2017). Sixth European Working Conditions Survey – Overview report (2017 update). Publications Office of the European Union, Luxembourg.

• National Center for O*NET Development. (2024). O*NET Online. Available at: https://www.onetonline.org/

• Torrejón Pérez, S., Hurley, J., Fernández-Macías, E. & E. Staffa. (2023). Employment shifts in Europe from 1997 to 2021: from job upgrading to polarisation, JRC Working Papers on Labour, Education and Technology 2023/05, European Commission, Seville, 2023, JRC132678.

• University of Cincinnati Business Analytics. (2017). “K-means Cluster Analysis”. UC Business Analytics R Programming Guide. Available at: https://uc-r.github.io/kmeans_clustering.

Stay connected!

StatsCAN app

Eh Sayers podcast

StatsCAN Plus

The Daily

Website

Surveys and statistical programs

Data service centres

My StatCan

Questions? Contact us: [email protected]

Statistics Canada— Your National Statistical Agency

Delivering insight through data for a better Canada

Integrating Immigration Administrative Data into the Canadian Census of Population

Kathryn Spence

Chief, Diversity and Sociocultural Statistics

Statistics Canada

Economic Commission for Europe: Conference of European Statisticians

Group of Experts on Migration Statistics

Geneva, Switzerland, 7−8 May 2024

Presentation Outline

1. Context of immigration in Canada and the Census of Population

2. Limitations with Census Questions 3. Limitations with administrative data 4. Methods used to integrate the administrative data 5. Results and lessons learned

2

Context of immigration in Canada

• Born in Canada • Born abroad to Canadian parent(s)

Canadian citizens by birth

• Economic immigrants • Immigrants sponsored by family • Refugees

Immigrants (includes Canadian citizens by

naturalization and permanent residents)

• Asylum claimants • Temporary foreign workers • International students

Non-permanent residents

3

Canadian Census of Population

• Key cross-sectional data source on the socio-economic outcomes of immigrants in Canada every 5 years

• The most recent Census was conducted on May 11, 2021.

• Immigration questions: immigrant status, year of immigration, place of birth, place of birth of parents and citizenship

• In 2021, Statistics Canada integrated administrative data to replace the questions on immigrant status and year of immigration

4

5

Overview of administrative data ▪ Immigration, Refugees and Citizenship Canada (IRCC) administrative data integrated into the Census

▪ Immigration landing file (1980 to present)

▪ Historical landing file (1952 to 1979)

▪ Asylum claims and temporary resident permits (1980 to present)

▪ Visitors file (2004 to present)

▪ Citizenship file (2004 to present)

▪ The main purpose of replacing the immigrant status and year of immigration question was to reduce response burden and improve the quality of the data.

▪ IRCC administrative data was also used to enhance 2016 Census

▪ Used to improve 2016 Census data processing (e.g. used as matching variables for imputation)

▪ Used to add admission category variables onto the 2016 Census

6

Limitation of immigrant status question: Immigrants who have been in Canada longer are less likely to think of themselves as immigrants

Percentage of census respondents linked to IRCC immigrant records who responded that they are not immigrants by years since admission 7.6% of 2016 Census

respondents who were linked to an administrative immigration record responded “No” to the immigrant status question on the Census. Most are corrected during processing.

7

Limitation of year of immigration question – Immigrants who have been in Canada longer are less likely to report their exact year of immigration

Match rate for census reported year of immigration by linked administrative value

Over 21% of immigrants in the 2016 Census had a year of immigration response which did not have an exact match to their linked administrative value. Most immigrants are within 5 years of accurately reporting.

Limitations of administrative data

• Administrative data are not updated to reflect deaths or emigration (The data need to be linked to estimate the current population of immigrants)

▪ No administrative records for immigrants admitted before 1952 ▪ Low linkage rates for immigrants admitted between 1961-1972

(see appendix) ▪ Lower coverage of non-permanent residents (e.g. Dependants of

temporary resident who don’t have their own permits, or other non-permanent residents)

▪ Missing links

8

9

Limitations of administrative data - Weaker linkage quality for years 1961-1972

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

90.0% 1

9 5

2

1 9

5 4

1 9

5 6

1 9

5 8

1 9

6 0

1 9

6 2

1 9

6 4

1 9

6 6

1 9

6 8

1 9

7 0

1 9

7 2

1 9

7 4

1 9

7 6

1 9

7 8

1 9

8 0

1 9

8 2

1 9

8 4

1 9

8 6

1 9

8 8

1 9

9 0

1 9

9 2

1 9

9 4

1 9

9 6

1 9

9 8

2 0

0 0

2 0

0 2

2 0

0 4

2 0

0 6

2 0

0 8

2 0

1 0

2 0

1 2

2 0

1 4

2 0

1 6

2 0

1 8

2 0

2 0

Percentage of administrative immigration records linked to the 2021 Census by administrative year of immigration

10

Method to replace Census questions on immigrant status and year of immigration ▪ Record linkage to administrative data identifies immigrants and non-permanent

residents ▪ Year of immigration for immigrants, permit or claim for NPRs

▪ Responses to the citizenship question are used to derive immigrant status and determine who is in scope for imputation for missing links

▪ Edit and imputation processes to address missing links or inconsistencies between linked values and other census responses (e.g. age).

Citizenship status Immigrant Non-immigrant

Canadian by birth Canadian by birth

Canadian by naturalization Canadian by naturalization

Not a Canadian citizen Permanent resident Non-permanent resident

Solutions to administrative data limitations

11

No available administrative data source

•Past census responses are used to supplement administrative data for years prior to 1952 (2016, 2011, 2006, or 2001)

Lower linkage rates •An algorithm is used to convert some records that are not linked to administrative data into immigrants for years 1961

– 1972

Lower coverage of non-permanent residents •Linking to permits which were no longer valid in 2021 •Linking to visitors' records

•Dependents of temporary resident permit holders •Parent and grandparent super-visa holders

Missing links •Responses to the citizenship question are used to determine which individuals require imputation (e.g. a person who

responded Canadian by naturalization who was not linked would have their year of immigration imputed)

12

Certification of the 2021 Census data: Comparability with other data sources

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2021 Census of Population Immigration, Refugees and Citizenship Canada (IRCC) Longitudinal Immigration Database (IMDB)

Year of immigration for the immigrant population who were admitted between 1980 and 2021, from different immigration data sources, Canada

13

Certification of the 2021 Census data: Comparability over time

-

50,000

100,000

150,000

200,000

250,000

300,000

1 9

8 0

1 9

8 1

1 9

8 2

1 9

8 3

1 9

8 4

1 9

8 5

1 9

8 6

1 9

8 7

1 9

8 8

1 9

8 9

1 9

9 0

1 9

9 1

1 9

9 2

1 9

9 3

1 9

9 4

1 9

9 5

1 9

9 6

1 9

9 7

1 9

9 8

1 9

9 9

2 0

0 0

2 0

0 1

2 0

0 2

2 0

0 3

2 0

0 4

2 0

0 5

2 0

0 6

2 0

0 7

2 0

0 8

2 0

0 9

2 0

1 0

2 0

1 1

2 0

1 2

2 0

1 3

2 0

1 4

2 0

1 5

2 0

1 6

2 0

1 7

2 0

1 8

2 0

1 9

2 0

2 0

2 0

2 1

2021 Census of Population 2016 Census of Population

Year of immigration for the immigrant population who were admitted between 1980 and 2021, Canada, Census of Population, 2016 and 2021

Notable shift for the period from 1989 to 1993. This is caused by respondents providing their year of arrival of their asylum claim (in the 2016 Census), as opposed to their year of immigration (in the 2021 Census).

