Session Number: 4B
Session Title: Household Budget Expenditures and Budget Standards
Paper Number: 1.
Session Organiser: David Johnson, Bureau of Labour Statistics, US
Discussants: Patricia Ruggles (U.S. Department of Health and Human
Services), Joel Popkin (Joel Popkin and Co.)

 

Paper Prepared for the 26^th General Conference of
The International Association for Research in Income and Wealth
Cracow, Poland, 27 August to 2 September 200

PETER SAUNDERS, JONATHAN BRADSHAW AND MICHAEL HIRST

For additional information please contact:

Jonathan Bradshaw
Social Policy Research Unit
University of York
Heslington
York
YO10 5DD
UK
jrb1@york.ac.uk
Fax: 44 (0)1904 433477
Phone: 44 (0) 1904 433480

This paper is placed on the following websites: www.stat/gov.pl

www.econ.nyu.edu/dept/iariw

Any attempt to measure the adequacy of resources or the prevalence of poverty requires that a threshold be established against which the living standards of an individual, family or household can be measured. This threshold is necessary to determine whether they are in poverty (poverty rate) and/or how far below the poverty threshold they are (poverty gap). A large variety of thresholds have been established since a century ago Seebohm Rowntree (1901) in his pioneering study of poverty in York fixed his primary poverty threshold, as the income required to purchase mere physical necessities. This minimum budget standards approach dominated conceptions of poverty in the first half of this century. However, in the second half, a wider range of methods have been developed to establish thresholds, many in response to the advance in our conceptual understanding of poverty, not as an absolute or physical lack of basic necessities, but as a socially determined or relative lack of resources, as relative deprivation. Among these approaches there have been
* budget standards studies which have followed the essence of the normative Rowntree approach, but in their modern guise have sought to establish thresholds less concerned with physical necessities, including low cost and modest but adequate budgets (Bradshaw 1993, Saunders et al 1998, Parker 2000, Bernstein et al 2000).
* The US poverty standard was developed using a related concept - what one might call a ‘component-and-multiplier’ approach. Orshansky took the costs of a minimal food budget for different family sizes and derived poverty thresholds by multiplying these costs by three – that being the inverse of the share of money income spent on food by the average family (Orshansky 1965, Ruggles 1990). Bradbury and Jantti (1999) applied the US poverty standard to (circa 1995) Luxembourg Income Survey data using purchasing power parities). .
*A variety of poverty standards can be derived from expenditure data. So for example the point of the income distribution where households spend more than a given proportion on necessities can be used (Bradshaw, Mitchell and Morgan 1987).. Similar methods based on budget shares are reviewed later in this paper
*One of the earliest poverty threshold established in Britain (Abel Smith and Townsend 1965), and used by the government until 1984 in the Low Income Families series, was based on the social assistance scales rates plus a margin (usually 40 per cent). Heikkila and McCausland (1997) tried this technique using OECD data. Another technique combining expenditure and benefits has been used to estimate the budget shares spent on necessities (food, fuel and clothing) of those on Income Support and fix an income poverty line based on that budget standard (Bradshaw and Morgan 1987)
* Townsend (1979) pioneered the use of social indicators to establish a poverty threshold - relative deprivation was defined when a family lacked three or more deprivation indicators.
* Mack and Lansley (1983) and Gordon and Pantazis (1998) improved that approach by democratising the choice of deprivation indicators by asking the general population what they considered were “socially perceived necessities”. They also distinguished between items that were lacking because they could not be afforded, and items that were lacking because they were not wanted. More recently the list of socially perceived necessities has been expanded to a range of items, activities, circumstances and opportunities that better reflect the range of resources that ought to be included in a relative conception of poverty or social exclusion (Gordon et al 2000).Nolan and Whelan (1996) developed the technique by using social indicators in combination with income thresholds. At present there is a Survey of Poverty and Social Exclusion underway in Britain which seeks to extend the range of indicators to encompass social exclusion as well as poverty. The ECHP also has questions based on the social indicator methodology and Dirven et al (2000) have been using them to establish a basic index of deprivation.
* the threshold most commonly used by national governments and international organisations such as the EU or OECD has been to take a point on the distribution of equivalent income (or more rarely expenditure) - normally 40, 50 or 60 per cent of the mean or median. Eurostat have recently settled on a threshold of 60 per cent of the median.
*Subjective measures where the population determine a poverty income threshold can also be used to measure absolute poverty. Thus for example after the World Summit on Social Development in Copenhagen in 1995, 117 countries adopted a declaration and programme of action which included commitments to eradicate absolute and reduce overall poverty, drawing up national poverty alleviation plans as a priority (UN, 1995). Absolute poverty was defined by the UN as a “condition characterised by severe deprivation of basic human needs, including food, safe drinking water, sanitation facilities, health, shelter, education and information. It depends not only on income but also on access to services”. (UN, 1995, p. 57). Townsend and others (1997) have attempted to operationalise this notion of absolute poverty (and overall poverty) using subjective methods.

