Concepts and Methods Guide
3. Survey design

Skip to text

Text begins

3.1 Target population and coverage

The target population of the 2017 Aboriginal Peoples Survey (APS) was composed of the Aboriginal identity population of Canada, 15 years of age and over as of January 15, 2017, living in private dwellings excluding people living on Indian reserves and settlements and in certain First Nations communities in Yukon and the Northwest Territories (NWT). These exclusions are the same as those in the 2012 APS. The census subdivisions (CSD) covered in Yukon and the NWT are listed in Table 2.

The CSD types can be found in Table 1.5 of the Census Dictionary.

3.1.1. Identifying the Aboriginal population

The APS selected its sample from respondents who reported either Aboriginal identity or Aboriginal ancestry to the 2016 Census questionnaire. More precisely, the APS sample was selected from respondents who gave specific answers to four screening questions on the census long-form questionnaire, which had two main versions, the 2A-L form and the 2A-R form.

The 2A-L form was completed by self-enumeration and was administered to approximately one in four households in most parts of Canada (2A-L regions). Other than the basic census demographic questions (name, sex, date of birth, legal marital status, common-law status, relationship to person 1, various language questions and the consent question to release the data in 92 years), the 2016 Census 2A-L form included questions on labour market activity, income, education, activity limitations, citizenship, housing, ethnic origin, and so on.

The 2A-R form, identical in content to the 2A-L form, except for some adapted examples and excluded questions^Note , was administered by personal interview to all households in remote areas, Inuit communities and Indian reserves and settlements (2A-R regions).

The four screening questions used to identify the Aboriginal population were ethnic origin (question 17), Aboriginal self-reporting (question 18), Status Indian (Registered or Treaty Indian, question 20), and First Nation/Indian band membership (question 21).

The reporting of an Aboriginal origin in question 17 defines the Aboriginal ancestry population (or ancestry population).

The Aboriginal identity population of the 2016 Census is derived from three questions: questions 18, 20 and 21. The concept of Aboriginal identity refers to those persons who either (a) self-reported as at least one Aboriginal group, namely, First Nations (North American Indian), Métis or Inuit; and/or (b) reported being a Status Indian (Registered Indian or Treaty Indian, as defined by the Indian Act of Canada); and/or (c) reported being a member of a First Nation or Indian band.

Within the context of the APS, individuals with an Aboriginal ancestry who did not have an Aboriginal identity in the census are defined as the Aboriginal ancestry-only population (or ancestry-only population). For the purposes of APS sampling, the Aboriginal population includes both the identity population and the ancestry-only population. Although, similarly to the 2012 APS, the ancestry-only population was not part of the 2017 APS target population, it was still sampled in the Census because it was noted that in past survey iterations, slightly less than one-third of the census ancestry-only population reported Aboriginal identity in the APS.

The Aboriginal identity concept in the APS is the same as that in the Census, but is defined based on a slightly different set of questions (see Table 1 in section 2.2).

2016 Census – 2A-L, question 17

What were the ethnic or cultural origins of this person’s ancestors?

An ancestor is usually more distant than a grandparent.

For example, Canadian, English, Chinese, French, East Indian, Italian, German, Scottish, Cree, Mi’kmaq, Salish, Métis, Inuit, Filipino, Irish, Dutch, Ukrainian, Polish, Portuguese, Vietnamese, Korean, Jamaican, Greek, Iranian, Lebanese, Mexican, Somali, Colombian, etc.

This question collects information on the ancestral origins of the population and provides information about the composition of Canada’s diverse population.

1. Specify as many origins as applicable using capital letters.

2016 Census – 2A-L, question 18

Is this person an Aboriginal person, that is, First Nations (North American Indian), Métis or Inuk (Inuit)? Note: First Nations (North American Indian) includes Status and Non-Status Indians.

If “Yes”, mark the circle(s) that best describe(s) this person now.

1. No, not an Aboriginal person. Continue with the next question.
2. Yes, First Nations (North American Indian). Go to question 20.
3. Yes, Métis. Go to question 20.
4. Yes, Inuk (Inuit). Go to question 20.

2016 Census – 2A-L, question 20

Is this person a Status Indian (Registered or Treaty Indian as defined by the Indian Act of Canada)?

