Imputation rates

View the most recent version.

To compensate for non-responding households in the SLID sample, a non-response adjustment is applied to SLID weights. However, partially responding households are kept in the sample and any income data that is missing for individuals within responding households is imputed. These individuals may require complete imputation of all income variables or they may require only certain fields to be imputed. Imputation rates in SLID may be thought of as a measure of partial non-response in the survey.

Two methods of imputation are used in SLID: Longitudinal Imputation and Crosssectional imputation. Cross-sectional imputation of income variables in SLID is done using a nearest neighbour approach. Longitudinal imputation of income is done by using last wave's income to impute for the current wave income. Some variables are also imputed using a deterministic approach.

For the nearest neighbour method, a set of basic consistency rules is defined and for a given record requiring imputation a set of consistent donors is identified. A set of matching variables, each of which are correlated with the variables to be imputed, is also defined. Through combined use of both a score function (for categorical matching variables) and a distance function (for numeric matching variables), the most similar consistent donor record is identified and used to impute data for the record.

The percentage of persons within responding SLID households that were subject to total or partial imputation is shown in Table 8.1. Recall that a responding SLID household is one in which at least one household member has responded partially or completely to either the labour or income questions of the survey.

In Table 8.2 we compare the percentage of tax data records requiring imputation to the percentage of records for which data is collected through the telephone interview. The need for partial imputation is determined after combining responses to both the labour and income questions. Inconsistencies are corrected through the imputation process.

Table 8.2 also shows the percentage of individuals subject to partial imputation who require between one and seventeen variables to be imputed.

In 2002, new housing content relevant for housing research and policy development was added to SLID in cooperation with the Canada Mortgage and Housing Corporation (CMHC). The survey now collects information for the following sub-populations beginning with the 2002 reference year: the need for repairs (as determined by the dwelling occupant); the principal heating fuel of the dwelling; and whether a farm or home business is operated from the property. Also from homeowners the amount of regular mortgage payments; the amount of annual property taxes; and whether the dwelling is part of a registered condominium is collected. From renters the following is collected: the amount of monthly rent, what amenities are included in the rent (e.g., heat, water, electricity); and whether the rent is subsidised by government or an employer.

The above information is in addition to information about home ownership and type of dwelling (since 1994) and information on the presence of a mortgage and the number of bedrooms in dwellings (since 1999).

Because of non-response to specific questions, imputation of housing related content was introduced in SLID in 2002. Two methods of imputation were used, longitudinal imputation and cross-sectional donor imputation. The cross-sectional donor imputation uses a similar method to that used in the income imputation, making use of the score function described above. Table 8.3 shows the percentage of responding SLID households that were subject to total or partial imputation.

In total twenty housing variables are imputed during SLID housing imputation. Many households require only partial imputation. Table 8.4 shows the break down of those requiring partial imputation.

Table 8.1 Income-variable imputation for respondents in 2006 (%)

Table 8.2 Breakdown of partial or total imputation in 2006 (%)

Table 8.3 Household-variable imputation in 2006 (%)

Table 8.4 Households requiring partial imputation (%)