14

Benefits and challenges of administrative data integration

Challenges ▪ Limitations of administrative data (missing links, missing information, acquisition of

administrative data sources) ▪ Historical comparability ▪ Loss of certification sources (linked data)

Benefits ▪ Improve data quality

▪ Improved precision of year of immigration since 1980 ▪ Better coherence with other data sources such as IRCC and IMDB ▪ Reduce response burden ▪ Addition of new immigration content

▪ Admission category and applicant type, year of arrival, pre-admission experience, province or territory of intended destination, non-permanent resident type

THANK YOU!

For more information, visit www.statcan.gc.ca

15

or contact

Kathryn Spence Chief, Diversity and Sociocultural Statistics Statistics Canada [email protected]

Measuring the emigration of immigrants in Canada using longitudinal administrative data

Julien Bérard-Chagnon

Statistics Canada

1) International migration is by far the main driver of demographic growth in Canada

• 2023: +3,2 % annual growth, strongest growth since 1957

• Record-high levels of immigration, both long-term and short-term

• Almost 1 in 4 Canadians is born abroad (2021 Census)

• Linked with various issues (eg: housing, aging, infrastructure, official languages)

• Increasingly strong interest by users for accurate and timely migration statistics

2

0

200000

400000

600000

800000

1000000

1200000

1400000

0.0%

0.5%

1.0%

1.5%

2.0%

2.5%

3.0%

3.5%

P e rs

o n s

D e m

o g ra

p h ic

g ro

w th

(% )

Natural increase Migratory increase Demographic growth

Source: Statistics Canada, Demographic Estimates Program.

Measuring immigrant emigration with accuracy and timeliness is challenging in Canada

• Immigrant emigration

– Impacts Immigration Levels Plan and immigration policies.

– Strong interest by users and stakeholders.

– Wasn’t studied much.

– Immigrants more likely to emigrate. Could lead to an increase of emigration in upcoming years given the increase of immigration.

• Measuring emigration

– Canada does not have a population register.

– Very few Canadian sources available.

– Emigration must be measured indirectly.

• Objective: develop an indirect measure of immigrant emigration using admin data

3

2) Longitudinal Immigration Database (IMDB)

• Comprehensive, detailed longitudinal file to shed more light on immigrant behaviour

• Linkage between immigration data, yearly tax returns and other admin data (eg: vital stats)

• Characteristics at admission (eg: admission category and country of birth) and on a yearly basis (eg: tax address, income)

4

1998 1999 2000 2001 2002 …

Immigration permit

Tax data Tax data Tax data Tax data …

Admission year

Measuring immigrant emigration indirectly using the IMDB

• Permanently stopped filing a yearly tax return.

• Not always a signal of emigration!

• Additional criteria:

– were 18+ years of age at the time of admission;

– have not died since landing;

– filed a tax return at least once after their admission;

– stopped filing a tax return for at least three consecutive years;

• Unless they put a departure date

– are assumed not to be non-tax filers.

• Studied cohorts: 1982 to 2017

5

Some examples!

6

1990 1991 1992 1993 1994 1995 1996 Status

Admitted in

Canada

In tax In tax Not in tax Not in tax Not in tax Not in tax Emigrant

Not in tax Not in tax In tax In tax In tax In tax Not emigrant

Not in tax Not in tax Not in tax Not in tax Not in tax Not in tax Not emigrant

In tax In tax Not in tax Not in tax Not in tax In tax Not emigrant

In tax In tax Died Not in tax Not in tax Not in tax Not emigrant

• 2 notable limitations of this indirect definition

– Only measures “permanent emigration”

– Not very timely

Our indirect definition provides similar levels of immigrant emigration than other sources

7

0

1

2

3

4

5

6

1991/1996 1996/2001 2001/2006 2006/2011 2011/2016

E m

ig ra

ti o n

ra te

s (%

)

Intercensal periods

Emigration rates (%) of recent immigrants

Reverse Record Check Longitudinal Immigration Database Sources: Statistics Canada, Longitudinal Immigration Database and Reverse Record Check.

3) Immigrant emigration peaks a few years after admission

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

P ro

b a b ili

ty o f

e m

ig ra

ti o n

(% )

Year since admission

8

Source: Statistics Canada, Longitudinal Immigration Database.

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

C u m

u la

ti v e p

ro b a b ili

ty (i n %

)

Year since admission

More than 15% immigrants emigrated within 20 years of admission

9

Source: Statistics Canada, Longitudinal Immigration Database.

Immigrant emigration fluctuates substantially by country of birth…

0

5

10

15

20

25

30

Sri Lanka Viet Nam Philippines Jamaica United Kingdom

Lebanon Taiwan Hong Kong France United States

C u m

u la

ti v e p

ro b a b ili

ty (i n %

)

Country of birth

Cumulative probability (in %) of immigrant emigration 10 years after admission

10

Source: Statistics Canada, Longitudinal Immigration Database.

Note: The study focused only on the 20 most frequently mentioned countries in the IMDB from 1982 to 2017. These 20 countries account for nearly 70% of all immigrants in this study.

…as well as by several other characteristics

0

5

10

15

20

25

30

Caregivers Entrepreneurs Investors Refugees No Yes Master’s degree Doctorate Study permit

Admission category Presence of child on tax data Level of education at admission Non-permanent resident status

before admission

C u m

u la

ti v e p

ro b a b ili

ty (i n %

)

Characteristics

Cumulative probability (in %) of immigrant emigration 10 years after admission

11

Source: Statistics Canada, Longitudinal Immigration Database.

4) Conclusion

• Immigration is a key demographic dynamic of Canada.

• Increased interest for immigrant emigration.

• Few Canadian data sources inform on emigration.

• Developed an indirect measure of immigrant emigration using admin data

– Some limitations, notably related to timeliness

– Results are consistent with those from other sources

• Some immigrant characteristics correlated with emigration

– More results, including from regression models, available in the full paper

• Working to acquire Entry/Exit Program data (border data)

– Long process

12

Some lessons learned

• Relevance: users needs evolve quickly. Challenging to stay relevant.

• Timeliness: increasingly important => trade-offs needed.

• More and more demographic models are required.

• Accuracy: indirect method -> more uncertainty (and potentially critics) -> more evaluations needed and made available.

• International collaboration

– North America Collaborative Agreement (Canada, the U.S. and Mexico) on Migration Statistics

13

Thank you for your attention! ☺ Merci pour votre attention ! ☺

Julien Bérard-Chagnon

Chef, Développement et Évaluation du Programme des estimations démographiques — Centre de démographie, Secteur de la statistique sociale, de la santé et du travail Statistique Canada / Gouvernement du Canada [email protected]

Chief, Demographic Estimates Program Development and Evaluation — Centre for Demography, Social, Health and Labour Statistics Field Statistics Canada / Government of Canada [email protected]

14

*Prepared by Kathryn Spence and Athanase Barayandema NOTE: The designations employed in this document do not imply the expression of any opinion whatsoever on the part of the Secretariat of the United Nations concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.

