Appendix 4: TaxMod and the HES

This paper draws on estimates calculated with TaxMod (which is a micro-simulation model of New Zealand’s income tax and social assistance systems) [Prebble et al (eds.), 1992, pp. 29-44].[21] TaxMod estimates are also often used when considering potential social assistance reforms.

Survey data (the 2000-01 Household Economic Survey) obtained using TaxMod and administrative data on the recipients of social assistance programmes both have their areas of relative strength. For instance, while administrative data on social assistance programmes contain little information on the large number of people who do not participate in such programmes (but who may nevertheless be affected by a policy change), detailed information on these people can be generated with estimates based on a general survey of households. These estimates may, however, not provide information on the recipients of social assistance programmes as accurately as the administrative data on these programmes (although at times administrative data is limited in the degree to which the characteristics of individuals (e.g., extent of work effort, level of investment income) are recorded).[22] One approach to address these limitations is to draw on a combination of both data sources.[23]

TaxMod calculates income tax liabilities and social assistance entitlement based upon characteristics of the population and rules regarding eligibility and abatement of income tax and social assistance programmes. A population of families is derived from demographic, income, and expenditure data contained in the HES. The HES was established to measure the Consumers’ Price Index and was conducted annually from 1983-4 to 1997-98. The HES is now conducted every three years with the most recent survey being completed in June 2000-01. TaxMod has 12 databases of HES data from 1987-88 through to 2000-01.

The HES collects demographic, income, and expenditure data from approximately 3,000 households over the course of a year [Gordon, 1997]. Although the HES collects a large amount of data from each household and surveys a large number of households by New Zealand standards, the survey is designed to measure CPI accurately and so the sample design does not always provide accurate results on social assistance entitlement. For example, in comparison to administrative data collected by Work and Income New Zealand the HES data consistently contains relatively low numbers of ex-beneficiaries. Thus in TaxMod each surveyed household is given a weighting representing the degree to which households of that type occur in the total population. This technique allows the sample to be weighted up to estimate the entire New Zealand population.

The HES collects information on family composition and other demographic data. The HES records blood relationships between people in each household. TaxMod, however, allocates some people in such relationships into separate families if this is the treatment that applies under income tax and social assistance rules.

The HES collects expenditure data for entire households and income data for individuals within households. In HES each person aged 15 or above is asked to record income from various sources received over the previous 12 months. These various income sources include up to three current jobs, six jobs that ended in the previous 12 months, social assistance payments, interest income, dividends, and other income from sources. Each household is asked to keep an expenditure diary recording all expenditure for two weeks. Each household is also asked to record all items of expenditure over $200 that occurred in the previous twelve months. Both the location and cost of accommodation are recorded. Some expenditure items (particularly alcohol and tobacco consumption) tend to be inaccurately reported.

TaxMod calculates benefit receipt based on data on benefit duration (not benefit income) contained in HES and assumes complete participation (100% take-up) in programmes. TaxMod contains incomplete information on wealth and does not model entities such as companies or trusts. TaxMod does not model people’s behavioural responses to income tax and social assistance programmes. Estimates of the fiscal costs of policy changes are not, therefore, adjusted in the light of any behavioural changes that these policy changes may induce.

When modelling the effects of a policy change it is important to recognise the limitations of estimates of outcomes and the risks that consequently apply when using these estimates. This modelling risk was illustrated in the case of the Parental Tax Credit, where estimates of the cost of the Parental Tax Credit prepared during its establishment differed significantly from the programme’s actual expenditure due to the complexity of the programme’s entitlement criteria and the consequently small number of eligible families contained in the HES survey data used to prepare these estimates. Initial estimates used during the development of the Parental Tax Credit were based on 1995-96 HES data, which contained 66 families potentially eligible for the proposed programme. This sample was scaled up to give an estimate of an eligible population of 26,000. In contrast, later estimates based on 1997-98 HES data contained a sample of 51 potentially eligible families, which was scaled up to give an estimate of an eligible population of 20,000 families.