Census time

The West Island had their census last night, so the Australian Bureau of Statistics is in the process of collecting millions of little bits of paper from around the country. It’s a good occasion to think about what the census is actually good for, because this goes further than you might think.

The most difficult problem with surveys is uneven response.  Whether you just stick some buttons up on a web site, or mail out questionnaires, or send interviewers around in person, some people will respond and others won’t.  Survey methodologists have come up with lots of ways to increase response, but the most effective ones tend to be expensive.

If we know enough about who is more likely to respond and who was less likely, it’s possible to correct for a lot of the bias.  For example, news.com.au have carried what they are calling an “Alternative Census”, tackling the big issues like AFL vs rugby, or how people find Facebook friends.   The people who responded to this survey are not remotely a representative set of Australians.  65% of them were men, and the state with the largest number of responses was Queensland, with 27%.    The over-representation of Queenslanders and men means that the respondents are likely to think rugby is more important than the typical Australian does.   Since we know the proportion of men and the proportion of Queenslanders in the whole Australian population, we can rescale the responses so they only represent their fair share of the population.

For this sort of non-response adjustment to be useful we need to collect quite a bit of information, and we need it to be complete, or at least representative.  That means we need to run some sort of regular survey where the non-response is very low.  The way we do this in NZ and Australia is with a census, where participation is mandatory and lots of resources are devoted to making sure nearly everyone takes part.  In the Scandinavian countries there are large adminstrative databases that routinely collect so much individual data that a census is unnecessary.  The US has a hybrid system, where they have a census every ten years that asks just a few questions, and a more detailed micro-census of 1% of the population per year (with mandatory participation) called the American Community Survey. And Canada went for a similar system to the US except that they got it wrong.

It’s not clear yet what will happen to the NZ census after the ChCh earthquake. We might just resume the five-yearly schedule in 2016, but the Government might decide to switch to a ten-year cycle, perhaps with interim surveys following the US model.   In any case, some form of regular, fairly detailed, survey of the population, with near-100% participation by the people who are selected, is important. It’s not just to find out the answers to the census questions, but to  help improve the accuracy of all the government, commercial, and academic surveys conducted in NZ.