Economic Commission for Europe Conference of European Statisticians Group of Experts on Migration Statistics Geneva, Switzerland, 7−8 May 2024 Item 6 of the provisional agenda Improvements in use of administrative data for migration statistics

Replacing immigration question with administrative data in the Canadian Census of Population

Note by Statistics Canada

Abstract

The Canadian Census of Population is a key data source on the socio-economic outcomes of immigrants in Canada. Census questions related to immigration include immigrant status, year of immigration, citizenship, place of birth, and place of birth of parents. As a strategy to reduce burden and improve data quality, Statistics Canada replaced the immigrant status and year of immigration variables on the 2021 Census questionnaire with administrative records. Immigration is a process administered by Immigration, Refugees and Citizenship Canada (IRCC), administrative data is collected for temporary (non-permanent) and permanent residents in Canada. Having successfully replaced income questions with administrative data and integrating new immigration variables in 2016 there was a strong precedent for this approach. This paper will outline the results from using administrative data to replace immigration question on the 2021 Census of Population providing an overview of the benefits and challenges with administrative data integration.

Working paper 12

Distr.: General 29 April 2024 English

Working paper 12

2

I. Introduction

1. The Census of Population is a key data source on the socio-economic outcomes of immigrants in Canada. Census questions related to immigration include immigrant status, year of immigration, citizenship, place of birth, and place of birth of parents. The 2016 Census also included variables related to admission category (e.g., refugees, economic immigrants, etc.) using administrative data (McLeish 2017).

2. As a strategy to reduce respondent burden and improve data quality, Statistics Canada integrated administrative data from Immigration Refugees and Citizenship Canada (IRCC) to replace the questions on immigrant status and year of immigration building on the success of integrating administrative data in the 2016 Census (Statistics Canada 2017b; McLeish 2017).

3. This paper will present the results of integrating the administrative data to replace the immigration questions, assessing the quality of the administrative data, and examining how replacing the questions could affect historical comparability. This paper will also assess the quality of the existing questions to understand the differences in data quality between asking questions and using administrative values outlining the benefits and challenges.

II. Limitations on questions and administrative data

A. Limitations on questions

4. Data on immigrant status and year of immigration have been collected on the census questionnaire since 1901. As has been documented in previous papers (McLeish 2017; McLeish 2014), questions are not always answered by respondents, and answers provided are not always precise. For immigration and citizenship questions specifically, there are some common issues that have been observed during the certification of the data quality of census results:

a. the longer immigrants have been in Canada, the less likely they are to respond affirmatively to the immigrant status question,

b. respondents do not always provide the precise response to the year of immigration question, and accuracy appears to decrease the longer the respondent has been in Canada,

c. certain respondents appear to provide their year of arrival instead of their year of immigration, and

d. Canadian citizens by birth born abroad, that is those who are entitled to Canadian citizenship because of their parentage, appear to sometimes respond that they are a Canadian citizen by naturalization.

5. During data processing, edits are applied to ensure consistency between responses and donor imputation methods are used to address item non-response (Crowe 2017; Guertin 2014).

6. Figure 1 illustrates issue 4a) where 7.6% of 2016 Census respondents who were linked to administrative immigration records (for immigrants admitted since 1980) responded “No” to the immigrant status question. This proportion increased the longer the linked respondents have been in

Working paper 12

3

Canada. Most of these cases were resolved during edit and imputation. However, since the immigrant status questions serves as a filter for the year of immigration question, this issue results in a higher imputation rate for the year of immigration1.

Figure 1

Percentage of census respondents linked to IRCC immigrant records who responded that they are not immigrants by years since admission.

Source: Statistics Canada, 2016 and 2006 Censuses linked to IRCC administrative immigrant data since 1980 and 2011 National Household Survey (NHS) linked to IRCC administrative immigrant data since 1980

7. Figure 2 illustrates issue 4b), where over 21% of 2016 Census respondents linked to immigration records since 1980 provided a year of immigration response which did not match their administrative value. This proportion increases the longer the respondents have been in Canada. However, most of these differences in value are less than 5 years.

8. Issue 4c) is also shown in Figure 2, where there is a notable drop in the congruence between administrative and census year of immigration for the years 1991 to 1993. This is explained by a large number of people who claimed asylum in Canada in 1989 and 1990. While their official years of immigration are between 1991 and 1993, their responses reflect their year of arrival or asylum claim.

1 Imputation rates for the 2016 Census citizenship and immigration variables: https://www12.statcan.gc.ca/census- recensement/2016/ref/guides/007/98-500-x2016007-eng.cfm#tbl1

Working paper 12

4

Figure 2 Match rate for census reported year of immigration by administrative value.

Source: Statistics Canada, 2016 and 2006 Censuses linked to IRCC administrative immigrant data since 1980 and 2011 National Household Survey (NHS) linked to IRCC administrative immigrant data since 1980

B. Limitations of administrative data

9. IRCC administrative data currently available to Statistics Canada includes;

a. Detailed immigration records from 1980 to present, b. Limited immigration records from 1952 to 1979, c. Non-permanent resident permit records from 1980 to present, d. Citizenship records from 2004 to present, and e. Visitor records from 2004 to present.

10. While IRCC data reflect an administrative census of all incoming immigrants and non-permanent residents to Canada, they are not updated to capture deaths or outmigration. Therefore, they cannot be used in isolation to estimate the current population of immigrants living in Canada.

11. For the purposes of replacing the two immigration questions on the Census of Population, the coverage of IRCC data introduces three limitations.

a. Firstly, there are no immigration records available prior to 1952. While this is a decreasing population, it still represented 111,000 immigrants (or 1.5% of all immigrants) living in Canada according to the 2016 Census.

b. Secondly, non-permanent resident records contain information on permit holders only; any accompanying family members who do not hold permits are not covered.

c. Thirdly, another limitation rests with the quality of the immigration records prior to 1980. Overall, there is less information included on these records that can be used for record linkage purposes. Records from 1961 to 1972 contain an incomplete date of birth which leads to lower linkage rates.

Working paper 12

5

12. Figure 3 demonstrates the percentage of the administrative immigration records linked to 2016 Census respondents by administrative year of immigration. In general, the linkage rates are lower for older cohorts of immigrants, as these individuals are more likely to be out-of-scope for the 2016 Census (e.g., dead or no longer residing in Canada). However, there is a notable drop in the linkage rate between 1961 and 1972.

Figure 3 Percentage of administrative immigration records linked to the 2016 Census by administrative year of immigration.

Source: Statistics Canada, 2016 Census linked to IRCC administrative immigration data since 1952

III. Methods used for replacement, solutions, and challenges

C. Data linkage and methods used for replacement

13. A record linkage between the census or survey respondents and IRCC administrative data took place immediately after collection, and before processing. The probabilistic method was used to integrate IRCC’s immigration data to the 2021 Census data. The global linkage rate for immigration records was 97.3%. The linkage rate for permanent residents was 99.4% and 92.4% for non- permanent residents. Respondents to the census who are linked to an administrative record are either an immigrant if linked to the landing file or a non-permanent resident if linked to a temporary resident file (work or study permit) or asylum claim.

