Brian Rudman: Census stats down for the count

Illustration / Peter Bromhead

COMMENT: Statistics New Zealand churned out a press release the other day promising that "New Zealanders can be confident the 2018 Census will produce accurate and high-quality data which can be relied on by communities and decision makers." Yet the smoke signals rising from the Wellington HQ suggest otherwise.

The "digital-first" approach adopted for the March 6 2018 Census relied on Kiwis filing by computer in order "to improve data quality." The dregs were to be mopped up by door-knocking afterwards. But in the end, only 82 per cent of us went on-line, and the follow up produced "full or partial information" for only another 8 per cent. That left one in ten of people in New Zealand on census day - around 400,000 - unrecorded.

The release of any results has now been postponed six months until March 2019 while the statisticians seek help from "Big Brother" – the assorted databases and records assembled by various Government Departments who keep tabs on some or all of us.

In a process called "imputation," the statisticians will trawl through the records of Government Departments such as Internal Affairs, Business, Labour, Education, Inland Revenue - even dredging through Tenancy Bonds - in an attempt to create an accurate picture of the missing ten per cent.

Putting on a brave front, Census general manager, Denise McGregor, says her boffins will "produce a high-quality dataset by making use of reliable government data to fill in gaps." Yet a departmental imputation trial, comparing 2013 census data with data from government departments published in 2017, suggests otherwise.

It concluded that "the estimates are not currently meeting all the quality standards." It recorded evidence of both undercoverage and overcoverage of population groups, noting in particular, how basic address records are often out of date.

It also highlighted how, in using government records to try and fill in the gaps, sub-population groups can be inaccurately reported. Females over 70, for example, and student numbers.

In North Dunedin, for instance, students were underestimated by 27 percent, which, says the report, "strongly suggests that the administrative data sources are not accurately recording the correct addresses of all of those students." Other student towns like Palmerston North and Hamilton showed similar patterns.

Dry and boring as this might sound, Census information plays a vital role in New Zealand's governance. Over $10 billion of health funding allocated to district health boards, for example, is shared out on the basis of census statistics. Electoral boundaries, due for updating before the next election, are also redrawn on the basis of census figures. Ms McGregor says the delay "will have little impact" on the upcoming review of electoral boundaries by the Electoral Commission.

Former Labour Party President, Mike Williams disagrees. He says the creation of new boundaries, which should have begun in November, has now been delayed until April next year, "trashing the timetable political parties have traditionally operated in the selection of candidates in the lead-up to the next general election."

He reckons "the people who will have been missed by this failure of design and execution will be disproportionately Maori, Pacifica, and poor," arguing that the comfortably-off middle-class Wellington public servants who designed the on-line census project, "have only just worked out that sixty dollars a month for an on-line connection is simply not priority spending when you struggle to put food on the table and pay your rent."

Even though the National Government signed off on this failed digital experiment, State Services spokesman, Nick Smith can't resist leading with his chin, referring to the current "shambles."

He highlights "reports that high need people including the elderly, those in rural communities and those with disabilities faced greater problems in participating" as "particularly concerning." What a shame he and his Cabinet colleagues didn't anticipate this when they ticked through "the digital-first approach."

So what now? Williams suggests we "junk the flawed data and run the whole thing again." With much of the budgeted $121 million presumably already spent, that seems a bit extreme. But instead of trying to fill in the 400,000 gaps using Big Brother's flawed files, wouldn't it be simpler to return to basics and door knock the addresses concerned.