Appendix 2: Details on the data, measures and weights
Data
The data used in this paper are from the first and second waves of HILDA, a longitudinal survey of households focusing on the interactions between the labour market, families and social welfare. The survey commenced in 2001 with a two-stage probability sample. In the first stage, 488 Census Collection Districts (CD), based on 1996 Census boundaries, were randomly selected. Within each CD, all households (approximately 200 to 250) were enumerated and 22 to 34 dwellings randomly selected.18 An adult representative of the household was asked to answer questions on a household questionnaire about the household. Interviews were obtained from 7,682 households, or 66 per cent of all households identified as 'in scope'. For the household grid basic information was collected (age, sex and relationships between household members) from all 19,914 enumerated household members. Personal interviews were attempted with the 15,127 household members aged 15 years and over. Person questionnaires were completed for 13,969 household members, a response rate of 92 per cent.19 Respondents were also asked to fill in a self-completion questionnaire, which included questions on financial stress, subjective prosperity, and spending and saving behaviour. Of the 13,969 individuals who responded to the person questionnaire, 13,058 (or 93.5 per cent) provided useable data for the self-completion questionnaire.
For this and subsequent waves, three data files were created: a household data file derived from the household questionnaire, a responding person file derived from the person questionnaire and the self-completion questionnaire, and an enumerated person data file derived from the household grid. The survey instruments can be downloaded from the Internet.20
In 2002, all responding households from Wave 1 were contacted again. Sixty-nine households were out of scope due to deaths or moves overseas, and there were 713 new households arising from changes in household composition.21 Thus, a total of 8,326 households were in scope for Wave 2. Interviews with the household questionnaire were obtained from 7,245 households, a response rate of 87 per cent. Interviews were again sought with all household members aged 15 or over, including people who did not respond in Wave 1, as well as new household members. In total, person questionnaires were completed for 13,041 individuals. Of this group, almost 12,000 were respondents from Wave 1, which represented almost 87 per cent of the Wave 1 individual sample.22 A slightly lower response rate was obtained for the self-completion questionnaire at around 90 per cent. Of the 13,041 respondents who were interviewed, 11,691 completed and returned self-completion questionnaires (Watson & Wooden 2004).
Income poverty
The measure of income poverty is based on disposable household income. Household income is the annual income from wages and salaries, self-employment, investments, superannuation and government benefits for all household members. Disposable income is the income after taxes (federal tax and the Medicare levy) and government transfers. These amounts were imputed from gross income. For more details, see Headey (2003).
Deductions were not made for:
- employer, employee or private superannuation or insurance contributions
- land taxes and rates
- health insurance.
Table A1 presents the means, medians, 25th and 75th percentiles for annual gross income, disposable income and disposable income after deducting housing costs. For these measures, missing incomes are imputed (see Watson 2004a).
| Measure | Mean | Median | Bottom quartile |
Top quartile |
|---|---|---|---|---|
| Wave 1 | ||||
| Annual household income | 57,298 | 46,060 | 22,574 | 77,588 |
| Annual disposable household income | 44,490 | 37,657 | 20,521 | 59,804 |
| Annual disposable household income after housing costs | 38,284 | 31,120 | 16,406 | 52,005 |
| Wave 2 | ||||
| Annual household income | 58,943 | 47,500 | 23,500 | 79,758 |
| Annual disposable household income | 45,608 | 38,451 | 21,210 | 61,263 |
| Annual disposable household income after housing costs | 39,448 | 32,198 | 17,264 | 52,852 |
For comparison, the ABS estimate for gross household income in SIHC was $972 per week or $50,544 per year. The estimate for median income was $773 per week or $40,196 per year (ABS 2000–01; 2004b). Although the HILDA and ABS surveys are for different years and there are technical differences in their estimation procedures, the estimates they give are not too dissimilar. For example, Watson (2004a, pp. 9–10) estimates average household income (not including Family Tax Benefit A and B and Child Care Benefit) at $54,689 per year for HILDA Wave 1 and $57,810 per year for HILDA Wave 2.
