2.1 Estimating the direct costs of children
The methodology used in this study to estimate the costs of children generally follows that developed by Espenshade (1984) and Betson (1990). Following Engel, they estimated the costs of parental expenditures on children in the United States (1984). This methodology was subsequently used by Lee (1988) to provide estimates of the costs of children in Australia in 1984.
Broadly, the methodology estimates the cost of a child as the difference in average expenditures between households, with similar numbers of adults, but where some have no children and some have one or more children present, given that the households enjoy an equivalent standard of living. A key question is therefore how to define an ‘equivalent standard of living’?
As Espenshade points out (1984, p. 19) the central problem in estimating the cost borne by parents in raising children is that it is difficult to separate the costs of each family member from the total costs of the household. In many instances these costs cannot be easily assigned to particular individuals (e.g., housing costs or electricity costs). As well, many expenditure items that are directly for children are simply recorded as expenditure items by their parents (e.g., food costs).
One solution to this problem is to use an index of the material standard of living of different families. Such an index allows families with the same standard of living to be compared, even though their family composition and incomes may differ. The problem, of course, is how to determine when families have a similar standard of living — i.e. how to construct such an index. Simply comparing incomes is not sufficient, as families with the same income may have widely differing demands placed on that income and, hence, differences in their standards of living.
Once it is possible to tell if differently constituted families have the same standard of living, it is reasonably straightforward to estimate the costs of their children. Suppose we follow Engel and assume, for example, that the proportion of total expenditure devoted to food is an adequate indicator of the standard of living of a family. Then consider two families, the first a couple with no children and a weekly expenditure of $500 and the second a couple with one child and a weekly expenditure of $600. If both spend 15 per cent of their total expenditure on food then the difference in their expenditures can be said to equal the cost of the child in the second family — i.e., $100 per week.
For the purposes of the study this comparison was generalised in the form of two equations. The first was used to estimate total household consumption expenditure, given information on parental incomes and the number (and ages) of the children and the number of adults. The second estimated household living standards, given information on household consumption expenditure and the number (and ages) of the children and the number of adults (see Appendix 1 for more details).
The methodology used in this study to calculate the cost of children relies on being able to determine when families of different sizes and with different incomes have similar living standards. In his earlier work on costs of children Espenshade examined several different methods that had been used to determine family living standards (1972, pp. 63-74). These included per capita income, level of adult expenditure, proportion of income saved and the proportion of income spent on food. After considering each, Espenshade concluded that the most appropriate measure to use was the proportion of family income spent on food (the so-called 'Engel estimator').
However, using the proportion of total expenditure devoted to food as the indicator of living standards has been the subject of extensive criticism. In particular, Deaton and Muellbauer (1986) argued that the Engel method is fundamentally incorrect in assuming that food share indicates the welfare level of households of different sizes (p. 741). They also argued that it can be theoretically shown that the Engel method will produce estimates higher than the 'true' cost of children.
Since the publication of the paper by Deaton and Muellbauer, the Engel approach has often been described as establishing an upper bound to child cost estimates. However, there are grounds for viewing this widely held assertion with some caution. In particular, as Bradbury points out, the argument that the Engel approach establishes an upper bound to child costs relies on the assumption that food constitutes a relatively high proportion of total child expenditures. However, as this would more typically be the case in developing countries (which are the focus of Deaton and Muellbauer’s analysis), in more developed countries such as Australia such a conclusion is less certain (Bradbury 1994, p. 2).
What is perhaps more likely to be correct is that an Engel estimator will overestimate the costs of larger families. This is because there are likely to be greater economies of scale present in a household's non-food consumption than there is in their food consumption.
A number of other indexes of standard of living have been proposed, including the level of total expenditure devoted to ‘adult goods’ (Rothbarth 1943). However, Bradbury has shown that, theoretically, this approach will most likely overestimate the cost of children (Bradbury 1997, p. 76)
Given this uncertainty in the literature, this study has used the general Espenshade approach. However, the estimator of comparable living standards used was a variation on the Engel method, with the food-at-home share estimator being expanded to include other basic expenditure items (such an estimator is sometimes referred to as an ISO-PROP estimator, following Watts 1977).
In this study, our indicator of the standard of living of a family is the proportion of total expenditure devoted to:
- food at home;
- fuel and power;
- household non-durables for use inside the home (eg. disposable nappies) ;
- postal, telephone and telegram charges; and
- personal care products and services (eg. shampoo).
Not included in either basket were the costs of housing, clothing and health. While these have often been included by other researchers when compiling a basket of basic goods (see, for example, Betson 1990; Merz and Faik 1992), their use is problematic.
Housing was not included as it has characteristics that set it apart from other essential expenditures. As a recent study (which did not include the costs of housing in any of several baskets of basic goods) noted, expenditure on housing is 'peculiar' (Carlucci and Zelli, 1998). What makes it peculiar is that it can differ markedly between households that are similar in every other aspect other than their tenure type. As well, in contrast to other 'basket ' items, the housing costs of many families, particularly those with children, tend to diminish across the life-cycle. That is, families with older children are much more likely to have lower relative housing costs, given that expenditure by home purchasers will, over time, be reduced by the effects of inflation and, as well, will increasingly go to paying off the capital component of their mortgage. For these reasons, it would seem that the inclusion of housing in the measure of living standards is likely to introduce significant distortions.
Clothing was excluded from the basic basket of goods because, in contrast to other basic expenditures, its share of total expenditure rose as household expenditure increased. This mirrors the finding of Carlucci and Zelli (1998) who categorised clothing as a 'comfort' good, rather than a 'necessity' good. (Some clothing is, of course, a necessity. However, it is also in many instances very much a luxury item. Unfortunately, when compared to food, it is more difficult to separate out what is basic expenditure on clothing from what is luxury expenditure. There would also appear to be more scope for wealthier households to increase their expenditure on clothing than there is to increase their expenditure on food, particularly that consumed at home.)
Finally, household expenditure on health was not included in the basic basket of goods, given the role played by the Medicare system in Australia to meet all or most of the basic components of these expenses.
Thus, to summarise, our indicator of the ‘standard of living’ was the proportion of total current expenditure devoted to a typical basket of goods, including food, personal care items, power costs and telephone charges. Where a couple only household and a couple household with two children had the same ‘standard of living’ using this measure, then the difference in the total spending of the two households was assumed to be the cost of the two children. This assessment of the ‘standard of living’ indicator should thus not be confused with the costs of the children. That is, we have not only estimated the food, power, and telephone costs associated with the children, but the total costs associated with the children. It is only when attempting to derive the comparable ‘standard of living’ indicator that we use the restricted basket of expenditure items outlined above.
[ top ]
2.2 Data source
The data source used in this study was the publicly released 1998-99 Household Expenditure Survey (HES) unit record file (ABS 2000). This Australian Bureau of Statistics (ABS) survey collected information on household expenditures, incomes and a wide range of other socio-economic characteristics. In total, 6892 confidentialised records were included in the publicly released file, representing some 7.1 million households.
However, for this study only households where the reference person or spouse was aged between 20 and 59 years and where there was either no other persons present or where there was one or more dependent children under 18 years of age were included. This age restriction for the adults was imposed to reduce any age-related effects that might be associated with families—those with and those without children — having different characteristics.
Also excluded from the analysis were households:
- whose principal source of income was from self-employment or reported as being from 'private sources' (eg. from interest or superannuation);
- with negative or zero consumption expenditure;
- with negative or zero income; and
- where the ratio of household total expenditure to household total income was greater than two.
The final data set thus defined contained 3340 records, representing some 3.234 million households.
One slight complication was that the age ranges in the original HES file did not exactly match with the age ranges required for this study, which were selected to conform with the current age ranges for Family Tax Benefit Part A and child support assessment. The relevant age ranges for FTB (A) are aged 0 to 12 years, aged 13 to 15 years, aged 16-17 years and aged 18 to 24 years. For child support assessment, it is to be less than 18 years of age. Age ranges in the HES are comparable except for children between the ages of 13 and 17 years, who are classified into the ranges 13 to 14 years and 15 to 17 years. To simulate the costs for children aged 13 to 15 years, we took 2/3 of the cost from the estimates for 13 to 14 year olds and 1/3 of the costs from 15 to 17 year olds. 16 to 17 year olds were held to have the same costs as 15 to 17 year olds. Given our earlier research which showed clearly that the direct costs of very young children were less than those of other children (Percival and Harding, 2002), we also estimated costs separately for 0 to 4 year olds and 5 to 12 year olds: these age variables were present in the HES so no imputation was required.
To ‘age’ the 1998-99 data to the desired 2005-06 world, we simply indexed both incomes and expenditures by the projected increase in the Consumer Price Index during that period, thus maintaining the relationships apparent in 1998-99 but expressing spending in current dollar values. While there may have been some changes in income and expenditure patterns during this period, particularly given the introduction of the Goods and Services Tax in July 2000, a more comprehensive analysis will have to await the release of the next Household Expenditure Survey.
2.3 Definition of expenditure
In this study, expenditure was defined as current household expenditure. This is total household expenditure as recorded on the HES less the following:
- repayment of mortgage principal for the family home;
- other capital housing payments (includes additions, extensions and renovations to family home, and purchase of other dwellings ; and
- expenditure on superannuation and life insurance.
The reason for excluding these items is that they represent saving, rather than consumption.
However, it needs to be acknowledged that expenditure on at least some of the above items is likely to be influenced by the presence of children in the family — that is, they could be said to have a non-discretionary element. For example, in the case of life insurance, it is quite possible that families with children are more likely to purchase these products.
Similarly, in the case of housing, once the decision has been made to become a home owner, the subsequent housing costs are not discretionary, including those that represent a form of saving.
What would be the effect of including items of non-current expenditure in the cost estimates? Most likely this would see some increase in the costs of older children. For example, typical housing repayment patterns see, over time, an increasing proportion go towards paying-off the mortgage principal. As a result, the housing costs of older families (who would be more likely to have older children) could be biased towards greater repayment of this mortgage component. (This may, however, be offset by the methodology not directly comparing families at the same point in their lifecycles. That is, couples with older children, who would be older themselves, are not compared with older couples without children. Instead, they are compared to couples without children of all ages.) As well, it is also possible that families with older children could be more likely to be paying for home extensions and renovations.
An important difference between this study and our earlier estimates of the costs of children published in the AMP.NATSEM Income and Wealth Reports series (Percival and Harding, 2002) is that we have excluded child care costs from the expenditure estimates. This is because the Task Force has noted the very major changes in government subsidies to childcare that have occurred since 1998-99 and has thus determined that major child care costs should be dealt with separately in any revised Child Support formula. The labour force status of parents with children can vary greatly, ranging from parents who do not participate in the labour force at all and use no child care to those who work full-time and use extensive child care. Accordingly, simply including the average costs of childcare within the child cost estimates overstates the child costs of those who do not incur child care costs and understates the costs of those facing substantial child care costs.
A further important difference to earlier work is that in this study we have attempted to estimate the costs of children in sole parent families, rather than concentrating only on the costs of children in intact couple families. This extension was undertaken to provide the Task Force with some guidance about the likely costs of children in separated families. The child costs for the two different family types were estimated independently, by separating the data according to the number of adults present in the household—i.e., one or two.