In summary there have been four different approaches: budget standards; relative deprivation indicators, arbitrary relative income or expenditure thresholds; and subjective measures - though these are tending to converge with elements of each often used together.

There is no doubt that this century of endeavour has advanced our understanding of poverty, both conceptually and empirically. Nevertheless since the abandonment of absolutist notions of poverty something has been lacking. Rowntree’s approach was somehow unchallengeable - there was categorically no doubt that human beings needed the physical necessities for survival. His findings had clout, advancing both understanding of poverty - that it was structural - the result of low wages in 1899 and not behavioural as the Charity Organisation Society had opined. He also a had a profound impact on policy, influencing the social reforms of the 1908-1923 Liberal Government.

In contrast the other approaches have been easier to challenge -
* why should such and such an item be included in a budget standard, costed at that price and given that lifetime?
* Social assistance scales were established to prevent poverty, lift people above it. How can it be justified to use them as a poverty threshold?
* Why should 40, 50 or 60 per cent of average income be poverty thresholds - surely they are points on a distribution - merely measures of inequality, which is not the same?
* Why should lacking three socially perceived necessities rather than (say) one or six, be a good poverty threshold?
* Is not self perceived poverty flawed by false expectations?

These are some of the criticisms made of each of the methods outline earlier. There are of course some justifications in them, and indeed answers to them. But they illustrate that as the setting of poverty thresholds has moved away from poverty defined as physical necessities, in absolute terms, the moral authority of the word poverty seems to have diminished. Or at least, despite the range and variety of evidence that substantial minorities of our citizens (at least in Australia and the United Kingdom) are living in poverty, and that in the last two decades the problem in both countries has got substantially worse, policy makers in those countries have failed to respond to the evidence. They have failed to respond as effectively as they did, for example, in Britain in the great liberal social reforms of the Edwardian period or later in the Beveridgian welfare state legislation introduced after the second world war in Britain and in Australia.

Admittedly things may be changing. In Australia in 1987 Prime Minister Hawke pledged to end child poverty by 1990, and the Adequacy Project was launched under the Keating Government to re-establish some consensus about the level of pensions and other benefits. In Britain the new Labour Government is targeting poverty, has promised to abolish child poverty within 20 years and is to produce an annual report on the consequences for the poor of its welfare state reforms. This has inevitably led to a debate about what is meant by poverty - if targets are to be set, what measures should be used to establish whether they are achieved?

We have been involved in (re)pioneered budget standards research in our two countries and believe that they have potential for establishing thresholds which have transparency and can be readily understood by the person in the street. However, budget standards are enormously difficult to establish and keep up to date. This paper, which developed out of the experience of undertaking budget standards research, presents the results of an exploratory attempt to shortcut the full budget standards enterprise. It builds on and develops work originally undertaken by Saunders (1996 and 1999) to establish income adequacy benchmarks from behavioural data as part of the work to establish Australian budget standards.

This paper updates that work for Australia and incorporates a comparative element to it by applying identical methods to the UK. In the course of this collaborative work, we have established comparable poverty rates for the two countries and for sections of the population within countries, which have there own intrinsic interest. We have also been able to compare patterns of spending within and between countries. However the main empirical purpose of the paper is to discover whether we could estimate income thresholds for each country which could be used to measure poverty with more authority than the existing thresholds.

The paper is based on the secondary analysis of the 1993/94 Household Expenditure Survey (HES) in Australia and the 1993 Family Expenditure Survey (FES) in Britain. Both surveys are government funded general population surveys undertaken principally to establish the weights for the retail price index. The FES is carried out annually and the HES every five years. This analysis is based on the latest year for which the HES is available. Both are national surveys of a random sample of households interviewed throughout the year. As well as collecting data on the household circumstances and detailed data on income, both surveys employ expenditure diaries to collect information on household (and individual) expenditures for a two week period around the time of the interview.