1. No
2. Yes, Status Indian (Registered or Treaty)

2016 Census – 2A-L, question 21

Is this person a member of a First Nation/Indian band?

If “Yes”, which First Nation/Indian band?
For example, Musqueam Indian Band, Sturgeon Lake First Nation, Atikamekw of Manawan.

1. No
2. Yes, member of a First Nation/Indian band.
   Specify name of First Nation/Indian band.

3.1.2. Survey reference date

January 15, 2017 was used as the APS reference date. This date corresponds approximately to the beginning of data collection for the survey. Age is established based on this reference date.

3.1.3. The 2016 Census frame

The APS sample was selected from the unedited, non-imputed database of the 2016 Census, which is the census database referred to as the Response Database (RDB). Before selecting the sample, survey methodologists developed an editing and imputation strategy to deal with missing values in any of the four census screening questions or in the variables used during stratification (including age and certain education variables), as well as for individuals with certain characteristics which appeared to be inconsistent with being Aboriginal.

The first step in selecting the sample was to include on the survey frame all individuals reporting Aboriginal identity or ancestry who were aged 15 and over as of January 15, 2017.

In the second step, all individuals who were part of the same households as the units in the initial frame were added to the frame. These additional persons were retained on the frame only if their responses to screening questions were missing but the characteristics of the people in the household who had answered the screening questions indicated that they would have had a good chance of having Aboriginal identity and/or ancestry. Hence, an individual with missing screening questions on identity would normally have been imputed as having Aboriginal identity if at least 50% of the members of the same household who completed the identity screening questions had Aboriginal identity. Similarly, an individual with non-response to the ancestry screening question would normally have been imputed as having Aboriginal ancestry if at least 50% of the members of the same household who completed the screening question on ancestry had Aboriginal ancestry.

It was important to stratify the 2017 APS sample by age group (18 to 24 years of age, 25 to 54 years of age, 55 or more years of age). Estimates were targeted for each of these three groups. Although estimates were not specifically targeted for them, individuals aged 15 to 17 years of age as of January 15, 2017 were also included in the sample.

Once the processing was complete, individuals under 15 years of age as of January 15, 2017, those living on reserves or in certain specific communities in Yukon and the NWT and individuals who no longer self-identified as Aboriginal following processing were excluded from the survey frame.

3.2 Sampling design

3.2.1 Domains of estimation

An effective stratification uses domains of estimation. Domains of estimation are groups of units for which estimates are targeted. In the case of the APS, stratification-specific domains of estimation were used. These domains of estimation corresponded to geographical regions for which estimates with an “acceptable” level of precision for a particular Aboriginal group (i.e. First Nations, Métis or Inuit) and particular age group were targeted.

An example of a domain of estimation would be Métis in Alberta aged 25 to 54. During stratification, the Métis Aboriginal group was comprised of individuals reporting Métis identity alone to Census question 18 or, for individuals without Aboriginal identity (Aboriginal ancestry-only population), Métis ancestry alone to question 17 (with or without non-Aboriginal ancestry). In reality, Aboriginal ancestry-only individuals were not part of the survey’s target population but were sampled because they had a relatively important chance of reporting identity on the survey as described in section 3.1.1. This is why the term “stratification-specific domains of estimation” is used rather than the term “survey-specific domains of estimation”.

More precisely, the stratification-specific domains of estimation were created by cross-tabulating the following variables:

• Geography
  • Inuit regions
  • Outside Inuit regions
    • province/territory
    • Atlantic provinces grouped
• Age group
  • 18 to 24 years of age
  • 25 to 54 years of age
  • 55 or more years of age
• Aboriginal group
  • Inuit in Inuit regions
  • Inuit outside Inuit regions (rest of Canada)
  • Aboriginal groups (excluding Inuit) combined for Atlantic Canada (outside Nunatsiavut), Quebec (outside
  • Nunavik), Yukon and NWT (outside Inuvialuit)
  • For Ontario, Manitoba, Saskatchewan, Alberta and British Columbia
    • Status First Nations people living off reserve
    • Non-Status First Nations people living off reserve
    • Métis

Below is how the Aboriginal groups were defined during stratification based on responses to the Census:

• Status First Nations people living off reserve – persons answering First Nations alone to Census question 18 and answering YES to Census question 20
• Non-Status First Nations people living off reserve – persons answering First Nations alone to question 18 and answering NO to question 20 or persons without Aboriginal identity but responding First Nations alone to question 17 (with or without non-Aboriginal ancestry)
• Métis only – persons answering Métis only to question 18 or persons without Aboriginal identity but answering

Métis alone to question 17 (with or without non-Aboriginal ancestry)

• Inuit only – persons answering Inuit only to question 18 or persons without Aboriginal identity but answering

Inuit alone to question 17 (with or without non-Aboriginal ancestry)

• Multiple Aboriginal group – persons reporting more than one identity to question 18 or persons without Aboriginal identity but reporting more than one Aboriginal ancestry to question 17 (with or without non-Aboriginal ancestry)
• Status Indian or member of a First Nation / Indian band only – persons answering YES to question 20 or to question 21 but NO to question 18.

In Atlantic Canada outside Nunatsiavut, in Quebec outside Nunavik, in Yukon and in the NWT outside Inuvialuit, the number of Métis is generally too small to be able to produce separate estimates for First Nations people and Métis by age group. For this reason, all Aboriginal groups (excluding Inuit) were combined during stratification.

Note that estimates were targeted for Inuit outside Inuit regions nationally; therefore, a domain of estimation was created for this group.

In total, 72 target domains of estimation and 60 supplementary domains of estimation were created, for a total of 132 domains of estimation. Supplementary domains of estimation included domains where there was no plan to produce estimates for the APS but where a particular group should still be represented in the sample. Examples of supplementary domains of estimation included the non-Inuit population living in Inuit regions and the population aged 15 to 17 years old.

For each target domain, the goal was to estimate a characteristic present for at least 10% of individuals in the domain, with a given coefficient of variation (CV). Depending on the domain sizes, CV values were set to 20%, 22.5%, 25% or 33%. The CV is a measure of the precision of the estimate, which is described in section 7.2. The minimum proportion targeted is referred to as “min-p”. The ability to achieve the targeted CVs for a given min-p value depended on such factors as:

• the size of the population,
• the number of respondents available from the 2016 Census,
• the expected response rate,
• the expected number of false positives (persons reporting Aboriginal identity or ancestry in the census without Aboriginal identity in the APS, a very specific type of respondents outside the target population),
• the sample loss associated with the constraint of selecting a maximum of three persons per household and
• the loss associated with reducing overlap with other surveys.

The following table gives the targeted CVs and min-p for each target domain of estimation.

Clearly, any combination of domains will give smaller CVs. Because separate estimates were targeted for Status and Non-Status First Nations people for each province between Ontario and British Columbia, CVs of 25% or 33% were set for these two groups, which should produce CVs around 20% for First Nations as a whole. CVs of 22.5% were targeted for Métis in each province between Ontario and British Columbia.

3.2.2. Stratification

Stratification will produce more precise estimates if units are homogeneous within strata and heterogeneous between strata. In addition, the estimation weights associated with survey respondents should ideally be as close as possible within strata.

Three variables were used to create the 2017 APS strata:

• the region associated with the census form type (2A-L or 2A-R),
• whether the individual self-responded to the census or responded during non-response follow-up (NRFU), and
• the type of Aboriginal identification.

One of the assumptions used in the APS sample allocation method was that the census weights would vary as little as possible within an APS stratum. As previously mentioned, the region associated with the census form type (2A-L or 2A-R) and the collection method created a difference in census weights. In regions where the 2A-L form was administered, a systematic sample of one in four households was drawn. In contrast, in regions where the 2A-R form was used, the form was administered to all households.

The other two factors considered for the APS stratification, the type of census respondent (self-respondent or NRFU) and the type of Aboriginal identification (Aboriginal ancestry-only or Aboriginal identity) are unrelated to the variability of the census weights. It was important to consider these variables as stratification factors since past studies have shown that individuals who self-respond to the census have different characteristics than those who respond during NRFU and similarly, the Aboriginal identity (ID) group and the Aboriginal ancestry-only (AO) group also have very different characteristics. Incorporating these variables in the stratification will increase the homogeneity of the strata. Moreover, considering the type of Aboriginal identification (AO or ID) will allow the allocation of the sample to reflect the probability of each unit being part of the target population (having Aboriginal identity on the APS). Nationally, based on past survey data, individuals having identity on the census have a probability of about 88% of having identity on the APS, while individuals with ancestry-only on the census have a probability of about 32% of having identity on the APS.  

