Participation and Activity Limitation Survey (PALS)

Detailed information for 2001

Statistics Canada is conducting a survey on Canadians (adults and children), whose day-to-day activities may be limited because of a condition or health problem. Survey results will help to identify difficulties and barriers these Canadians may face.

Data release - December 3, 2002

• Questionnaire(s) and reporting guide(s)
• Description
• Data sources and methodology
• Data accuracy

Description

The Participation and Activity Limitation Survey (PALS) is a post-censal survey which collects information about persons whose everyday activities are limited because of a health-related condition or problem.

PALS is a post-censal survey because it uses the census as a sampling frame to identify its target population. For example: the 2001 Census questionnaire included two general questions on activity limitations and long-term disabilities. The 2001 PALS respondents were selected through the use of the census information on age, geography and the responses to these two general questions.

The Health and Activity Limitation Survey (HALS -- see the "Other reference periods" sidebar) was conducted by Statistics Canada about persons with disabilities in 1986 and 1991. In 2001 the Health and Activity Limitation Survey was renamed the Participation and Activity Limitation Survey. The new name reflected the fact that the new survey would focus on the participation of persons with activity limitations.

The data collected by the survey are used to plan services and programs required by persons with disabilities to participate fully in our society. PALS is funded by Human Resources Development Canada (HRDC).

Subjects

• Disability
• Equity and inclusion
• Health
• Society and community

Data sources and methodology

Target population

The PALS survey population consists of all persons who reported disabilities to the Census questions on activity limitations and who were living in Canada at the time of the Census.

The population covered by the survey was persons living in private and some collective households in the 10 provinces. The population living in the Aboriginal communities covered by the Aboriginal People Survey (APS), including all First Nations reserves, were excluded as well as the population living in the three northern territories and the residents of institutional collectives.

In addition, individuals living on military bases, Canadian Armed Forces vessels, merchant vessels and coast guard vessels, as well as campgrounds and parks were excluded for operational reasons. The target population of PALS is the subset of the surveyed population that also reported disabilities in PALS.

Instrument design

The adults and children's questionnaires were developed based on the review of the 1991 Health and Activity Limitation Survey (HALS) questionnaires and an input consultation with client (that is, HRDC), federal and provincial governments and community associations.

A pilot test was conducted in the fall of 2000 to test the content of the questionnaires. The questionnaires were tested in both official languages.

Sampling

This is a sample survey with a cross-sectional design.

Frame description:

The sampling frame used for PALS consisted of estimates of the population by Enumeration Area (EA), age group and severity of disability according to the census definition of disability. These estimates were obtained from projections of the Canadian population produced by the Statistics Canada's Demography Division to which disability rates (using the census definition) as estimated from the 2000 PALS pilot test were applied. The reference period for these projections was the time of the 2001 Census.

Stratification and Sample selection methods:

The strata are defined by the cross-classification of the ten provinces, the age and the severity of disability as defined by the census. The four age groups considered are under 25, 25 to 44, 45 to 64 and 65 years of age and older. The two levels of severity are "Often" and "Sometimes".

The sample allocation within each stratum was done using a simulation involving the 1996 Census data. Since the 2001 Census activity limitation questions were not on the 1996 Census form, some modeling assumptions involving the PALS pilot test were used. In order to determine the sample size in each stratum, a targeted minimum proportion and coefficient of variation (CV) were fixed. Some parameters were estimated using the 2000 pilot test, such as the response rate and the percentage of persons with disabilities in the 2001 Census who are also disabled in PALS.

The sampling design is a two-stage stratified design that uses the 2001 Census long form sample, which is administered to one in five households in Canada. The primary sampling unit (PSU) is made up of one or more Enumeration Areas and the secondary sampling unit is the respondent. A sample of PSUs is selected in each stratum using a probability proportional-to-size (PPS) sampling design.

The PSU size is the predicted size of the projected census disabled population of that PSU within the stratum for which the PSU was selected (combination of age group and severity). Note that a PSU can be selected for more than one age group and severity combination as independent samples are selected within each stratum. Some very large units were sampled with probability of one. PSUs for which no or very few disabled individuals were predicted, were selected using a stratified random sample of PSUs.

