Survey Manager: Philip Giles

Host Organization: Statistics Canada

Year Initiated: 1993

Funding sources

Statistics Canada receives ongoing funding from the federal government to conduct the survey.

Host organization

Statistics Canada conducts the study.

- Governing body and role of external research advisors

None

Sample selection

The sample is drawn from that of the Canadian Labour Force Survey (LFS). The LFS uses an area frame and covers the population of the 10 provinces, with the exception of Indian reserves, the military and residents of institutions. (The coverage of the SLID sample is identical, with one small difference: SLID includes Armed Forces personnel living out of barracks.)

The size of each six-year panel is 15,000 households. This includes about 40,000 persons, of which 31,000 are aged 16 years and over.

The reasons for a staggered design with more frequent introductions of new panels are improved sample representativeness over time and flexibility for introducing new survey content. The choice of a six-year panel duration depended on this design and considerations about respondent burden as well. When only a few consecutive years of longitudinal data are required, the sample size can be doubled by combining data from the last three years of one panel and the first three years of the next, overlapping panel.

- Following rules

All longitudinal respondents are identified at the beginning of the panel, and they are followed for six years, whether they move or not. If a household splits up, all "branches" are followed.

Likewise, new people who start to live with a longitudinal respondent during the six years are also included in the survey, although they do not contribute to the longitudinal sample. They are called "cohabitants" in SLID. The reasons for interviewing cohabitants are: a) to maintain complete family and household data on longitudinal respondents; b) to obtain data on a representative cross-section of the population each year. 

Cohabitants follow a very similar interview format to longitudinal respondents. However, they do not begin the process until they enter the household of a longitudinal respondent, and they cease to be interviewed or followed as soon as they cease to live with a longitudinal respondent.

- Oversampling

None to directly address research using SLID. However, the LFS has a complex survey design that attaches unequal probabilities of selection to the various regions in the country.

- Weights and attrition bias

The sampling weight attached to a record indicates the number of units in the population that are represented by that unit in the sample. Examples of units are the person, family, household or job. There are three types of weights in SLID: the longitudinal weight, the cross-sectional weight, and the cross-sectional labour weight.

The longitudinal weight reflects the population at the time of sample selection, and is generally the most appropriate for longitudinal analysis which compares data from more than one year. For the first panel, these weights sum to the population as of December 31, 1992. Only longitudinal respondents are assigned a non-zero longitudinal weight in any year. The panel’s longitudinal weight is recalculated each year to take into account further non-response or attrition occurring that year.

The cross-sectional weight is appropriate to calculate estimates based on data from single years.

A different cross-sectional weight is calculated for each year, reflecting the population at the end of the reference year. The cross-sectional population includes both longitudinal respondents who remain in-scope for the survey as well as cohabitants.

The number of cohabitants grows each year as the members of the original dwellings move apart and live with other people. As the total number of people interviewed grows, the cross-sectional weights are adjusted downwards to prevent spurious population growth in the results. However, the weights are also adjusted upwards each year to account for attrition (non-response and failure to trace).

In adjusting for attrition, the weights of non-respondent individuals to a wave are redistributed among respondent individuals. If the characteristics of respondents and non-respondents were the same, this adjustment would be straightforward; however, it is generally accepted that there is some differential non-response, i.e. their characteristics are not the same. For example, among respondent individuals there are more employed persons and more people who have not moved. As a solution, the adjustment for non-response is carried out separately for groups of individuals having certain (known) characteristics in common.

- Sample "refreshing"

A new panel of respondents is introduced every 3 years.

There are 14 themes in all, grouped into four broad categories:

X     personal characteristics

X     education

X     income

X     labour

SLID collects data on a wide range of topics. Some are inherently "dynamic", involving transitions and spells, and others have important explanatory value. The following are the content themes organized under the topics of labour, income and wealth, education, and personal characteristics, including selections of the variables they contain.

I. Labour

Job characteristics (all characteristics updated each year and dates of changes recorded; collected for up to six jobs per year)

Characteristics of work absences lasting one or more weeks (collected on first and last absence each year, for each employer)

Receipt of compensation (whether benefits were received from each source and, if so, in which months)

Educational attainment (updated annually)

Household and economic and census family information (annual summary information, e.g., size, type)

- Mode

Virtually all interviews are conducted by telephone. Up to collection in 2000, the interviewing was decentralized and conducted by the interviewers in their homes. Starting in January 2001, the collection will be centralized in seven Statistics Canada offices across the country, with all collection being done by telephone. Proxy response is accepted in SLID, i.e. the interview is usually conducted with only one household member, who answers on behalf of all members of the household, provided he or she is knowledgeable.

- Instrument design (paper and pencil, CATI, EHC)

SLID uses computer-assisted interviewing (CAI).