Housing costs were defined as the sum of the costs of home mortgages and rent. For 245 households in Wave 1 and 39 households in Wave 2, data for missing housing costs were imputed by the nearest neighbour method. A regression model of housing costs based on household income, tenure and household type was estimated and the predicted values were used to impute housing costs for these cases. The estimates are presented in Table A2.
| Measure | Wave 1 | Wave 2 |
|---|---|---|
| Proportion owning/paying off | 68.0 | 67.7 |
| Proportion renting | 29.5 | 29.2 |
| Proportion other tenure | 2.5 | 3.2 |
| Proportion of owners with no first mortgage | 59.4 | 59.7 |
| Proportion of owners with no second mortgage | 94.0 | 92.6 |
| Mean annual repayment, first mortgage (of those with outstanding mortgage) | $11,635 | $12,071 |
| Mean annual repayment, second mortgage (of those with outstanding mortgage) | $11,694 | $12,386 |
| Mean annual rent (of renters) | $8,134 | $8,240 |
| Mean annual housing costs (all households) | $5,956 | $5,981 |
Table A3 presents the correlations for individual and household income measures from the first two waves of HILDA. Generally, the correlations are between 0.50 and 0.65, indicating substantial year-to-year movement in incomes. Watson (2004a, p. 17) presents Wave 1 and 2 correlations for imputed and non-imputed wages and salaries, and benefit incomes.
| Measure | Correlation |
|---|---|
| Individual gross income | 0.60 |
| Individual disposable income | 0.61 |
| Annual household income | 0.61 |
| Annual disposable household income | 0.60 |
| Annual disposable household income after housing costs | 0.56 |
| Equivalised household income | 0.60 |
| Equivalised disposable household income | 0.57 |
| Equivalised disposable household income after housing costs | 0.52 |
Table A4 presents the mean and median equivalised household incomes for Waves 1 and 2. For comparison, the ABS estimate for mean equivalised disposable household income in 2000–2001 was $469 per week or $24,388 per year. The ABS used the same equivalence scale as used here. The ABS estimate for median equivalised disposable household income was $414 per week or $21,528 per year (ABS 2000–01; 2004b). Again, the estimates from HILDA are not too dissimilar from the ABS estimates.
| Measure | Mean | Median | 50% of median |
|---|---|---|---|
| Wave 1 | |||
| Equivalised household income | 33,609 | 27,467 | 13,734 |
| Equivalised disposable household income | 26,227 | 22,350 | 11,175 |
| Equivalised disposable household income after housing costs | 22,363 | 18,924 | 9,462 |
| Wave 2 | |||
| Equivalised household income | 35,282 | 29,412 | 14,706 |
| Equivalised disposable household income | 27,438 | 23,850 | 11,925 |
| Equivalised disposable household income after housing costs | 23,545 | 20,224 | 10,112 |
The last column of Table A4 shows the half-median equivalised incomes. These estimates are for single-person households. Therefore, in Wave 1, the before-housing equivalised disposable income poverty line was drawn at $11,175 and the after-housing poverty line at $9,462. For Wave 2, the poverty lines are slightly higher at $11,925 and $10,112. Note that if the poverty lines were drawn at 40 or 60 per cent of median income they would differ only slightly. The 40, 50 and 60 per cent after-housing poverty lines for Wave 1 are $7,570, $9,462 and $11,354. However, these poverty lines generate quite different estimates of the proportions in poverty.
The half-median poverty lines are higher for larger households. On the modified OECD equivalence scale, the poverty line for a couple is 1.5 times that for a single person; for a sole parent, 1.6 times; and for a couple with two children, 2.1 times. Table A5 presents the poverty lines for a selection of household structures calculated using the modified OECD equivalence scale. For example, for a family of two adults and two children, the cut-off line for before-housing income poverty was at $23,467.
| Household structure | Wave 1 |
Wave 2 | ||
|---|---|---|---|---|
| Before housing | After housing | Before housing | After housing | |
| Couple with no children | 16,762 | 14,193 | 17,887 | 15,168 |
| Couple with one child | 20,115 | 17,031 | 21,465 | 18,201 |
| Couple with two children | 23,467 | 19,870 | 25,042 | 21,235 |
| Couple with three children | 26,820 | 22,708 | 28,620 | 24,268 |
| Single adult | 11,175 | 9,462 | 11,925 | 10,112 |
| Single adult with one child | 14,527 | 12,300 | 15,502 | 13,145 |
| Single adult with two children | 17,880 | 15,139 | 19,080 | 16,179 |
| Single adult with three children | 21,232 | 17,977 | 22,657 | 19,212 |
Subjective poverty
The measure of subjective poverty was based on the following question in the HILDA responding person questionnaire:
C1: Given your current needs and financial responsibilities, would you say you and your family are:
- Prosperous
- Very comfortable
- Reasonably comfortable
- Just getting along
- Poor
- Very poor
Table A6 presents estimates of the response frequencies to the question on standard of living. Very few indicated they were prosperous, whereas over 60 per cent said they were very comfortable or reasonably comfortable. Less than 4 per cent indicated they were poor and a further 1 per cent indicated they were very poor. The frequency distributions for the two waves are very similar. If continuous variables are constructed from the responses to this question (ranging from a score of 1 for 'prosperous' to a score of 6 for 'very poor'), the mean scores for both years were 3.2.
| Measure | Wave 1 | Wave 2 |
|---|---|---|
| Prosperous | 1.6 | 1.2 |
| Very comfortable | 12.4 | 12.9 |
| Reasonably comfortable | 51.0 | 50.9 |
| Just getting along | 30.6 | 31.1 |
| Poor | 3.6 | 3.3 |
| Very poor | 0.7 | 0.7 |
Financial stress
For this paper, financial stress was measured using question C2 in the HILDA self-completion questionnaire. The question was designed to elicit information from respondents about their cash flow problems. It was asked in an identical form in both waves, and was worded as follows:
Since January (Year), did any of the following happen to you because of a shortage of money?
| a) Could not pay gas, electricity or telephone bills on time | Yes | No |
| b) Could not pay mortgage or rent on time | Yes | No |
| c) Pawned or sold something | Yes | No |
| d) Went without meals | Yes | No |
| e) Was unable to heat home | Yes | No |
| f) Asked for financial help from friends or family | Yes | No |
| g) Asked for help from welfare/community organisations | Yes | No |
This question is very similar to the one used in HES. The difference is the replacement of the item on car registration by an item on mortgage or rent. The timeframe for the HILDA question is shorter—since the beginning of the year rather than in the last 12 months.
These items were re-coded with a score of one if the respondent answered 'yes' to the item and zero if 'no'. Respondents who completed the self-completion questionnaire, but did not provide a valid response to the item, were also assigned a score of zero. Data for respondents who did not return a useable self-completion questionnaire were declared missing.
Table A7 presents the distributions of financial stress items for the first two waves of HILDA. In Wave 1, 19 per cent did not pay utility or telephone bills on time, 17 per cent asked for financial help from friends and family, 9 per cent did not pay their rent or mortgage on time, 9 per cent pawned or sold something because of a shortage of money, 5 per cent went without meals, 4 per cent were unable to heat their home and 5 per cent sought help from welfare or community groups. In Wave 2, the results were about 2 percentage points lower. The sample includes more than one respondent in many households; however, the frequency distribution changes only marginally if the sample is restricted to one adult randomly selected from each household (Table A8).