The major task in this project was to make the analysis of the separate surveys as comparable as possible. This involved making a host of choices (and indeed concessions) and, in the case of the expenditure data, recoding the expenditure items so that they were as comparable as possible. This involved time consuming and detailed work trawling through expenditure codes and their definitions, the tedium of which only those who have experience of handling expenditure data will be able to appreciate. It is not possible to summarise all the decisions that were made but suffice it to say that they ensure that in all important respects the components of the key variables (income, expenditure), the definition of the characteristic (household type, age, employment status), and the way they are manipulated (mean, median, equivalence scale, quintiles) are comparable. The analysis is restricted to single tax unit households. In the case of the HES, the few cases with very large negative incomes were excluded because they made a difference to the results obtained. In both cases the weighted data, using the official statistical service weights, are used to compensate for known response bias.

The analysis starts with a comparative presentation of descriptive data. Then conventional poverty comparisons are presented. Then there is a descriptive analysis of expenditure patterns in each country. Finally poverty thresholds are derived and estimated for a variety of different household types.

In Table 1 it can be seen that Australia is a rather younger country than the UK, with fewer aged and more couples with children. These differences in the structure of the population will be taken into account in subsequent analysis.

Table 2 provides a summary of the level and sources of income in each of the samples. The income levels are expressed in national currencies so that they are not comparable but the ratios of income to gross pre tax and transfer are remarkably similar. Overall households in the UK retain a slightly higher proportion of their income after transfers and after direct taxation.

Table 2b Comparison of income and sources

Tables 3a and 3b enable us to compare the proportions of each household type in each of the quintiles of equivalent disposable income. In Australia, a much higher proportion of the single person households - aged and non aged, are in the bottom quintile group. In contrast, in the UK a higher proportion of households with children particularly lone parent families are in the bottom quintile. In Australia childless couples are concentrated in the top quintile while in the UK there are a higher proportion of the aged in the top quintile.

Without further analysis it is only possible to speculate on why these differences are occurring, They are likely to be a function of the fact that all aged pensions are means-tested in Australia; the incidence of unemployment by family type - unemployment is possibly higher among families with children in the UK and among the single in Australia; the level of benefits payable to families with children in-work and out of work are lower in the UK; and the equivalence scale may also be having differential impacts in each country.

There is a debate in the literature about whether, when using an income threshold to set a poverty level, it is better to use a measure related to the mean or to the median. Table 4 estimates poverty rates for the mean and Table 5 estimates them for the median, both tables using the conventional thresholds 40, 50 and 60 per cent. Whatever the threshold and whether poverty is measured using the mean or the median, the UK has a higher overall poverty rate than Australia - particularly at the lower 40 per cent threshold. Comparing proportions of the mean, the difference is most marked using the 40 and the 50 per cent thresholds. At the 60 per cent threshold the difference in the poverty rate, though still higher in the UK is not as big. In contrast, when the median is used there is very little difference in poverty rates at the 40 percent threshold, but much more difference at the 50 and 60 per cent thresholds, with Australia having lower poverty rates. Relative to the mean in each country, the median is higher in the UK - ie there is greater income inequality in the UK than in Australia.

Although we can be fairly confident from this analysis that relative poverty rates in the UK are higher than in Australia these results illustrate the importance of the threshold and the nature of the income distribution in making comparisons of poverty rates across countries.

These are even more important in comparing the poverty rates of particular household types. For example taking the single aged, the threshold based on the mean gives Australia a lower poverty rate at 40 per cent of the mean, a higher poverty rate at 50 and 60 per cent of the mean and at the median a lower poverty rate at 40 per cent and 50 per cent but a higher poverty rate at 60 per cent. For aged couples the estimates are also very sensitive to the thresholds being used. This is likely to be due to bunching in Australia due to the pension system being a flat rate payment. However for the other households groups the poverty rate estimates are much more stable. The picture is particularly consistent for families with children - regardless of the threshold, Australian families have a lower poverty rate than they do in the UK.

An alternative to the analysis of income is to analyse expenditure. It has been argued that expenditure is a far better representation of permanent command over resources, enabling, in particular, borrowing capacity and dissavings to be incorporated into the analysis of consumption. In particular in comparing poverty rates between countries, one problem with income is that it fails to take into account comparisons in patterns of consumption which are the result not of command over resources but reflect differences in relative prices due to national, cultural, climatic or other factors. Thus in comparing income, because we are not comparing what the income has to be spent on, we are not comparing like with like.

