This chapter provides an overview of the general NTA and NTTA methodology and describes in detail the estimation of private transfers within households for Austria in 2015. The basic NTA methodology is described in detail in the manual of the UN (UN, 2013) and the NTTA methodology in Donehower (2019). Hammer (2020a) provides the description of data and specific methods for estimating NTAs for Austria 2015 except the details for private transfers.
National transfer accounts: a general overview
NTA data provide information on age-specific means of income, public transfers, private transfers, consumption and saving. The broad estimation strategy for NTAs is, first, to derive the aggregate values from National Accounts data and related sources. Aggregate values refer to the quantities for the total economy, i.e. total labour income or total consumption in Austria 2015. In the second step the distribution over age groups and genders is estimated using survey- and administrative data. All age profiles are adjusted by an appropriate factor, to ensure that the per-capita averages for each age group summed up over the total population match the aggregate values. Likewise, gender-specific averages are adjusted so that the sum of male and female-specific values in each age group equals the age group total. In a third step transfers within families are estimated. These estimates of intra-family transfers are based on the difference between consumption and income.
Table 1 illustrates the central patterns of NTAs for Austria 2015. The average yearly primary income is highest in the age group 25-59 with more than 45,000 Euros on average. Primary income refers to income generated by direct participation in the production process and asset income. On average, a 25-59-year-old pays more than 13,000 Euros to children and elderly persons via net public transfers, and about 4,000 Euros to children in form of net private transfers. For children and young adults aged 0-24 the private transfers are the most important source of income. They receive more than 8,000 Euros of net private transfers per year, and more than 6,000 Euros of net public transfers, with education as main component. Public pensions and health services are the largest components of public transfers and directed mainly to the retired population. In total, the population 60+ receive more than 17,000 Euros of net public transfers, on average. Adjusted disposable income (including public transfers in kind) is increasing with age. While adjusted disposable income of the population below 25 is less than 20,000 Euros, it is more than 34,000 Euros for the population 60+. Consequently, the population 60+ has the highest level of total consumption (public + private) and the highest level of saving.
Age and gender-specific NTA data for Austria is publicly available through the Austrian Social Science Data Archive (Hammer 2020b) and through the data base of the NTA project at www.ntaccounts.org. We changed the methodology for estimating private transfers in the course of writing and revising this paper, resulting in small differences between the data presented in this analysis and the NTA data included in the databases. The methodological changes are related to the imputation of consumption in income data, the original methodology is described in Hammer and Prskawetz (2020).
Microdata for the estimation of private transfers
Intra-family transfers in NTAs are not measured directly, but are based on the difference between income and consumption of individuals. It is assumed that consumption is financed through transfers from other household members, if own income is not sufficient. These transfers consist mostly of flows from parents to children and, in gender-specific NTAs, of transfers between partners with large differences in income. NTAs focus on ”current transfers”, i.e. regular transfers. Wealth transfers, such as bequests or dowries, are therefore not included. Although wealth transfers play an important role in intergenerational redistribution, the limited data availability prevented so far the extension of NTAs with wealth and wealth transfers.
The estimation of intra-family transfers in NTAs relies on micro-data that represents the household structure and contains information on the consumption of households and the income of individual household members. However, in most countries, microdata that include both, information on consumption and the income of individual household members, are not available. When income and consumption are from different datasets, the NTA manual (UN, 2013) suggests to impute age-specific means of the missing income and/or missing consumption into microdata that represents the household structure. Estimates of intra-family transfers are then based on these imputed data. Such an approach has been chosen for Austrian NTAs for the years 1995, 2000, 2005 and 2010, which impute age-specific means of income (from tax statistic) and consumption (from the consumer expenditure survey) into data from the micro-census.