14. The administrative data and the census question on Canadian citizenship were used to derive immigrant status variable to determine whether the person is a non–immigrant, an immigrant or a non–permanent resident as shown in Table 1. The responses to the citizenship question were used to determine which individuals required imputation to the immigrant status variable due to missing links. A person who is Canadian citizen by birth cannot be an immigrant, a person who is Canadian

Working paper 12

6

citizen by naturalization are immigrants and a person who is not a Canadian citizen is either an immigrant (permanent resident) or non-permanent resident.

Table 1 Immigrant status and citizenship, Canada

Ci�zenship status Immigrant Non-immigrant Canadian by birth Canadian by birth Canadian by naturaliza�on Canadian by naturaliza�on Not a Canadian ci�zen Permanent resident Non-permanent resident

18. In general, the same edit and imputation methods used in the past were still applied using administrative values in lieu of responses to questions. In processing, those who were not linked and who gave a citizenship response of Canadian citizen by birth were considered to be non–immigrants and required no donor imputation for immigration status. Canadian citizens by naturalization and non-Canadian citizens were considered in-scope and required imputation if no administrative record was available.

D. Solutions to administrative data limitations

19. To resolve the absence of administrative records prior to 1952, the administrative files were supplemented to include records for past census respondents who responded with a year of immigration prior to 1952. These values would include the same response errors described above and would only be available for a sample of the total population who reported having immigrated prior to 1952. However, they are the only records available for this subpopulation.

20. Concerning the low linkage quality for years 1961 to 1972, an approach of using the multipliers directly to assign a year of immigration in this period for a specified number of unlinked records was used. The multipliers are the ratios of predicted probabilities of being linked if the date of birth was not missing. Different counts of unlinked persons, who did not identify themselves as Canadian citizen by birth on the Census, were assigned to be converted to each specific year for different age / place of birth groupings (cells). This was done before any Census donor imputation was performed. This approach has the great advantage of implementing the spirit of the multipliers to alleviate the linked year of immigration distribution. Another advantage of this approach is that it would be immediately and clearly generalizable to future Censuses.

21. Additional administrative data from IRCC such as visitor records, were used to supplement the temporary resident records for family members of work and study permit holders since some family members such as children, parents and grandparents may not have their own permit.

Working paper 12

7

E. Key Challenges

22. The principal challenges associated with replacing the census immigration questions with administrative values pertain to the limitations of the administrative data itself. The absence of records prior to 1952 require the supplement of past census responses. However, past census responses will not cover this entire subpopulation (the immigration questions have only been asked of a sample of census respondents), and it could be underestimated, as a result.

23. The limited linked information on the immigration records from 1952-1979 (especially from 1961- 1972) could lead to an underestimation of this subpopulation as well. Since the final results (post edit and imputation) are anchored by responses to the citizenship question, any underestimation of certain periods of immigration may lead to an overestimation of others (especially recent immigrants, whose records are of higher quality for the purposes of linkage).

24. Replacing the questions with administrative data would affect historical comparability. In particular, shifts in the distribution of year of immigration may occur as perceived year of immigration (e.g., year of arrival) is replaced with actual year of immigration.

IV. 2021 Census Results

F. Data Quality

25. A number of quality indicators such as the non-response rate and the imputation rate per question were produced and analyzed during the 2021 Census of Population data quality assessment.

26. The non-response rate for the immigration variables largely measures the proportion of immigrants and non-permanent residents for whom an administrative value was not available because the record was not linked to an administrative record. It also includes some inconsistencies between the census responses to the other questions and the linked administrative values. For example, immigrants may have reported a birth year on the census questionnaire that was before their year of immigration.

27. At the national level, the immigrant status variable had an imputation rate of 2.2% and a non- response rate of 12.1%. The difference between these two rates is because non-responses were resolved early during data processing because a single resolution was possible based on the answers provided to other questions, such as the citizenship question, making imputation unnecessary. For more details please refer to Place of Birth, Generation Status, Citizenship and Immigration Reference Guide, Census of Population, 2021.

G. Comparability Over Time

28. The methods used to collect data for immigrant status and year of immigration are different (questionnaire response vs. administrative value), therefore historical comparability with previous censuses will be affected. In particular, there are shifts in the distribution of year of immigration, as perceived year of immigration (e.g., year of arrival) is replaced with the administrative year of

Working paper 12

8

immigration. A comparison of the number of immigrants by year of immigration between the 2021 Census and the 2016 Census (Figure 4) shows these shifts.

29. For the period from 1989 to 1993, Figure 4 shows that 2016 Census data (responses) were more heavily distributed in 1989 and 1990, while the 2021 Census data (from administrative data) were more heavily distributed from 1991 to 1993. This is caused by respondents providing their year of arrival or asylum claim (in the 2016 Census), as opposed to their year of immigration (in the 2021 Census).

Figure 4 Year of immigration for the immigrant population who were admitted between 1980 and 2021, Canada, Census of Population, 2016 and 2021

Source: Statistics Canada, Census of Population, 2016 and 2021

H. Comparability with other data sources

30. Many statistical sources provide information on immigration to Canada, covering different reference periods and different subpopulations, thereby meeting different informational needs which needs to be considered when comparing data sources. The 2021 Census of Population estimates the immigrant population living in private households in Canada on Census Day, May 11, 2021.

31. In comparison, the administrative data from IRCC provide information on the total number of immigrants admitted to Canada each year as permanent residents. The IRCC data cannot be used to estimate the population of immigrants living in Canada as they do not account for any outflows, such as deaths or emigration. Since they provide the total number of all those who have ever been permanent residents in Canada, the IRCC administrative data counts are higher than the census estimates of immigrants living in Canada at a given point in time.

0

50,000

100,000

150,000

200,000

250,000

300,000

19 80

19 81

19 82

19 83

19 84

19 85

19 86

19 87

19 88

19 89

19 90

19 91

19 92

19 93

19 94

19 95

19 96

19 97

19 98

19 99

20 00

20 01

20 02

20 03

20 04

20 05

20 06

20 07

20 08

20 09

20 10

20 11

20 12

20 13

20 14

20 15

20 16

20 17

20 18

20 19

20 20

20 21

2021 Census of Population 2016 Census of Population

Working paper 12

9

32. To some extent, the counts from the IMDB consider deaths and emigration. The 2019 IMDB combines linked administrative data from IRCC with tax data files. The counts presented in Figure 5 are for those immigrants who filed tax returns in 2019. Since not all immigrants living in Canada would have filed tax returns, because of age or other factors, counts from the IMDB are expected to be lower than the estimates from the 2021 Census.

33. The number of immigrants by year of immigration in the 2021 Census data, the IRCC data and

the IMDB data (Figure 5) shows similar trends in all three sources. In the earlier years of immigration, the census estimates are closer to the IMDB counts, as immigrants who landed between 1980 and 2005 would most likely be taxfilers still living in Canada at the time of the census. As the year of immigration moves towards 2021, the census estimates start to move closer to the counts from IRCC, as the number of non-taxfilers, such as children, increases, while the number of deaths and emigrants would be lower.

