An urgent program to collect up-to-date data about the whole community is the only way to know if it is safe to lift the costly lockdown without the virus resurging.

This plan will be informed in significant measure by a complex model being developed to forecast the evolution of the pandemic, particularly in Australia. And the quality of the forecasts will depend critically on the quality of the data supplied to the model.
Fortunately, there is a long-established and cost-effective statistical approach to acquiring reliable data for such purposes, and it is relatively straightforward to implement. It involves selecting people at random and testing them to identify which categories they fall into. While there is some complexity in the detail of how this is done, it would appear to be perfectly feasible to conduct such testing, given the legitimacy of random testing for alcohol or drugs. (A slightly unusual feature of such sampling would be that children are eligible to be selected as well.)
By randomly testing people every week, we can capture invaluable information.
The most obvious question that arises is: How many people would need to be tested? With Australias population approaching 25 million, would 5 per cent or 10 per cent or 20 per cent need testing? If so, the proposal is simply not feasible. However, this is not the case. We are talking about tens of thousands of tests over the coming months, not millions.
By randomly testing people every week, we can capture invaluable information. We can build up estimates of the proportions of the Australian population in each category, and how these proportions are changing with time.
We can use demographic information about people sampled to understand differential risks of different groups according to age, gender and socio-economic factors. For example, how many children are asymptomatic? What proportion have passed it on to their parents? Experts have suggested that the majority of “asymptomatic” cases are in fact “pre-symptomatic” cases but are uncertain. Our proposed survey would inform this important question. The answer is vital for epidemiological models.
Various enhancements might also be contemplated, such as following up some of the selected individuals who have tested positive to the virus, either asymptomatically or symptomatically, to check on the possibility of reinfection and, if it occurs, the rate.
Many of the basic skills and capabilities required to run a suitable sampling process are already available in the Australian Bureau of Statistics. These will need to be supplemented by a team of high-calibre statisticians and epidemiologists, especially for the analysis of the data.
Of course, there are issues that must be resolved, not least that of respecting, as far as possible, peoples privacy and an appropriate level of confidentiality about the data collected. Also, and obviously, whatever test swab or blood test is used to decide on the presence or otherwise of the virus must have a low level of false positives and a low level of false negatives in other words, correctly identify the individuals status most of the time.
Other countries are starting to move down similar paths. Germany plans to start a large-scale national survey at the end of April using random samples that will also include tests on the presence of antibodies. And a research firm in Iceland running a national survey has found that 0.3 per cent of the population had tested positive to COVID-19, much lower than previously thought.
We suggest the following actions as a matter of urgency. (1) Establish a team with relevant statistical and epidemiological expertise. (2) Commission it to produce a suitable sampling plan and protocols. (3) Deploy a public communications plan, including a prime ministerial announcement. (4) Start collecting data. These actions could be accomplished within a few weeks.
The government needs to act on this now. The welfare of all Australians is at stake.
Nicholas Fisher is a former chief research scientist in statistics in CSIRO. Dennis Trewin was the Australian Statistician from 2000 to 2007. Both are former presidents of the Statistical Society of Australia.