The NTA manual is clear about the weaknesses of NTAs that are based on microdata with imputed means, among them a bias in the estimates of transfers paid and transfers received and the impossibility to use these data to analyse subtypes of households. For example, many families with young children are characterized by large transfers of income from fathers to mothers, as mothers specialize in unpaid non-market work and fathers in market work. Age-specific averages of income greatly overstate the income of mothers and therefore underestimates the extent of intra-family transfers. Furthermore, to calculate age-profiles for sub-types of households, the differences across subgroups must be reflected in the imputed values. Abio et al. (2021) impute age-averages of consumption by subgroups of households, distinguishing by age, gender, education and family type. The disadvantage of this approach is that these estimates are partly based on few observations and require broad categories. For example, Abio et al. (2021) only distinguish between families with dependent children and without dependent children, ignoring the large differences within these groups, for example by age of the children.
The estimates of intra-family transfers and the scope of the NTA analysis can be greatly improved by combining data on income and consumption in a way that maintains the relation between household characteristics, income, and consumption. In particular, the imputation needs to account for those characteristics that have the greatest impact on consumption, most notably income. Such an approach enables the separate estimation of transfers paid and received and the analysis of intra-family transfers by characteristics other than age and gender. Since microdata with individual income is in most countries available the problem reduces to the imputation of consumption into income data.
Imputing consumption into income data
A range of different methods have been developed for imputing information on consumption into income data. In particular, we can draw on the work for integrating indirect taxes (consumption taxes) into the European tax-benefit simulation model EUROMOD, which is based on EU-SILC. For the 2015 NTAs, we use the information on consumption from the Austrian consumer expenditure survey (CES)Footnote 3 and impute this information into EU-SILC. The Austrian CES contains information on consumption of households as well as information on characteristics of households, such as household income, household size and the age of household members. EU-SILC contains information on income of each household member as well as characteristics of households. The information about households that is available in both surveys is used to estimate and impute consumption into EU-SILC data. Two broad types of imputation methods can be distinguished, regression based methods and matching methods. Regression models are estimated using CES, the predicted values of the consumption expenditure for specific household characteristics are then imputed into income data. The matching methods impute consumption expenditure into income data by using the values from CES-households with similar characteristics.
Decoster et al. (2007) compares different imputation methods, including both model-based and matching methods. In their analysis the model based methods produce a better fit of the imputed data to the observed data, while the matching methods are better in reproducing the original distribution of consumption. However, the differences between the distribution of consumption in the model-based imputations and the original data is a result of a deterministic imputation, which ignores the unexplained variation across households. Reproducing the distribution with a regression model requires adding an error term that captures the unexplained variation in the data. The first implementation of the indirect tax tool in EUROMOD is based on a linear model that relates the log of consumption expenditure to household characteristics (De Agostini et al. 2017). An interesting alternative is suggested by López-Laborda et al. (2020), who argue that the re-transformation of log consumption to nominal consumption expenditure, in the presence of heteroscedasticity, results in biased estimates. They suggest the use of generalized linear models and show that a GLM model with the log as link function and a distribution from the gamma family delivers the best results.
Matching methods have several advantages, among them that the imputed values inherit the distribution including the unexplained variation from the original data. For experimental data on income, consumption and wealth, Lamarche (2017) use a hot deck matching procedure, which uses for each household in the income survey the consumption expenditure of a comparable household in the consumption survey. The newest version of the indirect tax tool in EUROMOD uses a combination of model based approaches and an hot deck matching approach (Akoğuz et al. 2020). First, a regression method is used to impute consumption in EU-SILC and the data from consumer expenditure surveys (the source data for consumption). Second, the imputed values are used to match households in EU-SILC with those in the consumer surveys. The final imputation is not the predicted values from the first step, but the actual values in the matched households. The method is implemented in EUROMOD, but unfortunately not for Austria.
The literature provides no clear indication that one method would outperform others. We therefore applied several of these imputation methods and evaluate them according to how well the imputations reproduce the estimates of age-specific consumption from the original CES data. We use a linear model for the log of consumption, a GLM model with the log as link function and gamma distribution as well as hot deck matching of households. For our purpose a simple hot deck matching seems more appropriate as we aim to match households with similar characteristics, which is not guaranteed with the two-stage hot deck method. For all the methods we applied we use the same variables as explanatory characteristics of the level of consumption: household income per household member, the number of household members, the number of children below six, the number of children aged 6-14, and the number of persons of age 70 and older. Most research uses detailed characteristics of households, such as region and education level. However, these variables do not add to the explanatory power of the model once income and household size is accounted for and they do not change the results of our analysis. We therefore decided for a more parsimonious model. Details for each imputation method as well as the results of the regression models are given in Appendix 5.1.