Figure 5

Year of immigration for the immigrant population who were admitted between 1980 and 2021, from different immigration data sources, Canada

Source: Statistics Canada, Census of Population, 2021, and IMDB, tax year 2019; and IRCC

V. Conclusion

34. The replacement of census questions on immigrant status and year of immigration with administrative values in the 2021 Census of Population was a success. Considering the good quality of administrative immigration data (particularly from 1980 to present) and the limitations of Census questions on immigration, the benefits of the replacement are data quality improvement, reduction in response burden, and the integration of additional variables from the administrative data such as non-

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

19 80

19 81

19 82

19 83

19 84

19 85

19 86

19 87

19 88

19 89

19 90

19 91

19 92

19 93

19 94

19 95

19 96

19 97

19 98

19 99

20 00

20 01

20 02

20 03

20 04

20 05

20 06

20 07

20 08

20 09

20 10

20 11

20 12

20 13

20 14

20 15

20 16

20 17

20 18

20 19

20 20

20 21

2021 Census of Population Immigration, Refugees and Citizenship Canada (IRCC)

Longitudinal Immigration Database (IMDB)

Working paper 12

10

permanent resident type (asylum claimant, work or study permit holder), pre-admission experience in Canada, province or territory of intended destination and year of arrival.

35. Another advantage of the replacement of census questions on immigrant status and year of immigration with administrative values is the impact beyond the Census of Population. Other household surveys may also follow in the same direction. Similarly, it could lead to a reduction in response burden on those surveys, and open possibilities of integrating the additional immigration content available on administrative files (such as admission category).

36. Replacing immigration questions by administrative data has some disadvantages. First, because the method used to collect the data for immigrant status and year of immigration is different, the historical comparability will be affected for these questions. Specifically, the year of immigration will be more reflective of administrative value versus perceived value which reflects the true value of the concept being measured. Secondly, the replacement will imply a loss of the linked data as certification data source. However, Statistics Canada has the benefit of a wealth of available sources of data, such as Immigration Landing File, the Longitudinal Immigration Database (IMDB), and other social surveys such as Labour Force Survey and the General Social Survey Program that are used for certification.

References

Biernot, P. 2017. “External linkage between the Immigration File (1952-2016) and the 2016 Census Response Database.” Unpublished report. Ottawa: Statistics Canada.

Brennan, J. 2011. “CIC Landing File to Census 2006 Linkage.” Unpublished report. Ottawa: Statistics Canada.

Brennan, J. 2013. “CIC Landing File to Census 2011/NHS Linkage.” Unpublished report. Ottawa: Statistics Canada.

Crowe, S. and Janes, D. 2017. “Edit and Imputation Report: Ethnocultural Process.” Unpublished report. Ottawa: Statistics Canada.

Guertin, L., Bureau, M., and Morel, J. 2014. “Editing the 2011 Census data with CANCEIS and options considered for 2016.” 2014 Work Session on Statistical Data Editing. Conference of European Statisticians. United Nations Economic Commission for Europe.

McLeish, S. 2014. “Using administrative data to evaluate data quality: Immigrants in the 2011 National Household Survey”, Proceedings of Statistics Canada Symposium 2013, Catalogue no. 11-522-X. Ottawa: Statistics Canada.

McLeish, S. 2017 “2016 Census of Population of Canada: Integration of immigration administrative data” 2017 Work Session on Migration Statistics. Conference of European Statisticians. United Nations Economic Commission for Europe.

Statistics Canada. 2017. “Place of Birth, Generation Status, Citizenship and Immigration Reference Guide.” 2021 Census of Population. Catalogue no. 98-500-X2016007. Ottawa: Statistics Canada.

*Prepared by Julien Bérard-Chagnon, Chief of the Development and Evaluation Section, Centre for Demography, Statistics Canada. This note is a summary of a paper entitled Emigration of Immigrants: Results from the Longitudinal Immigration Database published on Statistics Canada website on February 2, 2024. NOTE: The designations employed in this document do not imply the expression of any opinion whatsoever on the part of the Secretariat of the United Nations concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.

Economic Commission for Europe Conference of European Statisticians Group of Experts on Migration Statistics Geneva, Switzerland, 7−8 May 2024 Item 5 of the provisional agenda Measuring emigration

Measuring the emigration of immigrants in Canada using longitudinal administrative data

Note by Statistics Canada*

Abstract

Canada has a rich history of immigration. In 2021, almost one-quarter of the Canadian population was born abroad. Moreover, Canada’s international migration, both permanent and temporary, is now reaching record-high levels and accounts for nearly all the country’s population growth.

However, the fact that Canada is often seen as a country of immigration tends to obscure the opposite trend: emigration. Measuring emigration accurately is a challenge in Canada because the country does not have a population register. Consequently, the study of emigration must rely on indirect sources and definitions.

Some studies report that immigrants are more likely to emigrate from Canada than the rest of the population. Given the marked increase in immigration, it is more and more relevant to develop a robust measure of the emigration of immigrants to evaluate and inform immigration policies.

This note shows how a linkage between immigration and tax data, called the Longitudinal Immigration Database IMDB), allowed the development of an indirect method to measure the emigration of immigrants. The key result from the study is that over 15% of immigrants who were admitted between 1982 and 2017 emigrated within 20 years of landing.

Working paper 6

Distr.: General 29 April 2024 English

Working paper 6

2

I. Introduction

1. Canada has a rich history of immigration. In 2021, almost one-quarter of the Canadian population was born abroad. Moreover, Canada’s international migration, both permanent and temporary, is now reaching record-high levels and accounts for nearly all the country’s population growth. In 2023, Canada posted its strongest demographic growth (+3.2%) since 1957 following the arrival of more than 1.2 million permanent and temporary immigrants.

2. However, the fact that Canada is often seen as a country of immigration tends to obscure the opposite trend: emigration. Measuring emigration accurately is a challenge in Canada because the country does not have a population register. Consequently, the study of emigration must rely on indirect sources and definitions (Bérard-Chagnon 2018).

3. Some studies report that immigrants are more likely to emigrate from Canada than the rest of the population (Finnie 2006). Moreover, while some characteristics associated with the emigration of immigrants are known, the underlying mechanisms are less clear. New demographic dynamics are also emerging, such as two-step migration and the increasing diversity of immigrants’ countries of origin, that could shape future emigration patterns. As a result, it is more and more relevant to develop a robust measure of the emigration of immigrants to answer emerging data users’ needs, to maintain accurate demographic statistics as well as to evaluate and inform immigration policies, such as Canadian immigration targets.

4. This note shows how Statistics Canada used a linkage between immigration and tax data, called the Longitudinal Immigration Database (IMDB), to develop an indirect method to measure the emigration of immigrants for several characteristics in the absence of a population register. It is a summary of a paper entitled Emigration of Immigrants: Results from the Longitudinal Immigration Database published on Statistics Canada website on February 2, 2024. Section 2 and 3 respectively introduce the IMDB and the indirect definition of emigration used for this paper. Then, section 4 presents the results of some evaluations to assess the accuracy of this definition. Lastly, section 5 shows selected results of the study.

II. The Longitudinal Immigration Database

5. The IMDB is the result of an ongoing collaboration between Statistics Canada and Canada’s immigration department (Immigration, Refugees and Citizenship Canada [IRCC]) to create a yearly. comprehensive and detailed longitudinal file to shed more light on immigrant behaviour.

6. The IMDB essentially combines administrative data from IRCC on the number of immigrant admissions and temporary resident permits issued with annual tax data from the Canada Revenue Agency (CRA). Immigrant admission data notably include the admission date and various immigrant characteristics, such as the level of education at admission, the country of origin and the admission category, whereas tax data provides data on the province/territory of residence and on marital status.