In the second stage of the sample design, all of the 2001 Census long form respondents having the characteristics of the stratum for which the PSU was selected were included in the 2001 PALS sample. This second stage took place in the five census Field Collection Units (FCU) across Canada.

The FCU operations constituted the last phase of the census data collection process. In the census field collection units, each EA box selected for PALS was identified by the stratum description for which the Enumeration Area was selected (age group and severity combination). The clerks then had to list all the individuals with the specified characteristics on forms designed for that purpose.

The PPS sampling design allows a better control of the sample size and provides higher probabilities of selection to the larger primary sampling units. Given the FCU sampling selection process, the PPS design was particularly useful (for example, the reduction of the number of EA boxes to be inspected).

Sample size:

The total sample size is 43,276. Of these individuals, 42,485 live in private dwellings, while 791 are residents of collective dwellings.

• Additional documentation

Data sources

Data collection for this reference period: 2001-09-10 to 2002-01-31

Responding to this survey is voluntary.

Data are collected directly from survey respondents.

The interviews were conducted by telephone with the interviewers completing a paper and pencil questionnaire.

Interviews by proxy were allowed. In some special cases, face-to-face interviews were carried out. The interviews for the children's questionnaire were conducted with the parent or guardian of the child.

View the Questionnaire(s) and reporting guide(s).

Error detection

The first phase of error detection was done during the data collection. At that stage, the interviewer's supervisors reviewed the completed questionnaires. Observed inconsistencies were discussed with the interviewer who conducted the interview and the respondent was called back if required.

The second phase of error detection was conducted during data processing which was made up of many steps. The first step was the data validation where, among other actions, multiple responses were blanked out and processed with the other missing responses. The second step of the data processing was the editing.

Edit rules were developed to identify and correct inconsistencies between responses within each section of the adults and children's questionnaires. For most of the situations corresponding to inconsistencies an automated correction was specified, as discussed in the IMPUTATION Section. Once this step was completed, a macro verification was done by analyzing frequency distributions to identify anomalies, for example missing categories or unusually large frequencies.

Imputation

For PALS, the only type of imputation done was deterministic imputation. Once inconsistencies were identified between responses, a corrective action had to be used on at least one of the responses. The approach used to determine the appropriate action was generally the "bottom-up" approach. With this strategy, questions related to each other were edited simultaneously. If the answer to Question A determined that Question B was to be asked, then Question B was edited first and the edited responses to B were then used to determine if the response to Question A was correct. If both responses were inconsistent, the response to Question A was modified deterministically if possible. Conversely, for a small number of questions, a "top-down" approach was used. With this approach, responses to previous questions determined whether a subsequent question was to be asked. Although the corrections were generally done in an automated way, analysts reviewed some problematic situations.

During this review, a valid response was deterministically imputed for the missing responses if sufficient information was available in the related questions. Otherwise, it was coded to "Not stated". In addition, the questions that were not to be asked were coded to "Valid skip". If a question with a missing answer (coded to "Not stated") should have been used to determine if subsequent questions were to be asked, these subsequent questions were set to "Path not known", because it was not possible to determine whether or not they should have been asked.

However, non-response was not permitted for the demographic information required for weighting, namely the age and sex of the respondent. This information was asked at the beginning of the interview to make sure the selected person had been reached. These two variables were imputed from the census if they were missing or invalid. In particular, an age was considered invalid if it was not consistent with the questionnaire used.

Estimation

In a sample survey, each respondent represents not only himself/herself, but also other persons that were not sampled. Consequently, a weight is associated to each respondent to indicate the number of persons that this respondent represents. This weight must be used for all estimations.

The weight is calculated in a three-stage process. The FIRST stage is the assignment of an initial weight based on the sampling design. The initial weight is the inverse of the inclusion probability. For the 2001 PALS, the initial weight was the product of three components: the inverse of the sampling fraction of the Primary Sampling Unit (called the PSU weight), the census weight and the subsampling weight. Following this calculation, individuals selected by mistake and those missed during sample selection were taken into consideration and appropriate weight adjustments were applied to the initial weight.