Table 6 compares the budget shares (the proportion of the total budget spent on different commodities by quintile groups and Table 7 compares the same by family type. Overall households in the UK spend a larger proportion of their budgets on housing costs, fuel and power and alcohol. Households in Australia spend a higher proportion of their budgets on transport in the lower quintiles and on medical expenses. Apart from these commodity items the overall budget shares are very similar.

The budget shares by quintile group are very similar between countries and for food expenditure they are nearly identical. In the UK lower quintile groups spend a higher proportion of their total expenditure on housing and fuel and power whereas the proportions are more even in Australia. The budget share on tobacco does not decline in Australia as income increases as it does in the UK.

As far as household types are concerned single pensioners in the UK spend a much higher proportion of their budgets on food and in the UK all household groups spend more on housing and fuel and power. Apart from this the differences in budget share by household group are very similar.

Table 8 compares the poverty rates using an expenditure threshold (below 50 per cent of the median) with an income threshold (also below 50 per cent of the median). It can be seen that the overall proportion living in poverty in the UK is higher for both measures but the difference is much more marked using income rather than expenditure. Indeed for all the household groups, other than families with children, expenditure poverty is higher in Australia than in the UK.

Given these findings, which measure of poverty is the most appropriate? It can be seen that the expenditure poverty rates give a substantially higher poverty level for both countries than the income poverty rates. This is because in both countries pensioners are not spending up to their income - they are saving. In contrast, households with children have higher income than expenditure poverty rates which indicates that they are spending more than their income - dissaving. Table 8 compares poverty rates using two other measures: income and expenditure poverty is when the household is below both the income and expenditure thresholds. Core poverty is even more restrictive in that it excludes those with income greater than expenditure on the grounds that they are still managing to save despite their income and expenditure being below both thresholds - they may conceivably be obtaining some of their resources in kind or as gifts from friends of relatives (Saunders 1998).

The income and expenditure poverty measure gives much higher poverty rates for the UK for all groups, except aged and non aged couples though the poverty rates for them is still higher. It is notable that lone parents have particularly higher poverty rates in the UK. The core poverty rates are much lower than the income and expenditure poverty rates for both countries. Again the poverty rate for Australia is only about half that for the UK and it is most different for the single non aged and lone parents.

Table 9 compares the composition of the poor using the various poverty measures. On expenditure poverty the composition of the poor is very similar but with a higher proportion of lone parents in the UK. There are more differences between the countries on the income measure with more aged singles and lone parents in the UK and fewer aged couples and couples with children. These differences are sustained using the income and expenditure measure. The core poverty measure gives a higher proportion of non aged couples in Australia and a higher proportion of lone parents in the UK.

Table 9 :Poverty rate: Various measures. Composition of poor households

CONSTRAINED EXPENDITURE
Expenditure data is of limited usefulness for deriving poverty thresholds because expenditure is the outcome of constrained choices and thus reveals little information about the nature and level of needs or the extent to which needs are being met by expenditure. In order to overcome this problem it is necessary to combine expenditure data with some external criteria (generally normative in nature) which allow conclusions to be drawn about needs from observed expenditure patterns. One possibility first proposed by the Watts Committee and explored using UK data by Bradshaw, Mitchell and Morgan (1987) is the S-curve approach. This is based on the idea that starting from a very low income, successive increases in income will first be devoted to increasing the quantity of necessities purchased but after a while further increases will be used to improve the quality of what is bought. If this is the case then plotting the quantity purchased against income will produce an S-shaped curve whose slope first rises rapidly but then rises more slowly as quality is in effect substituted for quantity. The income corresponding to the inflection point on this curve - the point at which the S starts to level out - is then defined as the income at which basic needs are met.

The main problem with the method are practical. Detailed data linking the quantities of specific items purchased to the level of expenditure are rarely available. Instead what tends to be available is data linking expenditures on specific commodities to total expenditure (and total income). Observing how these vary in relation to each other will only provide an insight into the shape of the S-curve if prices can be assumed to be constant, so that expenditure is proportional to the quantity purchased. But this is most unlikely to be the case and in practice attempts to identify inflection points on the S-curve have been largely unsuccessful (Watts1980, Bradshaw, Mitchell and Morgan 1987).

An alternative approach involves an explicit normative judgement designed to identify one particular living standard as corresponding to a level of inadequacy. Clearly this approach is arbitrary but at least the judgement on which it rests is straightforward, easily identifiable and readily varied. This is the approach used in deriving the Canadian ‘Low Income Cut Offs’ (LICOs). As originally developed by Podoluk (1967) the method involves defining the list of necessities to include food, shelter and clothing and then anchoring the LICOs at the point at which the proportion of total expenditure devoted to these items is at some arbitrary level above the average proportion of expenditure devoted to necessities by all households. The budget share has declined over time and the degree of complexity of the LICOs has also increased (Wolfson and Evans 1989) and has been applied to Australian data by Saunders (1996).