The results of our comparison of imputation methods indicate only minor differences between the various approaches (Fig. 8 in the Appendix). We finally decided for the matching method because it is the most intuitive one and the associated age-profile had the smallest absolute deviation from the age-profile based on original data.
Private market transfers: methodology for Austria 2015
After imputing consumption at the household level, consumption of households is allocated to its members based on the NTA consumption equivalence scale. This equivalence scale assumes that children until age four represent 0.4 equivalent consumers and for persons between age 4 and age 20 the equivalence scale increases linearly to one. After allocating household consumption to individual members, individual income and consumption is adjusted so that age and gender-specific values correspond exactly to the NTA age-profiles. This ensures that the estimates of intra-household transfers are consistent with the system of NTAs, i.e. that income plus public net transfers plus intra-family net transfers equals exactly consumption and saving at each age.
We need to emphasize that the estimates of consumption of individual household members and consequently the size and direction of intra-family transfers is influenced by the NTA consumption equivalence scale. The use of equivalence scales however is a common approach for evaluating costs of children (Humer and Rupp 2020) and the NTA scale has been intensively discussed within the NTA network. It represents a consensus that is based on the available evidence from different countries and facilitates cross-country comparisons, because the results are not influenced by country-specific methods and data for estimating consumption of children. Its basic features such as the strong increase of consumption with the age is also confirmed by the newest study of the costs of children in Austria (Bauer et al. 2021).
The further algorithm for estimating intra-household transfers in Austrian NTAs deviates slightly from the methods suggested in the NTA manual, because of the adjustments required for a gender-specific estimation and particularities of the data. We assume that income is shared within couples. The sharing among couples constitutes the first component of the intra-household transfers, and represents mostly transfers from fathers to mothers, because the latter reduce their employment (and income) because of care responsibilities. In the next step we calculated the difference between consumption and the shared income. If consumption exceeds income, it is assumed that the difference is financed by household members with income exceeding consumption. We assume that consumption of children is always financed by intra-family transfers, even when total income of households falls short of consumption and there is no household member with excess income. This assumption is necessary, because information on consumption expenditure is only collected for a two-week period and therefore subject to high random variation; the extrapolated consumption expenditure to yearly values frequently exceeds yearly income of households. Likewise, in other households the extrapolated consumption underestimates yearly expenditure. With our approach of accounting for transfers even when consumption estimates exceed income we receive unbiased estimate for transfer averages by age and parental status. This component of intra-household transfers constitutes mainly a redistribution from parents to children.
The main difference between the algorithm suggested in the NTA manual and Austrian NTA is the lack of an explicitly defined household head. The algorithm suggested in the UN, (2013) is quite complex and gives the household head a central role. It is assumed that only the household head owns asset. This assumption affects intra-household transfers, because all income that is not used for consumption needs to be transferred to the household head for saving, and only the household head can finance the consumption of children if total household income falls short of consumption. We decided to adapt the algorithm because it is likely to bias the gender-specific results. We know from a special module in the Household Finance and Consumption Survey that most assets are equally distributed among couples rather than belonging to a single person in the household (Groiß et al. 2018). By assigning all assets to a single person in the household and making her to the central person regarding intra-family transfers, results of gender-specific NTAs are likely to be strongly influenced by the choice of the household head. Because of this concern we treat all adult members in the same way. Our approach aims to improve the gender-specific estimates, but does not limit the comparison of non-gender-specific NTAs with the data from other countries.