Combining the factors form type (2A-L or 2A-R), type of census respondent (self-respondent or NRFU respondent) and type of Aboriginal identification (AO or ID) gave a maximum of six strata per domain of estimation:

• 2A-L, self-respondent, AO
• 2A-L, self-respondent, ID
• 2A-L, NRFU respondent, AO
• 2A-L, NRFU respondent, ID
• 2A-R, AO
• 2A-R, ID

For certain domains such as Status First Nations, there were only three possible strata because included individuals had Aboriginal identity by definition. Some strata may also have been empty, especially in the supplementary domains. For example, for a specific Inuit region, it was possible that no one classified as Aboriginal ancestry-only had Aboriginal ancestry other than Inuit.

3.2.3. Sampling design and sample allocation

The 2017 APS selected its sample from 2016 Census long-form respondents. Thus, the APS sampling design can be considered a two-phase design where the first phase corresponds to the census long-form sample and the second phase corresponds to the APS sample.

Once the frame had been constructed, it was stratified according to domain of estimation and then substratified by form type (2A-L vs. 2A-R), type of census respondent (self-respondent vs. NRFU respondent) and type of Aboriginal identification (ID vs. AO). A systematic random sample was then selected within each stratum, the frame having been pre-sorted by household and person number. The purpose of this was to ensure proper distribution of the sample geographically within the strata as well as across as many households as possible. This procedure was, however, limited by the fact that members of the same household can be in different strata.

A method for optimal allocation between the substrata of a particular domain was used by taking into account the different types of sample size loss as described in section 3.2.4 as well as the probability of each unit belonging to the target population in a given stratum. This allocation depended in part on the census weights. It should be noted that at the time of the allocation, these weights had not yet been calculated. Preliminary weights were therefore derived solely for the allocation. The definitive weights derived by the methodology team working on the census estimates were used during weighting (see section 6).

Allocation was done in the survey’s target domains first. For the non-targeted or supplementary domains (for example, non-Inuit in Inuit regions or persons 15 to 17 years of age), the sample size in a given stratum was calculated using a sampling fraction equal to or less than the sampling fraction of the corresponding stratum in the corresponding target domain. These supplementary domains had to be retained in order to cover the entire target population but did not require a specific sample size since no estimates were to be derived for them. For non-targeted Aboriginal groups, such as non-Inuit in Inuit regions, the same sampling fraction as for Inuit was used. For persons 15 to 17 years of age, half of the sampling fraction of persons 18 years of age or over (i.e. all other age groups combined) was used.

Although the plan was to select the 2017 APS sample in a single wave, supplementary sample was selected at two additional different occasions.

On the first day of collection, it was discovered that the Whapmagoostui reserve in Nunavik had been incorrectly included on the sampling frame and individuals living on reserve had been selected for the APS sample. These individuals were removed from collection, the sampling frame was corrected and all sample size calculations were repeated for Nunavik. This led to an additional sample of 8 individuals being selected in Nunavik to meet sample size requirements.

Midway throughout collection, it was noticed that the proportion of the sample selected among First Nations without Status in Ontario to British Columbia (rate of Aboriginal identity on the survey) was lower than expected, in particular for those who had Aboriginal ancestry-only on the census. Therefore, it was decided to select a top-up sample.

To calculate the size of the top-up sample, the identity rates on the survey were reviewed for First Nations without Status in Ontario to British Columbia based on the results observed during collection. The same allocation method was then used to determine new sample sizes. The difference between this sample size and the size of the initial sample gave the size of the top-up sample. The top-up sample contained a total of 822 additional individuals.

The initial APS sample consisted of 42,815 individuals across Canada. After the two additional samples were selected, the total APS sample consisted of 43,645 individuals.

3.2.4. Sample size adjustment

Probability of belonging to the target population

Past survey iterations found that approximately 88% of individuals with Aboriginal identity in the census report Aboriginal identity in the APS and approximately 32% of those with ancestry-only in the census report identity in the APS. These probabilities of belonging to the target population were expected to be lower in 2017 than in 2012. Accordingly, the probabilities of belonging to the specified 2017 target population were adjusted downward by multiplying 2012 figures by 95% in the census Aboriginal ancestry-only strata and in the census identity strata.

