There was lots of understandable angst on social media yesterday morning when it looked like the Abbott government was hell-bent on doing away with the national census. However by yesterday afternoon it seemed that things aren’t quite as bad as first feared.
It now appears the proposal on the table is coming from the ABS. The idea isn’t to do away with the census entirely but rather to shift from a five year to a ten year cycle as per the US and provide more frequent survey-based data in the intervening years. The 2016 census would be cancelled and the next one conducted in 2021.
There are still doubts about what’s really going on, but if we take the ABS at its word, it’s worth considering if there’s a case for the Bureau’s proposed changes. Here are what I see as the main “pro” arguments for a decennial census cycle:
- It would save a lot of money. The 2011 Census cost around $440 million and the 2016 census will probably cost considerably more. That money could be used for more productive purposes like providing even better data.
- The Census is very limited; it only provides a snapshot of the population on one day in every five years.
- It would free-up funding to enable data to be collected more frequently – possibly every three months for basic data – by sample surveys.
- A census provides very little methodological benefit; the error level in large scale sample surveys is very small. Like the census, participation in surveys can be made compulsory.
- The accuracy of survey data will be calibrated against the full census every 10 years.
- There’ll be no reductions in the scope or depth of the data collected.
- Getting finalised census information to users is currently protracted because of the scale of the undertaking: it takes 10 months for some of the basic data to be processed and 2-3 years for some of the more specialised data to be released.
- It would provide the Bureau with the resources to install new electronic technology that will reduce the cost of conducting the census in future years.
Many of the objections to the proposed changes are essentially political. They include suspicions that the government is driven by the need to cut expenditure and that it doesn’t understand the social value of statistical information.
Another plausible argument is the ABS might be motivated by self-interest. Its attitude could be biased by years of underfunding by successive governments and by the expectation it can’t hope to get the money it really needs.
It might be trying to avoid the embarrassment of not being adequately prepared for next year’s census. Perhaps it’s also playing a strategic game in the public arena in the hope of influencing the government on funding.
But if we continue to take the ABS at its word, the arguments “against” moving to a decennial census include:
- No survey is as good as a census; other jurisdictions have experienced low participation rates in voluntary surveys.
- Fine or disaggregated data might not be sampled at a sufficiently high level to provide reliable information e.g. small geographies.
- The half a billion dollars or so it costs to conduct the census is once every five years; that’s small change in the context of the total Commonwealth budget and the social benefits it provides.
- Where the stakes are high there’ll be politically motivated disputes based on the argument that survey data can never be 100% definitive e.g. grants to the states.
It’s hard to make a call on this one on the information that’s available. So far the government has kept mum and the ABS has hardly been as clear and straightforward as it should be. I also expect there’ll be further objections raised if this debate continues, particularly from specialist users of census data.
The value of rich and timely statistical information on the population is extraordinarily high. But I’m frequently reminded in my areas of interest of the limitations of data that only provides a snapshot of a single day every five years.
For example, the usefulness of the census question on the mode share of cycling for the journey to work is limited because census day is in winter when cycling levels are lower and when it’s more likely to rain in the southern states. Surveys conducted more frequently (e.g. annually or on a rolling basis) would provide more representative information.
On the other hand, the census provides data at a fine scale, like the mode share of cycling at SA1 level in inner city areas. Neighbourhood level data can be broken down (subject to privacy concerns) by other relevant variables like income and education; I’d need to be convinced that any survey envisaged by the ABS will do that well enough.
Information is so valuable socially that a good case can be made for the ABS to be funded to run an expanded program of surveys in addition to the existing five year census cycle. However continuing the quinquennial schedule makes no sense if the benefit of the census relative to other options is as marginal as the ABS seems to be arguing. Half a billion dollars or more might seem small in some contexts but it’s an enormously large sum if it’s effectively wasted.
The federal government needs to make its position clear well before this year’s budget so the terms of the debate are understood. I’m wary of taking a definitive view just yet but I don’t think it can be automatically assumed that moving to a decennial census cycle is self-evidently a bad thing.
It’s vital that any decision isn’t rushed and that the hundreds of thousands of direct users of census data are consulted. The idea of deciding forthwith to cancel the 2016 census just 18 months out is a bad one.
This is such an important debate it needs to be about what happens with the 2021 census, not next year’s. The financial strains on the ABS might not be solely or even primarily its fault, but the federal government needs to find the funds to make sure the 2016 census proceeds and is executed at the high standard we expect of the ABS.