- Dependent interviewing

"Dependent interviewing" becomes much easier in a computer environment. SLID especially benefits from this technique due to the long recall period required for most topics. Survey content for which dependent interviewing is used:

1. Household composition, demographics and relationships

2. Employer name and type of work done for each job held

3. Class of worker for each job

4. Wage rate or earnings at each job

5. School attendance

6. Receipt of benefits from "sensitive" government transfer programs (Employment Insurance, Workers' Compensation, Social Assistance)

In addition, dates of jobless spells are derived from the dates of job spells and verified with the respondent.

Respondents’ reaction to dependent interviewing has been favourable. In fact, respondents seem to expect interviewers to have available all the information collected at the previous interview.

- Calendar year, survey year and point of survey measures

Over the six years of a panel, up to 12 interviews are conducted.

A preliminary interview to collect background information.

The preliminary interview takes place at the same time as the first annual labour interview.

Six labour interviews (every January). The labour interview refers to the previous calendar year.

Six income interviews (every May). Like the labour interview, the reference period is the previous calendar year. The income interview is deferred until May to take advantage of income tax time when respondents are more familiar with their records. An income interview is not required if the respondent gives Statistics Canada permission to use his or her tax information from Revenue Canada for the purposes of this survey.

Since the labour and income interviews each year both refer to the previous calendar year, together they make up one wave. This represents a kind of split-interview format. The interviews are designed to maximize consistency between reported information in January and May. Many processing decisions also reflect the treatment of the two interviews as one.

Persons under 15 are not interviewed in SLID, but some information is collected on them from other household members. When young household members turn 16 years of age, they become eligible for the preliminary interview and the labour and income questions.

SLID processing can be described as being composed of the following steps: preprocessing, loading, coding, editing and imputation.

Preprocessing. This is a set of primarily manual steps. It involves the removal of duplicate data for the same person resulting from interviewer errors and transmission difficulties; ensuring that changes in household composition and move dates are consistent; reviewing interviewer notes which may lead to data corrections.

Loading. This is the process of moving and converting the collected question responses into the variables in the internal master database. Due to complex question flows and the many derived variables, this represents a large component of the processing.

Coding. Several fields collect textual information that must be translated into a numeric code before being useful for analysis. The assignment of these codes is a mixture of automated and manual procedures. The codes are then "loaded" into the database.

Editing. Compared with traditional data collection methods, the use of computer-assisted interviewing (CAI) allows much of the editing to be done during collection. These include:

sequence or question flow

checks for consistency with what was said in the same interview or an earlier interview

ranges which may themselves be calculated by the computer to take into account information reported during the same interview

However, not all editing is completed this way. In general, minimal verification is applied to survey data after the interview. One key element in the editing strategy is that all editing is preliminary until collection for the entire panel is finished. That is, changes may result after further data are collected. Data drawn from income tax files are complete (they do not require imputation) and are generally considered to be good.

Imputation. This is the process of assigning a value to variables which do not have a valid value, due to non-response at collection or to valid values suppressed as a result of data inconsistencies.

Currently, imputation is restricted to the following variables:

date of birth and sex

household relationships

geography of residence

all income components

With the exception of data files, SLID products are available free of charge on the Internet at the following addresses. Alternatively, one can use the menu structure, starting with the Statistics Canada web site "www.statcan.ca".

Survey Overview (basic survey description)

http://www.statcan.ca/english/freepub/75F0011XIE/free.htm

SLID electronic data dictionary (detailed information on all variables in database)

http://www.statcan.ca/english/SLID/diction.htm

SLID User's guide (detailed survey documentation)

http://www.statcan.ca/english/IPS/Data/75M0001GIE.htm

Working paper series (including print versions of questionnaire)

http://www.statcan.ca/cgi-bin/downpub/listpub.cgi?catno=75F0002MIE

Paradoxically, the comprehensive data that make SLID so valuable, also makes it more complex for Statistics Canada to ensure that confidentiality of respondents is maintained. In order to comply with the strict confidentiality provisions of the Canadian Statistics Act, SLID data are made available through various modes of dissemination:

Remote access

Computer programs are written by clients and sent electronically to Statistics Canada, where staff run the program(s) against the data base and apply confidentiality protection measures. If need be, data are suppressed from the output. Survey officers subsequently return results to clients. This is likely the most appropriate type of access for international researchers.

On-premises access

Under contract with Statistics Canada, researchers are given access to the data in one of the Statistics Canada offices across the country, where staff provide access to the data and implement confidentiality procedures.

Research data centres

Starting in Fall 2000, these will open in selected universities across Canada. These centres will act as extensions of Statistics Canada and provide access to researchers while protecting confidentiality.

For more information, contact the SLID Client Services at dynamics@statcan.ca, or by telephone at 613-951-7355 or toll-free at 1-888-297-7355.