The SECOND stage of the weighting process is the adjustment for non-response. More specifically, two adjustments were done at this stage to take into consideration the fact that the non-respondents can be classified into one of two main categories with very different characteristics: the persons not contacted and the persons contacted but who did not respond.

The weights were adjusted first for non-contact and then for non-response. As the adjustment method is being the same for both types of non-respondents, it is described here only for non-response. The non-response adjustment is done by forming non-response adjustment classes in such a way that the records in each class have similar response probabilities. The estimated response probabilities were obtained by developing a logistic regression model to predict the response probability using explanatory variables.

Many explanatory variables can be used since all census long form information is available for each respondent and non-respondent. Separate models were used for children and for adults. Approximately ten classes of roughly the same size were obtained for each logistic regression model. The inverse of the weighted response rate in a class is used as the weighting adjustment factor for that class and the initial weights of the respondents within the class are adjusted accordingly.

The THIRD stage of the weighting adjustment is the post-stratification. This adjustment ensures that the sum of the final weights for the respondents is equal to the population counts obtained from the census. This adjustment is done for groups (called post-strata) defined by the combinations of province, gender, age group (generally 5-year age groups) and severity of the limitation reported in the census. The weights corrected for non-response are then adjusted using the ratio of the census count to the sample count for each post-stratum.

Quality evaluation

Two types of error occur in surveys, namely sampling and non-sampling errors. As opposed to the sampling error, non-sampling errors are not explained by sample-to-sample variability. These errors can occur at any step of the survey process and actions have been taken to reduce them to a minimum. Following is a description of measures that were put in place for that purpose.

A pilot test was conducted one year before the survey to evaluate all the survey process, from the questionnaire content to the data processing.

The sample selection procedures were also tested using questionnaires from the 1998 National Census Test in order to identify difficulties that would be faced by the sample selection clerks in the census Field Collection Units (FCUs). In addition, quality control procedures were used during the sample selection in these FCUs.

High response rates are essential for quality data. To reduce the number of non-response cases, the interviewers were all trained by Statistics Canada's staff, provided with detailed Interviewer Manuals, and were under the direction of interviewer supervisors. Refusals were followed up by senior interviewers to encourage respondents to participate in the survey.

In addition, some measures were taken to identify and correct errors that could result from misinterpretation of a question by the respondent or from a wrong flow followed in the questionnaire. The questionnaires were first reviewed by the interviewer's supervisor. A detailed set of edit rules were then used during data processing to identify and correct any inconsistencies between the responses provided. These edit rules were exhaustively tested before being applied to the data.

Disclosure control

Statistics Canada is prohibited by law from releasing any information it collects that could identify any person, business, or organization, unless consent has been given by the respondent or as permitted by the Statistics Act. Various confidentiality rules are applied to all data that are released or published to prevent the publication or disclosure of any information deemed confidential. If necessary, data are suppressed to prevent direct or residual disclosure of identifiable data.

In order to prevent any data disclosure, confidentiality analysis is done using the Statistics Canada Generalized Disclosure Control System (G-Confid). G-Confid is used for primary suppression (direct disclosure) as well as for secondary suppression (residual disclosure). Direct disclosure occurs when the value in a tabulation cell is composed of or dominated by few enterprises while residual disclosure occurs when confidential information can be derived indirectly by piecing together information from different sources or data series.

Revisions and seasonal adjustment

This methodology does not apply to this survey.

Data accuracy

The overall response rate was 82.5%.

The data accuracy measure used for each table produced is the estimated standard error of the estimate (sampling error measure), which is the square root of the estimated sampling variance of the estimate. However, the estimated standard error is usually expressed relative to the estimate to which it pertains, and the resulting measure is the estimated coefficient of variation (CV).

The estimated CV is obtained by dividing the estimated standard error of the estimate by the estimate itself and is expressed as a percentage of the estimate. For PALS, all estimated CVs were obtained using the bootstrap method described in the ESTIMATION Section.