7. The immigration data date back to 1980 while the linked tax return data start in 1982. The IMDB is updated annually through record linkages to add new information from IRCC and the CRA. The 2020 version of the IMDB was used in this study.

Working paper 6

3

III. Measuring immigrant emigration using the IMDB

8. Emigration is a very difficult demographic event to measure accurately in Canada. Canadians who emigrate are not required to report their departure, making it very difficult to track them accurately in different Canadian data sources. Moreover, immigrants who emigrate from Canada don’t appear in many international databases since they use the country of birth to compute Canadian emigration.

9. In this study, immigrants are identified as emigrants if they have permanently stopped filing a tax return. The majority of Canadians have to fill tax returns every year so the coverage of tax data is high for adults.

10. Immigrants who do not file a tax return are not necessarily emigrants. Additional criteria were developed to better identify emigrants among these immigrants. Immigrants who have permanently stopped filing a tax return are considered emigrants if they:

a. were 18 years of age or older at the time of admission;

b. have not died since landing;

c. filed a tax return at least once after their admission;

d. stopped filing a tax return for at least three consecutive years;

e. are assumed not to be non-tax filers.

11. The reasons behind these criteria are the following.

a. The age criterion was set because of the very low coverage of children in tax returns.

b. Dates of death were derived from tax and vital statistics data. The estate of a deceased tax filer must register the date of death when filing the deceased person’s last tax return.

c. This study is limited to immigrants who completed at least one tax return after admission and who were matched by the IMDB team. This decision was made to ensure that immigrants had indeed settled in Canada before leaving and to avoid classifying immigrants who could not be linked by the IMDB as emigrants.

d. The three-year criterion was chosen to minimize the risk that the immigrant was either a late tax filer or a non-tax filer, rather than an emigrant. In those cases, the immigrant is likely to file a tax return in the following years.

e. Some of these tax filers permanently stop filing a tax return while still residing in Canada. In this study, two groups were identified as especially likely to be in that situation: immigrant women of child-bearing age (19 to 45) and immigrants who are not eligible to receive Canada Old Age Security pension (65 years of age or older).

12. While the definition chosen for this study makes it possible to measure immigrant emigration indirectly, two limitations cannot be overlooked.

13. First, the definition used in this study only captures “permanent emigration” (those who left as of the 2020 version of the IMDB) and omits immigrants who left and subsequently returned to Canada.

14. Second, the 2018 and 2019 immigration cohorts were excluded from the study because they arrived too recently in the country to properly measure emigration according to the definition used.

Working paper 6

4

IV. Evaluating the definition of immigrant emigration

15. This section presents one of the evaluations made to assess the accuracy of the definition used in this analysis.

16. The results obtained using the IMDB and the definition chosen were compared with those other sources. The evaluation showed in this note is done using the Reverse Record Check (RRC), the postcensal survey that estimates census undercoverage. The RRC can be used to estimate the level of emigration of recent immigrants admitted to the country (within the last five years before census as Canada conducts a census every five years). The following chart compares emigration rates from the IMDB and from the RRC for several censuses.

Figure 1

Emigration rates (in %) of recent immigrants from the Reverse Record Check and the Longitudinal Immigration Database, 1996, 2001, 2006, 2011 and 2016

Sources: Statistics Canada, Longitudinal Immigration Database and Reverse Record Check.

17. Both sources suggest that the emigration rates of recent immigrants range from 1.7% to 5% from 1991/1996 to 2011/2016. The main differences observed between the two sources concern the 1996/2001 and 2006/2011 periods. In the first case, the IMDB reports a higher emigration rate than the RRC by 1.6 percentage points. Conversely, in 2006-2011, the IMDB reports a lower emigration rate than the RRC by about two percentage points.

18. Similar conclusions can be drawn from the results of additional comparisons, suggesting that the criterion developed in this document produces emigration levels that are reasonably accurate.

0

1

2

3

4

5

6

1991/1996 1996/2001 2001/2006 2006/2011 2011/2016

Em ig

ra tio

n ra

te (i

n %

)

Intercensal periods

Reverse Record Check Longitudinal Immigration Database

Working paper 6

5

V. Selected results of immigrant emigration

19. This section presents selected results from this study.

A. Emigration probabilities

20. The following chart illustrates the probability of emigrating in each year after landing. Note that these results were obtained using life tables in order to take into account that some immigrants have died and thus are no longer at risk of emigrating.

Figure 2

Annual probability (in %) of the emigration of immigrants, 1982 to 2017

Source: Statistics Canada, Longitudinal Immigration Database.

21. The main finding emerging from the analysis of the probability of emigrating is that immigrants are much more likely to emigrate within the first few years after admission. The annual probability of emigrating reaches the highest level from three to seven years after admission, and peaks at almost 1.4% in the fourth and fifth years after admission. Thereafter, the annual probability of emigrating falls and holds steady at 0.6% to 0.7%. These findings echo those of other studies on this topic, i.e., that recent immigrants are more likely to emigrate from Canada than are immigrants from older cohorts.

22. We can obtain the cumulative emigration probabilities by summing the annual probabilities just presented. Five years after admission, just over 5% of immigrants have emigrated. This probability increases to more than 10% a decade after admission. Slightly more than one in five immigrants have emigrated 25 years after their admission to Canada based on the IMDB and this study’s definition. These results indicate that while emigration of immigrants is quite low annually, it becomes a relatively significant phenomenon over the long term.

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

Pr ob

ab ili

ty o

f e m

ig ra

tio n

(in %

)

Year since admission

Working paper 6

6

B. Emigration probabilities for key immigrant characteristics

23. The following table displays the cumulative emigration probabilities ten years after admission for selected immigrant characteristics.

Table 1

Cumulative probability (in %) of the emigration of immigrants ten years after admission for selected characteristics and categories, 1982 to 2017

Characteristic Cumulative probability (in %)

Total 10.5 Country of birth Taiwan 20.1 Jamaica 4.6 France 24.0 Lebanon 18.3 Viet Nam 4.1 United States 27.5 Sri Lanka 3.5 United Kingdom 16.1 Hong Kong 20.1 Philippines 4.5 Admission category Caregivers 3.7 Entrepreneurs 17.0 Investors 21.9 Refugees 4.9 Presence of child on tax data No 27.9 Yes 6.2 Level of education at admission Master’s degree 16.4 Doctorate 20.5 Non-permanent resident status before admission Study permit 20.9

Source: Statistics Canada, Longitudinal Immigration Database.

24. The propensity to emigrate varies widely by country of birth, admission category, presence of child on tax data, level of education at admission and non-permanent resident status before admission.

25. More than 20% of immigrants whose country of birth is Taiwan, the United States, France, or Hong Kong emigrated within 10 years after admission. Two key dynamics could be at work among immigrants from these countries. First, immigrants from countries with a standard of living similar to Canada’s are more likely to emigrate. These countries may continue to hold a strong attraction for their nationals even years after their arrival in Canada. The geographic proximity of Canada and the United States may also encourage US- born immigrants to emigrate. Second, immigrants from locations like Hong Kong or

Working paper 6

7

Lebanon may have been admitted to Canada amid political instability and may view their settlement and departure as two stages of a more complex migration strategy.

