Wealth Variables and Wealth Imputation (Wave 2 and 6 Special Topic)

Wealth Model

In wave 2 a special wealth module was incorporated into the questionnaires. The Household Questionnaire contained the majority of the wealth questions and we endeavoured to ask these of the person knowing the most about the household finances. These questions covered the following topics:

• Cash and equity investments, trust funds, life insurance;
• Home and other property assets and debts;
• Business assets and debts;
• Children’s bank accounts¹⁸;
• Collectables and vehicles, and
• Overdue household bills (in wave 6 only)¹⁹

Also, each respondent was asked some questions about their personal wealth in the Person Questionnaire, including:

• Bank accounts and credit card debt;
• Superannuation;
• HECS debt: and
• Other personal debts²⁰

Figure 22 shows how the wealth components are combined together to form the total household wealth. The boxes with the broken lines highlight the variables that come from the Person Questionnaires. From Release 6, the imputation for non–respondents has been conducted at the wealth component level, so the household–level components are the sum of all persons in the household.²¹

Figure 22: Wealth Model Diagram, Household–level

Several equity variables (assets less debts) not described in the previous figure are provided on the household file. These are business equity, home equity, other property equity, and total property equity.

These variables, together with the unimputed versions of the sub–totals described in Figure 22 are provided in Table 32 (variables relating directly to the wealth imputation are provided later in Table 35).

Table 32: Other derived variables at household–level

Imputation Method

With the repeat of the wealth questions in wave 6, we can now impute the wealth variables via a longitudinal method.²²

Person–level wealth variables are imputed using the Little and Su method. The nearest neighbour regression method²³ is used as a fall–back method when the Little and Su method does not work. Wealth variables for non–respondents have been imputed at the wealth component level (rather than for total assets and total debts as was done in Release 2 through to 5) so that the household wealth can be provided at the component level by summing the person–level wealth of the adults in the household.

Household–level wealth variables are also imputed using the Little and Su method, where the household can be longitudinally linked²⁴ between waves 2 and 6 (bhwlink is an indicator variable for whether the wave 2 household could be linked to a wave 6 household). Otherwise the nearest neighbour regression method is used.

The household–level wealth variable for home value was collected in all waves of the survey. This variable was imputed through waves 1 to 6 via the same method used in the two wave wealth module approach. Households were longitudinally linked from one wave to another (_hhlink is an indicator variable showing whether a household was linked to another household in the next wave) with the Little and Su method applied wherever possible. The nearest neighbor regression method was used as a fall–back method again.

Where we knew the wealth component was non–zero (for example, a screener question identified them as having the particular wealth component), only a non–zero donor was permitted. For situations where it was unknown if the individual or household had the asset, a filter process²⁵ was used to determine if they received a non–zero value or not.

Table 33 and Table 34 show the proportion of missing cases that were imputed by each imputation method.²⁶ In the first table the proportions are summarized across all wealth variables that have been imputed. As with income it is preferable to have all records imputed by the Little and Su method but with only two waves of wealth data sufficient information was not always available. Non respondents in the enumerated person group were less likely to be imputed by the Little and Su method (for similar reasons as in income imputation) and any households not linked between the two waves, regardless of their situation, were imputed via the nearest neighbor regression method. Table 34 shows a much higher proportion of imputed records via the Little and Su method for home value due to better household linking between consecutive waves (rather than the four–wave difference experienced with the wealth module imputation).

Table 33: Proportion of missing cases imputed by imputation method (wealth), waves 2 and 6

Table 34: Proportion of missing cases imputed by imputation method (home value), waves 1 – 6

Imputed Wealth Variables

Table 35 outlines the imputed wealth variables included on the wave 2 and 6 files.

Table 35: Imputed wealth variables

Endnotes: