Household Expenditure Survey data - Report on user views

The household expenditure survey (HES) is the only comprehensive data collection of household spending in Australia linked to household type (single person, two parents and two children etc). The data is collected every six years, but a two-year delay to the 2020-21 collection means the ABS is reimagining how to deliver the household expenditure data.

To support this work, the ABS set up a temporary team to investigate the future of the HES. The project is applying human-centered design techniques to interview HES data users and understand their needs. Conducting 62 interviews with 93 individuals over a three-week period, the team identified use cases and the context of HES data, from the perspective of those outside the ABS. This paper summarises the views, opinions, and conclusions expressed by the users of Household Expenditure Survey data who were interviewed in this study. The ABS has not taken any decisions yet about the future of household expenditure data and will undertake a second round of interviews with users before the end of October. 

There are six main uses of household expenditure data, outlined below:

Throughout the interviews we heard that the link from expenditure to household type and other data is the key for the analysis and needed to be retained as a feature of the data. There were no users who simply used expenditure data in isolation.

We also heard that the ABS is trusted to deliver a quality product, and that users are ultimately agnostic about the source – administrative, big data or surveys – used to produce expenditure data. Users expect the ABS to explain differences across sources and methods and manage changes over time.

Users would like the quality of the data to remain high, however different users prioritised different aspects of quality. There is demand for more frequent data collection. Aspects of accuracy and sample size came up, particularly where advanced users conduct detailed and topical analysis.

Users from academia, government, non-government organizations and research institutes all want a rich and detailed set of variables at a common unit of analysis (currently households), available in both aggregate and as microdata.

None of the users interviewed considered other surveys to be viable alternatives to HES data. The Household, Income and Labour Dynamics in Australia (HILDA) survey was the most cited alternative, as many users put it: “Usually I will revert to HILDA if I can’t use HES”. However HILDA “couldn’t do what the ABS does” because the HES diary (requiring respondents write down all spending over a week) is too burdensome. While the HILDA data is more accessible than the HES data, the expenditure data is not as detailed

HES is well understood, both in terms of its content and methods. As stated by one academic “surveys have limitations, but these limitations are understood.”

Because HES data is the only comprehensive expenditure dataset, users are nervous about any change to HES. The HES output is “especially important, as other bespoke surveys have disappeared” (Peak body). Government agencies noted that without HES data they would be dependent on “past experience and theoretical relationships between economic concepts” in their forecasting and analysis. Several academics pointed out that for research they will “go to where the data is”, shifting to international data sources, if HES were not available or considered unreliable.

There are more than 250 users of the current HES data, which include academics, nongovernment organizations (NGOs), research institutes and government departments, both State and Commonwealth. The table below provides a summary of how the data is used, based on the completed interviews.

The frequency is the primary criticism of HES. Nearly all users state the six-year frequency limits the relevance, value and “credibility” of HES data, with one policy maker frustrated that “when accounting for timing and release time, we can be looking at 10-year old data”.

Many interviewees took this drop in relevance as a “given” but some users volunteered their reasoning for the reduced relevance. One driver was the perception of changes in the structure or nature of expenditure such as moves toward a cashless economy and shifts in expenditure levels (Academics, Government agencies)

A second driver for increased frequency is measuring short term impacts such as natural disasters, recessions, the COVID pandemic; and related policy interventions (Academics, Government agencies)

When asked how often would be frequent enough, the answer is simple: “the more frequent the better”. When pressed for specific answers, nearly all users desired increased frequency, but not at the expense of quality. For example, one government user had the view that “The highest desired frequency is monthly. It would significantly improve modelling… [but we] still need household/demographic characteristics and geography”. Some suggested administrative data would be “more timely, [and] help you get it NOW… But we still need household breakdown.” (Academic)

Some Government users noted that greater frequency would change their outputs and products, not only delivering increased accuracy but also new analysis.

Comparing the Australian HES with international counterparts, initial research from the ABS suggests that the 6-yearly cycle is among the least frequent, although Eurostat’s five-year cycle will become a minimum standard in the EU (see Figure 1).

Because of these concerns about relevance, several users age or update the data to undertake analysis on the current period. These users typically author their own products and report biannually, or more frequently.

Updating/uprating methods provide a buffer to the current timeliness demands for these users, and can be applied to manage the recent delay of HES to 2023/24. However, users are uncertain about how these may need to change to reflect the larger gap.

All users of HES data analyse expenditure with other variables. No-one looked at expenditure in isolation. Users therefore require expenditure data and coincident measures, including household characteristics, for a given unit (i.e. household, currently)

Users generally desire greater detail and disaggregation than currently provided. Multiple users raised, some with frustration, the difference between the detail collected by the ABS compared to the detail being published. Some academics made the point that “If you collect detailed demographic info but only publish the aggregate, why bother?”. Several stakeholders across multiple user types expressed they “would love to have individual level as well as household level data”.

Some users discussed improvements to the expenditure classification scheme (e.g. disaggregating electricity and fuel), and discussed the processes needed to align HEC coding with other classifications in use (e.g. CPI/COICOP).

While users did not raise the relationship between output granularity and sample size (or the implications on errors or confidentiality), sample sufficiency was implied in the conversation.

Several government and academic users explored the usefulness of a longitudinal/panel design to deliver frequency and detail, while adding greater explanatory power than the current crosssectional design.

Overall, substituting survey data for admin data was not inherently a concern for users so long as quality standards were maintained, and the ABS could explain the data differences and manage changes over time. Some noted the potential improvements to accuracy by replacing or combining items with admin data, and one Government agency “generally prefer admin data”. Some suggested the ABS should be looking to “utilise the power of a survey in the big data world”, such as using bank data to show inflows and outflows, linked to the survey data.

Most users had not directly used admin data in their analysis. For those who had not used big data there were mixed views of its potential. Some suggested combining the survey with admin data to improve quality or provide data for small area/regional/low-population estimates. Multiple academics stated that there are “no other sources of admin data” because they lack the same data richness as the survey.

This was particularly clear for subjective/qualitative items, such as perceptions of financial stress, intentions, or questions about why respondents’ behaviour is as it is. Many users were concerned about the representativeness of Administrative/Big data for their purposes.

Those who had used administrative data held similar views. These users explained the difficulty of integrating/linking data sources to produce an adequate and representative dataset to meet their needs. These include lack of linkage details, different levels of representation (e.g. individual vs household; transaction vs individual), varying degrees of timeliness, and conceptual or definitional differences across sources and time.

Users experienced with admin data were therefore sceptical (sometimes outright negative) about the ability to deliver the required richness. The most optimistic users evaluated the landscape as “a lot of work” (Academic, Government agency, Peak Bodies) which prevents some users from engaging with the data.

As stated by one researcher: “I never have to clean ABS data, it’s always perfect *kiss*”. While not all users agree with this notion of perfection, HES data is certainly seen as a high quality, unbiased source of expenditure data in Australia.

While users are open to improvements to the HES, and some would welcome a more frequent and bigger HES, there were clear reservations about frequency coming at the cost of quality. There were differences, depending on who we spoke to: Macro policy analysts wanted more frequent and less detailed information, but very quickly needed to be able to get into the detail of why things had changed. A number of modellers were relaxed about a HES every 5 years, or at a 2-4 year interval, as they would not be updating their models more frequently. The highest demand for a more frequent HES was from government departments, who require annual outputs and account for infrequent HES output in a variety of methods.

Many users, of all user types, reported the HES sample size is too small for specialised or small populations of interest, and volunteered populations of interest that might be targeted in future surveys, such as immigrants, high net worth individuals, and benefit recipients.

A number of users volunteered additional data items – either spontaneously or as context for other answers. These items are captured here for reference and transparency; this does not constitute formal consultation for future HES content.