4 Household debt from the Survey of Family, Income and Employment

The preceding section has provided an overview of the assets and liabilities of the household sector. However, aggregate data can obscure a wide dispersion of individual circumstances. An exploration of the distribution of liabilities requires data measured at the unit record level. This section begins with an introduction to the Survey of Family, Income and Employment (SoFIE). Following from this, the SoFIE-based estimates of aggregate assets, liabilities and wealth are compared with the RBNZ aggregates that featured in the previous section. Basic descriptive statistics for the distribution of liabilities are then presented, followed by regression models that have been estimated to identify factors associated with debt levels.

4.1 Background to the SoFIE

SoFIE is a longitudinal survey where the original sample members are tracked and surveyed each year. The target population for SoFIE is the usually resident population of New Zealand living in private dwellings. The survey began in October 2002 with an original sample size of about 11,500 households, amounting to over 22,000 individuals 15 and over. Children younger than 15 who were living in households selected for the survey will also be tracked and will be surveyed from age 15. The survey will be run for a total of eight years. The core survey collects information on family characteristics and labour market and income spells. An assets and liabilities module and a health module are included in alternate years. The analysis that follows is based on the first release of the second wave of SoFIE, which covers the year ending 30 September 2004, and contains the first assets and liabilities module. We have carried out our analysis at the economic family unit level. An economic family unit can be either a non-partnered individual or a couple, with or without dependent children.[9]

The advantage of unit record data over aggregate data is that it enables analysis of associations between variables at the unit level. In particular, we have been able to examine the levels and variability of debt, debt servicing and the proportion of families with “high” debt servicing costs across characteristics such as age and income. On the other hand, survey data can suffer from sampling errors and potential biases.

Sampling error can be quantified. Bias is more difficult to deal with and can arise for a number of reasons. The main factor is non-response bias that arises if those who don't respond have different characteristics from the respondents. The response rate for the first wave of SoFIE was 77% (ie, 11,500 of the randomly chosen 15,000 households agreed to participate). Further, attrition bias may enter as people drop out of the survey in subsequent waves (ie, 87% of those responding in wave 1 also responded in wave 2). Missing data within records is another type of non-response bias. To the extent that they can, Statistics New Zealand attempts to adjust for non-response bias by adjusting the weights so that the data match targets for selected demographic characteristics such as age and sex. Imputation is also carried out to fill in missing data for key variables. More information about SoFIE is available on the Statistics New Zealand website.[10]

Our preliminary analysis of the raw assets and liabilities data revealed some inconsistencies and probable errors in the data. In a small number of cases, it was relatively clear what the intended values were and in these cases we applied an edit. This had an immaterial effect on the estimates of total assets and liabilities.[11]