Saunders has also developed a new method which addresses the problem from a somewhat different direction. The method is similar to the S-curve method in that it attempts to draw conclusions about the standard of living by observing changes in the nature of expenditures and how these change as total expenditure changes. The basic idea is quite simple. It is that where resources are already severely stretched, it will not be possible for households to undertake either major purchases (e.g. of household durables and the like) or to engage in what can be broadly though of as expenditures on luxury items. The former will be postponed until circumstances improve, while the latter are likely to be permanently deferred. Existing durable goods will be maintained longer than would otherwise be desired because priority will have to be given to meeting basic daily consumption needs. Households will tailor their expenditure to fit current constraints by in effect supplementing current consumption by running down their stock of capital goods. The approach involves identifying the income level at which all income is being spent, but none of it is devoted to purchasing any of the major durable (or luxury) items. This income is then equated with the level at which all resources are being used to meet immediate needs with nothing left over to replace the household’s capital stock, nor to consume luxury items. The key point to note about this method is that it attempts to utilise data on the absence of expenditure on durable goods and luxury items as a way of identifying inadequately low income, as opposed to using information on the lack of ownership of certain durable items to indicate the presence of deprivation. The method developed here thus focusses on analysing the current flow of expenditure over a period of time, rather than patterns of ownership of the stock of durable goods at a point in time.

The approach can be applied in several different formulations depending on how expenditure on durable and luxury items are defined. In this experimental analysis the results for one formulation are presented. This involves no expenditure on household goods and equipment, motor vehicles, bikes, white goods, audio equipment, luxury goods such as watches and clocks (See appendix). Table 10 provides a summary of the proportion of households with nil expenditure on those items in each country. Overall, the proportion is higher in the UK than Australia which is to be expected given the higher poverty rates found in the UK. The proportion of non aged childless couples with zero expenditure is identical in both countries and for couples with children it is very similar. The differences in the proportions are greatest for the aged, the single non aged and lone parents. Of course the aged are more likely than the other groups to have zero expenditure due to choice on the grounds that they have already obtained these items and may be more likely to depreciate these assets in the final stages of their life course.

Table 10: Constrained expenditure items

 For each country a regression equation was estimated in which total household expenditure was related to a number of household characteristics, to total disposable income and to the list of commodities and services. Once estimated the model was then solved to determine the income level at which total income is equal to total expenditure (implying that there is no scope for saving at the household level) and total expenditure on the list of identified commodities is zero. In deciding on the specification of the model, variables were chosen which were prima facie likely to be associated with variations in living standards, were relevant to policy and available in both data sets. Table 11 summarises the regression results for both countries. In both countries total expenditure, private renter, number of adults and number of dependent children make a significant contribution to the model. In Australia being an employee and being under 25 are also significant as are being a purchaser, being self employed and the number of earners in the UK. The proportion of variance explained by the model is rather higher in the case of the UK model than that for Australia.

Table 11: Estimations of income at which expenditure on non constrained items exceeds zero.

*=statistically significant at least at the 0.05 level.

The next stage involves using the estimates to derive the income levels at which all household income is spent and none of it is used to purchase any of the identified durable items. This is taken to correspond to a low-income cut-off at which all resources are being devoted to meeting the basic on-going consumption needs of the household, with nothing left over to engage in large expenditures, no matter how urgent the need for the items which these could secure, or for expenditures on what would generally be agreed to be closer to luxuries than to necessities.
There is an enormous range of household characteristic combinations for which the equation can be solved to give this income level but only a selection are derived to give a flavour for what the estimates imply. These are shown in Table 12. It can be seen that there is a good deal of variation in the income thresholds both within and between countries. The estimates for an employed single person are much higher in the UK than in Australia and for the aged are higher in Australia than in the UK. The relativities to a couple also vary a good deal in each country - for example lone parents in the UK with one child, not in the labour force, public renter has a threshold less than half that of a couple and a lone parent in Australia with one child , not in the labour force and a private tenant has a threshold more than 50 per cent higher than a couple. It is also difficult to explain why a two earner couple in Australia has a threshold only half that of a one earner couple.

Table 12 Income threshold based on constrained expenditure model (Couple with one earner (employee) purchaser=100)