The incidences of cash flow problems are higher in HILDA than in other surveys (Table A9). The figures for ASLS are not really comparable due to differences in question wording and the longer (two-year) timeframe. Even so, they indicate higher levels of financial stress; 16 per cent of ASLS respondents had trouble paying utility bills on time in the previous two years compared to over 20 per cent of HILDA respondents since the beginning of the year.
| Measure | Wave 1 |
Wave 2 | ||
|---|---|---|---|---|
| n | % | n | % | |
| Could not pay electricity, gas or telephone bills on time | 2,463 | 18.9 | 1,902 | 16.4 |
| Could not pay mortgage or rent on time | 1,151 | 8.9 | 891 | 7.7 |
| Pawned or sold something | 850 | 8.8 | 608 | 5.2 |
| Went without meals | 603 | 4.6 | 448 | 3.9 |
| Was unable to heat home | 482 | 3.7 | 369 | 3.2 |
| Asked for financial help from friends or family | 2,151 | 16.5 | 1,579 | 13.6 |
| Asked for help from welfare/community organisations | 687 | 5.3 | 451 | 3.8 |
| Measure | Wave 1 |
Wave 2 | ||
|---|---|---|---|---|
| n | % | n | % | |
| Could not pay electricity, gas or telephone bills on time | 1,461 | 20.7 | 1,188 | 18.6 |
| Could not pay mortgage or rent on time | 682 | 9.6 | 564 | 8.9 |
| Pawned or sold something | 505 | 7.1 | 373 | 5.9 |
| Went without meals | 396 | 5.6 | 278 | 4.3 |
| Was unable to heat home | 310 | 4.4 | 243 | 3.8 |
| Asked for financial help from friends or family | 1,187 | 16.8 | 893 | 14.0 |
| Asked for help from welfare or community organisations | 396 | 5.6 | 276 | 4.3 |
| Measure | ASLS (1986) | HES (1998–99) | GSS 2002 |
|---|---|---|---|
| Could not pay utility bills on time | 16 | 16 | 13 |
| Could not pay mortgage or rent on time | - | - | 5 |
| Pawned or sold something | - | 4 | 3 |
| Went without meals | - | 3 | 2 |
| Was unable to heat home | - | 2 | 1 |
| Asked for financial help from friends or family | 19 | 10 | 8 |
| Asked for help from welfare or community organisations | 3 | 3 | 3 |
Similarly, the proportion seeking financial help from friends and family is the same in HILDA since the beginning of the year as in ASLS over the previous two years. Comparisons with HES and GSS also show that HILDA yields higher levels of financial stress.
Differences in data collection modes are the most likely explanation for higher levels of financial stress in HILDA than other surveys examined (HES and GSS). The HES and GSS surveys were conducted by personal interview, whereas information on financial stress in HILDA was obtained from self-completion questionnaires. It is plausible that respondents are less likely to admit cash flow problems in face-to-face interviews than in self-completion questionnaires.
Table A10 investigates whether the financial stress items all relate to the same underlying or latent concept. These analyses serve only as a guide since such analyses are not usually performed on dichotomous variables. Factor analysis is used to determine whether a group of items relate to a single underlying concept (or dimension) or relate to two or more underlying concepts. It is based on the pattern of responses: if a group of items elicit similar responses, they are likely to be tapping the same underlying concept.