Response rates

Response rates from the 2012 APS were used to allocate the sample for the 2017 APS. As a preventive measure, these rates were reduced by multiplying them by an adjustment factor of 90%. Note that increasing the sample size based on an expected response rate is a protective measure to reduce sampling error but does nothing to reduce non-response bias, a particular type of non-sampling error (see section 7.3). Clearly, it is preferable to obtain a higher response rate on a smaller sample size than a lower response rate on a larger sample size.

Households with more than three individuals selected

Because the sample unit was the individual and not the household, the sample may have included several individuals from a single household. To limit response burden, one of the constraints imposed during collection was to select no more than three individuals per household. As a way to estimate this loss, preliminary samples were selected and the average number of persons who would have been removed from the sample because of this constraint was calculated. The relative loss obtained was multiplied by 1.1 for each combination of domains and strata and was incorporated in the sample size adjustment factor.

Overlap with other surveys

The APS collection period overlapped with several other Statistics Canada surveys. It is generally understood that an individual contacted for one survey is less likely to agree to participate in that survey if he or she has just been interviewed for another survey. To compensate for this potential loss, the total sample size was increased uniformly by about 5%. Several of the overlapping surveys also cover the territories where the Aboriginal population represents a high portion of the total population. Therefore, there was a higher possibility of overlap between the different surveys in the territories. Special attention was given to this matter to reduce overlap as much as possible in order to reduce the potential burden on respondents. Overlap in the territories was examined for the Canadian Community Health Survey (CCHS), the Labour Force Survey (LFS) and the Survey of Household Spending (SHS) and steps were taken to minimize response burden where possible.

For example, overlap with the CCHS was of particular concern. Some of the APS questions on health are drawn directly from the CCHS, including several highly sensitive questions on mental health. No measures were taken for the CCHS interviews held before and after the initial APS collection period before January 2017 and after June 2017. However, the identifiable overlap in households to be interviewed between January 2017 and June 2017 for the CCHS was eliminated. Households selected for the CCHS between January and June, and which had responded to the census long-form were removed from the APS survey frame. Within each community covered by the CCHS, the weights of the remaining census long-form households were adjusted to represent all of the community’s households. Since these households had been removed from the APS survey frame, they therefore had no chance of being selected.

3.3 Sample size and response rate

The final sample of the 2017 APS contained a total of 43,645 units, 405 of which were not sent to collection (see section 6.2.). Table 4 shows the allocation of the 43,240 units sampled by geographical domain and the corresponding response rates.

A response rate is defined as the number of eligible respondents divided by the number of eligible units in the sample.

Persons living outside of Canada or in an institution at the time of the survey or under 15 years of age as of January 15, 2017 are examples of ineligible units for the APS.

Two definitions of eligible units were used for the APS. In the first definition, individuals without Aboriginal identity on the APS were deemed ineligible (in addition to the other types of ineligible units). In the second definition, these individuals were deemed eligible and were included as respondents. These individuals agreed to participate in the survey and completed all of the questions administered to them, that is, the screening questions determining whether or not they have Aboriginal identity.

These two definitions of eligible units and respondents therefore offer two ways to measure the response rate. The first response rate of 70.1% is generally the response rate used during collection. This is a response rate relative to the sampled units falling within the APS identity population. The second response rate of 76.0% is the one used by Methodology and is a response rate which is more relative to all sampled units (with some exclusions). Of course, in both cases, non-responding units cannot always be classified as being eligible or not eligible. Given that the probability of having identity on the APS differs substantially between individuals with Aboriginal identity on the census and those with ancestry-only on the census, the response rate used during collection is highly influenced by the allocation of the sample between these two groups, which is not the case for the response rate used by Methodology. In Inuit regions, because most individuals have identity, the two response rates are quite similar, which is not the case in the other regions. The maximum difference between the two rates occurs in Quebec outside Nunavik. In this region, the Aboriginal ancestry-only population forms a large part of the total Aboriginal population and, according to past APS surveys, the probabilities of having Aboriginal identity on the survey are the lowest in the country regardless of whether the individual has Aboriginal ancestry-only or Aboriginal identity on the census.

• Previous
• Next