Footnotes

¹. SPSS portable files provided on previous releases are no longer provided on the CD but can be obtained by special request.

². You will need to use STATA SE as there are more than 2047 variables in the datasets.

³. A trial copy of StatTransfer Version 8 can be downloaded from www.stattransfer.com or purchased online at www.stattransfer.com/html/store.html. DBMS/Copy Version 8 can be purchased online at www.dataflux.com/product-services/products/dbms.asp.

⁴. Users of the In-confidence Release (unconfidentialised) files can alternatively use _hhid to match the household and person files, and _hhpid to match the person files. In wave 1, the household identifier is six digits long, corresponding to area (three digits), dwelling number (two digits) and household number (one digit). The person identifier in wave 1 is then eight digits long – the first six are the household identifier, followed by two digits for the person number. In subsequent waves, the household identifier is five digits long, and the person identifier is seven digits long.

⁵. Note that this definition of a dependent student is different to the full-time student identifier provided on the Responding Person File.

⁶. The Maternity Allowance is allocated to all families with newborn children and included in the Australian pensions and benefits.

 ⁷. This is a change for Release 4.0. In earlier Releases, the Child Care Benefit was included in the total financial year income and the Maternity Allowance was only recorded if the respondent reported it.

 ⁸. Age groups were used to create the imputation classes because it is a simple characteristic and it is known for almost all donors and recipients. For a few cases, age was missing and was therefore imputed from a person with a similar relationship structure to the missing case. Not all income variables were imputed using imputation classes. The variables where donors and recipients were matched with imputation classes were current wages and salaries, current benefits, financial year wages and salaries, Australian Government pensions and rent income.

^8A. Zeros are carried forward or backward from the surrounding waves with the same probability as that observed in the complete cases.

 ⁹. For the number and proportion of cases which are missing, see Table 39.

^9A. A detailed review of the performance of the imputation methods is reported in Starick and Watson (2007).

 ¹⁰. That is, bank accounts of people in the household aged under 15.

^10A. Overdue household bills were explicitly asked for in wave 6. It was assumed that this was captured in the ‘any other debt’ question asked in wave 2 (though perhaps not well).

^10B. In wave 6, these other personal debts were asked for at a more disaggregated level and overdue personal bills were also explicitly asked for.

^10C. For Release 2 to 5, please note that the imputation for non-respondents was only conducted at the total assets and debts level. As a result the household-level components that summed these person-level components was just the sum of responding persons only. This will explain some of the differences observed for these variables between releases.

^10D. For Release 2 through to 5, the wealth imputation for wave 2 used a nearest neighbour regression method (see Watson, 2004a for more details).

^10E. The reported value of a donor with the nearest predicted value from a regression model to the recipient’s is used to replace the missing value.

^10F. Households have been linked over time for the purposes of imputing wealth if they have the same adult (aged 18+) membership. Deaths are not counted as a membership change.

^10G. The nearest neighbor regression method was run prior to the normal imputation step. All unknown records who were assigned a donor with a zero value were considered not to have the wealth item. All other records were imputed with a non–zero amount and the Little and Su method was used where possible to determine the non-zero amount.

^10H. For the proportion of cases which are missing, see Table 41 and Table 42.

 ¹¹. For example, the number of people living in a household with two people. This can be derived from the household file by estimating the number of two person households and multiplying by two. Alternatively, it can be estimated from the enumerated file by summing the weights of people living in two person households.

¹². Prior to Release 4, the household benchmarks were number of adults by number of children by broad geography and State by part of State (the bolded text indicates what has been dropped).

¹³. The Wave 2 benchmarks do not include labour force status as this was not collected on the Household Form that wave.

¹⁴.Prior to Release 4, the enumerated person benchmarks were State by part of State by sex by broad age, and State by part of State by labour force status (the bolded text indicates what has been dropped, but note that State by part of State is now included as a separate benchmark). The marital status benchmark has been added from Release 4.

^14A. Prior to Release 6, the responding person benchmarks were State by part of State by sex by broad age; State by part of State by labour force status; marital status by broad age; and occupation by broad geography (the bolded text indicates what has been dropped, but note that State by part of State is now included as a separate benchmark). From Release 4 the marital status and occupation benchmarks have been included.

¹⁵. No benchmarks for the responding person weights have been dropped, but the marital status and occupation benchmarks have been included.

¹⁶. Prior to Release 5, only household estimates based on the 1996 Census were available.

¹⁷. Prior to Release 5 only the sparsely settled parts of the Norther Territory were excluded.

¹⁸. While this paper is written in relation to the wave 2 weighting, the process in later waves follows the same methodology.

¹⁹. Prior to Release 4, the longitudinal enumerated person benchmarks were State by part of State by sex by age broad, and State by part of State by labour force status (the bolded text indicates what has been dropped, but note that State by part of State is now included as a separate benchmark). The marital status benchmark has been added from Release 4.

^19A. Prior to Release 6, the longitudinal responding person benchmarks were State by part of State by sex by broad age; State by part of State by labour force status; marital status by broad age; and occupation by broad geography (the bolded text indicates what has been dropped, but note that State by part of State is now included as a separate benchmark). From Release 4 the marital status and occupation benchmarks have been included.

^19B. Prior to Release 6, weights were only provided for the balanced panel of respondents or enumerated persons from wave 1 to every other wave.

²⁰. No benchmarks for the longitudinal responding person weights have been dropped, but the marital status and occupation benchmarks have been included.

²¹. In Adobe Acrobat, you would begin a search by clicking on the button that looks like this: .

^21A. A wealth band is two values which the respondent believes their actual value to be within. The bands differ between some variables.

²². See the section on Following Rules for more information about who is temporarily and permanently added to the sample.

²³. This approach is consistent with the British Household Panel Study (BHPS), with the difference that in the BHPS only people aged 16 years and over are interviewed. The lower age chosen for the HILDA Survey simply reflects our desire to conform to Australian Bureau of Statistics (ABS) standards adopted in its Labour Force Survey.

²⁴. ABS, Labour Statistics: Concepts, Sources and Methods (ABS Cat. No. 6102.0), ABS, Canberra , 2001.

²⁵. ABS, Statistical Concepts Library (ABS Cat. No. 1361.30.001), ABS, Canberra , 2000.

²⁶. Note that if a child CSM moves without any other adult CSMs, they are followed to their new household and the eligible members of that household are then interviewed

²⁷. Six or more calls were made to all selected households until a final household outcome was achieved. These calls were made over a minimum of a five-day period, with typically three calls on weekdays and at least three calls on weekends.

²⁸. For details on the tracking procedures adopted, see Watson and Wooden (2004b).

²⁹. When initially making contact with a household, the interviewer had up to six calls to make contact and a further six calls to undertake all of the interviews once contact had been made. If a household had to be put into tracking and was found, the initial call allocation to make contact with the household was carried over to the next stage of the fieldwork. When following up a household, the interviewer had a total of five calls to finalise the household.

³⁰. See Watson and Wooden (2002b) for details of these monitoring methods.

³¹. Only responding households in wave 1 were issued in wave 2, so the closest comparison in the household response rate to be made in later waves is for households resonding in the previous wave.