| Measure | Wave 1 |
Wave 2 | ||||
|---|---|---|---|---|---|---|
| Mean | Correlation | Load | Mean | Correlation | Load | |
| Could not pay electricity, gas or telephone bills on time | 0.21 | 0.56 | 0.65 | 0.16 | 0.56 | 0.64 |
| Could not pay mortgage or rent on time | 0.17 | 0.52 | 0.61 | 0.08 | 0.49 | 0.56 |
| Pawned or sold something | 0.07 | 0.45 | 0.54 | 0.05 | 0.43 | 0.50 |
| Went without meals | 0.05 | 0.46 | 0.57 | 0.04 | 0.49 | 0.57 |
| Was unable to heat home | 0.05 | 0.39 | 0.46 | 0.03 | 0.37 | 0.45 |
| Asked for financial help from friends or family | 0.17 | 0.54 | 0.60 | 0.14 | 0.54 | 0.60 |
| Asked for help from welfare or community organisations | 0.06 | 0.41 | 0.50 | 0.04 | 0.42 | 0.49 |
| Cronbach’s alpha | - | - | 0.74 | - | - | 0.74 |
Factor analysis showed that the items loaded a single factor. Each item had a factor loading of over 0.4, which can be understood as the correlation between the item and the latent factor. Loadings over 0.4 are usually considered as part of the latent construct. The item on not paying utility bills on time had the highest loading. Correlation analyses, which estimate the correlation of each item with the sum of the other variables, also indicated substantial interrelationships between the financial stress items. Cronbach's alpha, a statistic that indicates the consistency with which respondents answer the items, was 0.74 in both waves.23 Because of the high inter-item correlations, there is no need to discard items or to discard the non-core items and focus only on the five core items identified by Saunders (2004).
Table A11 presents the number of incidences of financial stress reported by households. In Wave 1, 70 per cent had not experienced any financial stress, 13 per cent answered 'yes' to one of the seven financial stress items, 8 per cent two items, 5 per cent three items and a further 5 per cent four or more items. In Wave 2, 75 per cent had not experienced any financial stress. Across both waves, 63 per cent of households reported no incidences of financial stress, 11 per cent one incidence and 8 per cent two incidences. Direct comparisons with the GSS data are not possible since financial stress in that survey was measured by eight rather than seven items. Even so, financial stress is higher in the HILDA data with lower proportions indicating no stress.
| Number of incidences | Wave 1 |
Wave 2 | ||
|---|---|---|---|---|
| n | % | n | % | |
| 0 | 9,185 | 70.3 | 8,700 | 74.8 |
| 1 | 1,645 | 12.6 | 1,234 | 10.6 |
| 2 | 999 | 7.7 | 848 | 7.3 |
| 3 | 625 | 4.8 | 432 | 3.7 |
| 4 | 315 | 2.4 | 220 | 1.9 |
| 5 | 171 | 1.3 | 111 | 1.0 |
| 6 | 72 | 0.5 | 50 | 0.4 |
| 7 | 46 | 0.4 | 41 | 0.4 |
| Total | 13,058 | 11,636 | ||
Weights
Several weights were constructed for each wave of HILDA, including cross-sectional weights for households; enumerated person and responding person weights; longitudinal enumerated person and longitudinal responding person weights; and weights adjusted to either the sample or population size.
The household weights for Wave 1 were constructed in three steps. The design weights adjusted for the probability of selection of households into the sample. The second step was to model response/non-response probabilities by a number of household and neighbourhood characteristics. The inverses of the probabilities of response were included in the weights, so that responding units with characteristics associated with non-response received larger weights. The third step was to adjust for differences in the distributions of benchmark variables between the sample and the population. The final household weights provided the basis for the enumerated person and responding person weights. The responding person weight was further adjusted by using information on responding and non-responding persons. The enumerated person weights were adjusted in relation to state or territory, region, sex and age. The responding person weights were adjusted by these variables as well as by labour force status. Further details on the construction of the Wave 1 household and person weights can be found in Watson and Fry (2002).
The weights for Wave 2 were based on the Wave 1 weights.
The Wave 2 household weights adjusted for:
- changes in selection probabilities resulting from new household entrants to the survey
- characteristics relating to Wave 2 response/non-response among Wave 1 responding households
- differences in the distributions of benchmark variables in the sample and population using ABS estimates for 30 September 2002.
The Wave 2 person weights adjusted for:
- person and household characteristics relating to Wave 2 response/non-response among Wave 1 respondents.
- differences in the distributions of benchmark variables in the sample and population using ABS estimates for 30 September 2002.
The enumerated person weights were adjusted in relation to state or territory, region, sex and age. The responding person weights were adjusted by these variables as well as by labour force status.
See Watson (2004b) for further details on the construction of Wave 2 weights.