26. In contrast, immigrants from other countries show significantly lower propensities to emigrate. ten years after being admitted to Canada, less than 5% of immigrants born in the Philippines, Sri Lanka, Viet Nam or Jamaica had left the country. The circumstances in which immigrants born in these countries are admitted to Canada may explain these results. They tend to belong to admission categories with lower emigration rates, such as the refugees and caregivers categories.

27. More than 20% of immigrants admitted as investors and 17% admitted as entrepreneurs emigrated within ten years of admission. These categories include wealthy immigrants who tend to be highly mobile and who may intend to leave Canada even at the time of their admission.

28. On the other hand, less than 5% of the immigrants admitted as refugees and caregivers emigrated within ten years of admission. Refugees are admitted to Canada with reasons to fear returning to their country of origin and are more likely to stay in the country. Immigrants admitted as caregivers are predominantly women from the Philippines. The Philippines encourages some of its workers to emigrate so that they can send their foreign earnings back to their families in the Philippines. These immigrants may be less likely to emigrate because one of the admission criteria for this category is to have a job in Canada and because immigration in these categories is often seen as a gateway to Canada for immigrants who are not eligible in other categories (Bonifacio 2008).

29. The presence of children in the tax family relates closely to emigration. Around 28% of immigrants who never had children in their tax family emigrated within one decade of admission compared with less than 7% of those who already had children in their tax family.

30. The propensity to emigrate follows a clear gradient based on level of education at admission. Just over 20% of immigrants who had a doctorate at the time of admission and slightly more than 16% of those with a master’s degree emigrated within ten years of admission.

31. Some immigrants arrive in Canada as non-permanent residents as part of a two-step migration approach. A little bit more than 20% of immigrants who held a study permit prior to being admitted as immigrants left Canada within ten years of admission. Because they were students while in Canada as an non-permanent residents, many of these immigrants may have had fewer ties to the country, such as family in Canada or stable employment. They may also have originally planned to return to their home country after their studies or to continue their studies in another country.

VI. Conclusion

32. Immigration is an increasingly important facet of Canada’s migration dynamics. As the country’s population growth becomes increasingly reliant on international migration and as policies are developed to attract and integrate immigrants, the emigration of immigrants is becoming an issue of interest for users and policy makers. The purpose of this analysis was to highlight certain factors associated with this phenomenon using IMDB data. Since Canada does not have a population register, emigration was examined indirectly through immigrants who permanently stop filing a tax return.

33. Based on the criterion developed for this study, the emigration of immigrants is a numerically significant phenomenon. A little bit more than 10% of immigrants admitted

Working paper 6

8

between 1982 and 2017 emigrated within ten years of admission. The propensity to emigrate varies widely by country of birth, admission category, presence of child on tax data, level of education at admission and non-permanent resident status before admission.

A. Refining the measurement of emigration in Canada

34. Migration is a very complex phenomenon. Technological advances in communication and transportation, as well as globalization, foster international migration and the development of economic and social ties in more than one country. Some immigrants may not emigrate permanently, but rather as part of a more complex migration strategy. Since the emigration criterion used in this study was obtained by examining immigrants who permanently stopped filing tax returns, return migration and circular migration were not analyzed. These forms of migration could increase in the future not only for immigrants but also for people born in Canada.

35. Border data could shed new light on the emigration trajectories of Canadians. Data from the Entry/Exit Program could potentially be leveraged in this regard similarly to what is done in New Zealand and Australia, two countries also without population registers.

36. Finally, the accurate measurement of a complex phenomenon such as emigration must rely on clear concepts and data appropriate to its measurement. Canadian administrative data do not always measure emigration in the same way as other key demographic databases such as censuses. A better understanding of how tax data (and admin data in general) reflect different facets of emigration is needed to improve measurement.

References Bérard-Chagnon, Julien, 2018, “Measuring Emigration in Canada: Review of Available Data Sources and Methods”, Demographic Documents. Statistics Canada.

Bonifacio, Glenda Lynna Anne Tibe, 2008, “I Care for You, Who Cares for Me? Transitional Services of Filipino Live-in Caregivers in Canada”, Asian Women.

Finnie, Ross, 2006, “International Mobility: Patterns of Exit and Return of Canadians, 1982 to 2003”, Analytical Studies Branch Research Paper Series. Statistics Canada.

1

UNITED NATIONS ECONOMIC COMMISSION FOR EUROPE CONFERENCE OF EUROPEAN STATISTICIANS Expert Meeting on Statistical Data Collection and Sources 22-24 May 2024, Geneva, Switzerland

19 April 2024

Optimizing Collection Strategy- Labor Force Survey

Cindy Ubartas and Sylvie Cyr (Statistics Canada, Canada) [email protected] Abstract In the context of declining response rates for the Labour Force Survey (LFS), our organization is implementing innovative strategies to improve engagement and participation. This includes optimizing our collection strategy, enhancing our outreach efforts, and leveraging technology to make the survey process more user-friendly and accessible. Statistics Canada’s new model for LFS will leverage pre-contact and pre-collection activities in advance of the regular 10 day collection cycle, allowing for further respondent engagement and positive contact opportunities and support the promotion of electronic questionnaire (EQ) self-response as the primary collection mode. The agency continues to utilize experimental design principles, para data analysis and collection experience from other household surveys and Census to evaluate and shape our vision and balance the needs of data quality, cost and response burden. We will present an outline of the challenges and opportunities to help increase our response rate, while taking into consideration the possible bias of non- response.

Group of Experts on National Accounts April 2024

Regional economic accounts in Canada

Overview ▪Regional economic accounts in Canada

▪Background of the PTEA

▪ Income and expenditure accounts • Data sources • Deflation • Trade flows by region

▪ Industry accounts • Key concepts and data sources

▪PTEA challenges

▪Additional information

Regional economic accounts in Canada ▪ Regional economic accounts in Canada include:

- Provincial-Territorial Expenditure account - Provincial-Territorial Income account - Provincial-Territorial Household Current and capital accounts - Provincial-Territorial Industry account - Provincial-Territorial Supply and Use tables;

▪ Time series consistent Income and expenditure accounts and household sector accounts for the 13 regions start in 1981*;

▪ Industry accounts start in 1997, Supply & Use start 2010;

Provincial and territorial economic accounts,

2022

▪ Coordinated release in early November of each year: November 2023 included Supply Use tables for RY2020, regional Income, expenditure and sector accounts and Industry accounts up to RY2022

*Regional incomes and expenditure accounts were originally available back to 1961, but these were discontinued with SNA1993 integration.

Background of the PTEA ▪ The Income and expenditure accounts are fully consistent with the Supply

and Use tables by province/territory for the SUT reference year and many series use the SUT estimate as the benchmark • Current period estimates use more timely data sources;

▪ The Industry accounts are a projection of the SUT estimates to the current period for value-added using a broad range of indicators (e.g., production, labour inputs, physical volumes, etc.) • Assumption of fixed technology in the short term for GDP;

▪ Nominal estimates of GDP by incomes and by expenditures are reconciled, whereas volume growth rates from GDP by expenditures and GDP by industry are reconciled • Real GDP by expenditure is a chained fisher estimate, where GDP by industry is a

‘simulated fisher’.