A further difference to the guidelines in the NTA manual is that inter-household transfers are not estimated in Austrian NTAs for 2015. Data on inter-household transfers between generations in Austria are of low quality and do not permit age-specific estimates. As we mentioned above, the transfers measured in NTAs include only ”current transfers”. Wealth transfers, such as bequests, are not included. The negative residual (”saving”) for young adults (Table 1 and Hammer (2020a)) suggests that current transfers and wealth transfers between households may play an important role in financing consumption in young adulthood. Negative values indicate that part of consumption is not financed out of income and most likely through transfers, because Austrians usually do not finance their consumption through credit. Unfortunately, the limited data availability prevented a more detailed analysis of inter-household current transfers and of wealth transfers.
Private non-market transfers: national time transfer accounts methodology
National Time Transfer Accounts measure production, consumption and transfers of services produced by non-market work. The most important non-market services include cooking, cleaning, shopping and childcare. Non-market production of households is not captured in the National Accounts core system, but is occasionally estimated in so called household satellite accounts (Poissonnier and Roy 2017; Communities 2003). The estimations are challenging, since there are no data on the output of non-market production activities of households, let alone their market value. Furthermore, there is not even data on the value of the inputs, consisting mostly of unpaid work. Most household satellite accounts measure non-market production by valuing the time input with wage rates for comparable activities.
Non-market production for other household members is not only an important part of total production, it constitutes a fundamental type of transfer between gender and generations. Donehower (2019) developed the NTTA methodology to integrate non-market work into the NTA framework. NTTAs measure production, consumption and transfers of non-market services by age and by gender. We estimate NTTAs for Austria using the most recent time use survey (TUS) of 2008/09Footnote 4.
While allocating non-market production to individuals is straightforward, the estimates of non-market consumption and non-market transfers require several assumptions. TUS micro-data provides information on time use for non-market production at individual level. Production estimates by age, gender or parental status simply use this information. However, the data do not provide information about who in the household consumes the produced services. To allocate non-market consumption to individuals, NTTAs use several methods and assumptions. These methods depend on the type of the activity and whether it is carried out for household members or other households. Non-market transfers are calculated as difference between production and consumption.
NTTAs assume that each household member profits equally from general household services such as cooking or cleaning. Unfortunately, we cannot observe who in the household consumes a certain service, the assumption that all members profit equally represents a reasonable minimal consensus that is applied in all countries and thereby ensures the comparability of NTTA data across countries. Consequently, consumption of household services that are produced for the own household is estimated by adding up the time used for production within households and distributing it to all members in equal shares. The consumption of non-market services produced for other households assumes an equal distribution over the total population, i.e. independent of age and gender. Non-market services for other households include voluntary work for non-profit organizations as well as help in household work.
Most childcare services are consumed by children in their very first years of life. To estimate the consumption of childcare services provided by household members, we assign the total time used for childcare in the household to the children. This approach is straightforward if there is only one child in the household. In case there are more children in the household, we use age-specific weights. The age-specific weights account for the higher need of younger children and are based on age-specific amount of childcare in households with only one child. The Austrian TUS includes information on childcare provided for other households. We allocate these care activities to the children outside their own household using age-specific weights, which are derived from information on the use of childcare in EU-SILC. The weights reflect that non-household members, such as grandparents, use more time to take care of older children and less for very young children. For adult care we assume that it is provided to the oldest person in the household. In general, total time that is identified as adult care in the time use survey is small.
To assign a monetary value to non-market work we value each hour of non-market work either with the minimum wage rate for housekeepers or childcare workers. The activities other than childcare are valued with the hourly costs of a housekeeper employed at minimum wage in 2015, accounting for Christmas and holiday allowances, holidays and the employers social contributions. Childcare is valued with the costs of a childcare worker, which is somewhat lower than the costs for a general housekeeper.Footnote 5 Donehower (2019) suggests the use of average wages to value unpaid work. Because we assume the minimum wage of experienced workers and top-ups for household staff over minimum wages are low, these values should represent a good approximation for average wages in this sector. The rationale behind the use of gross wages for valuing non-market work is that a shift from non-market work to market work should not affect the measure of total production and transfers. If parents decide to employ a nanny instead of providing childcare by themselves, this should not change the transfer measure. Since parents must pay gross wages for the nanny, we must value their own work by the same amount.