Background of the PTEA, continued

▪ Provincial-territorial data sources underwent a massive overhaul in the 1990s with funding from the federal government under the ‘Project to improve provincial economic statistics’ • Improvements to existing annual surveys to capture provincial-territorial

information, development of enterprise-based business surveys;

▪ As a result, the PTEA estimates are used for the calculation of important financial transfers between federal and provincial/territorial governments • Allocation of sales tax revenues to jurisdictions participating in the Harmonized

sales tax program (collected by federal government) • Fiscal arrangements allocation to provinces (through Equalization) and

territories (through Territorial formula financing) to maintain an average fiscal capacity across all jurisdictions.

Income and expenditure accounts (IEA) ▪Presentation for Provincial-territorial (PT) income and expenditure

accounts follows the same as the quarterly national program, except for interprovincial trade flows;

▪Over 800 data (nominal) estimates by region within the IEA (360 for trade alone), approximately 500 volume estimates (and prices);

▪ Sequence of accounts is available for 14 regions (13 provinces & territories and outside Canada); • Sector accounts are only available for the household sector given that numerous

corporations and non-profits would be considered national in scope;

• Federal government is allocated to regions on an arbitrary basis (often using population) and is considered illustrative.

IEA: regional data sources

▪Certain series (such as housing, compensation, rent) are developed quarterly by PT, where the national is the sum of the regions;

▪Others have regional details on an annual basis for the current period;

▪Series without data sources by PT use the benchmark shares from the SUT, projected forward using the national growth rate.

IEA: Income accounts data sources GDP by income

Compensation of employees income tax data includes the province of employment, whereas province of residence is used for sector accounts

Operating surplus of corporations establishment based surveys, tax filings, energy surveys, projected from SUT benchmark

Consumption of fixed capital from capital stock model, built using PT data on investments, depreciation profile in national by asset, prices are PT by industry

Gross mixed income (non-farm, non- rent)

tax filings (business declarations), labour force indicators used for some industries

Gross mixed income (rent) developed at the PT level using a variety of data sources (housing stock in units, average rent from CPI, etc.)

Gross mixed income (farm) compiled at the PT level from our Agriculture division (farm cash receipts, expenses)

Taxes less subsidies reported data for local and provincial/territorial governments, federal government spread based on population

IEA: Expenditure accounts data sources GDP by expenditure

Household consumption survey of household spending, vehicle purchases by registration, surveys on retail and restaurants, energy admin data, financial based on bank filings & employment, insurance based on employment (as examples)

Government final consumption

by level of government (local, provincial/territorial), federal government split arbitrarily

NPISH final consumption based on tax filings for non-profits

Capital investment: construction

sub-annual surveys for buildings has PT details, engineering from annual CAPEX survey with PT details

Capital investment: M&E annual CAPEX survey, projected forward from SUT benchmark

Capital investment: IPP PT ratios projected from SUT benchmark using national level surveys, own-account using PT jobs and wages

Non-farm inventories business establishment-based surveys, energy surveys, tax data

International trade Canada’s trade with rest of world is total, PT splits are based on SUT benchmark projected using Final demand and energy/mining surveys

Interprovincial trade SUT benchmark projected forward using Final demand and energy/mining surveys

IEA: Deflation of expenditures

In most cases, the nominal is derived first which is deflated with prices to estimate the volumes;

Examples of prices with provincial-territorial dimension: • Consumer price index • Machinery and equipment price index (by industry, by PT) • Building construction price index • Own-account capital uses movement of wages

➢There is no data source for prices of trade flows by region, national prices are used for the most part

IEA: Trade flows by region ▪ Inter-provincial trade flows measure the annual sales of goods and services among the

provinces and territories, whereas international trade by province reflects annual sales between individual provinces/territories and the rest of the world • Both are estimated for Expenditure based GDP by region;

▪ Developed by balancing the supply and demand for goods and services by province and within the SUT framework; beyond the SUT year, the modelled flows are confronted against energy/mining surveys and final domestic demand;

▪ Nuances with PT trade flows: • definition of certain services is different in national versus provincial system therefore international

trade of services is not equal between the two systems (total international trade is),

• Inter-provincial services require special attention: for example, insurance – there could be a disaster in one province, but the insurance company (and staff) are in another province. Therefore, we estimate an interprovincial trade flow of insurance services.

Industry accounts (IA)

▪GDP by industry, by province and territory is estimated for 226 industries plus 111 aggregates thereof, including special aggregations such as ‘goods-producing industries’, ‘energy sector’, ‘public sector’ for each region • various levels/aggregations permit targeted analysis and

industrial performance reviews

▪Estimation method: • SUT year – Gross output (GO) and GDP from SUT • Beyond SUT year – project GO and GDP in real terms using

various indicators, national totals benchmarked to monthly GDP by industry.

Industry Accounts: key concepts and data sources ▪ Assumption of fixed real technology coefficients -> changes (real) in the

indicator used (output, labour or usage) correspond to changes in (real) value added

▪ Due to limited data sources, cannot estimate value-added directly so partial indicators of an industry’s production function are used

• Principal outputs, employment, some intermediate inputs

▪ Deflation: unit values (where quantities and prices are available) or price indexes

▪ Main data sources: • Supply and use tables for SUT reference year • Output indicators (Statistics Canada surveys, financial reports), employment, GST,

price indexes, etc. • Use of Income and expenditure accounts components: Gross output indicators such

as investment (construction), household final consumption expenditures (personal services industries), government labour income and depreciation.

PTEA: Challenges Although having three measures of GDP (income, expenditure and production) is a source of strength in the PTEA, it can also pose unique challenges:

▪ Some regions are relatively small and/or are focused in one or two main sectors, in these cases, it is tricky to come to the correct growth rate, adjusting these small regions can also be tricky;

▪ Derived estimates can show conflicting signals (for example surplus, gross output and trade);

▪ For years beyond the SUT year, the importance of an industry or sector at the PT level might not always be apparent within the national estimates (PT IA is benchmarked to the national GDP by industry), and so when provincial has a certain target growth there can be a trade-off between revisions at the national level or another region compensating;

PTEA: Challenges, cont. The pandemic introduced new challenges with regional estimation:

▪ Supply chain disruptions were particularly difficult in smaller regions including rapid price increases • Using a national deflator would minimize these regional impacts;

▪ Large impacts were felt in services industries • regions that are heavily dependent on those industries had a more difficult experience

both during the pandemic and with the recovery;

➢Given the projection beyond the SUT year, revisions are larger than before the pandemic and quite different across regions.

Additional information PTEA annual releases: ➢Provincial and territorial economic accounts, 2022: https://www150.statcan.gc.ca/n1/daily-

quotidien/231108/dq231108b-eng.htm

➢Provincial and territorial economic accounts: Interactive tool: https://www150.statcan.gc.ca/n1/pub/71-607-x/71-607-x2019022-eng.htm

➢Gross domestic product (GDP) by industry, provinces and territories: Interactive tool: https://www150.statcan.gc.ca/n1/en/catalogue/71-607-X2019024

➢Supply and use tables, 2020: https://www150.statcan.gc.ca/n1/daily-quotidien/231108/dq231108f-eng.htm

Sub-annual components: ➢Stock and consumption of fixed capital (quarterly capital stock program):

https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=3410016301

➢Wages and salaries: https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=3610020501

➢Housing stock in units: https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=3610068801