How the SNP botched Scotland’s census

Scotland’s first census results have finally been released: just 444 days after England managed to publish theirs. The once-a-decade count of the population was disastrous at worst and botched at best. As the first deadline for returning the census loomed last April, some 700,000 households – a quarter of the country – were threatened with £1,000 fines for not completing it. It had taken over a month to reach a 74 per cent response rate. Eleven years ago it took just ten days. Now that the results are in, the final response rate was 89 per cent: well short of the official target of 94 and the 97 per cent achieved in England and Wales.

When the first deadline was extended, the SNP did not apologise. Instead they blamed the media and ‘anxiety’ caused by Russia’s invasion of Ukraine, the cost-of-living crisis and Covid. They are yet to explain why none of this put 97 per cent of English households off.

Something had to be done to boost the return rate. The census is the basis for a whole range of national statistics that guide policy and direct decision makers. Examples of these estimates failing include Scotland’s Covid vaccinations statistics. After ministers boasted impressive vaccination rates it emerged that, impressively, some 103 per cent of over-70s in Scotland had been vaccinated. The error had been caused by an underestimation of the headcount in certain age groups. It’s easy to laugh at, but it’s also worrying because it meant we didn’t know the true vaccine coverage among vulnerable groups. The lower return from this census will now lead to more uncertainty, meaning it will be harder to judge how many schools, hospitals and libraries should be built.

Yesterday the National Records of Scotland’s new chief Janet Egdell was put in front of cameras to tell us there are ‘lots of lessons to learn’. As to what went wrong, that was not what her organisation is currently ‘focusing on’. Was it embarrassing that England and Wales achieved 97 per cent compared with Scotland’s 89, a reporter asked. ‘We will certainly be looking at that.’ Quite. And on whether the next Scottish census should realign with the census across the UK? The woman in charge did not yet have a view.

Scotland’s low response rate was dreadful by international standards too. Canada managed 98 per cent, Australia 96 and the USA used a follow up programme of non-respondents to get to nearly 100 per cent.

Multiple things seem to have gone wrong. Scots complained of spending hours filling out the online form, only to receive angry letters weeks later reminding them to fill it out in time. Certain women’s groups boycotted the poll over the wording of gender self-ID questions.

The nature of the census – carried out online for the first time ever – seems to have contributed to the problems too. People struggled to use the website, there were issues with translations and marketing wasn’t clear. Some more emotive questions (like the below) did not have a box for people to write their own answer in, as was found in the paper version. Options provided for white ethnic background don’t include English, for example, despite some half a million Scots being of English origin in the previous census.

The most obvious contributor to the botched outcomes must surely be the decision to delay the count. During the first lockdown Nicola Sturgeon’s government chose to delay the census citing Covid restrictions. It meant it would be the first census in 150-years that didn’t simultaneously cover all parts of Britain (Scotland always had a separate census but it was timed to align with England’s). At the time the decision was described as ‘an act of scientific vandalism’. The result was missing out on the UK-wide advertising campaign which would have considerably helped take-up.

Indeed, in recent years the NRS, ONS and Northern Ireland’s stats body (the NISRA) agreed a four nations aim that stated: ‘The 2021 Census outputs should constitute consistent, coherent and accessible statistics for the UK, individual countries and geographic areas within each country.’ Northern Ireland used the same computer systems as the ONS, saving taxpayers millions. Would Scotland have avoided its problems if Scotland shared UK systems too?

Warnings from abroad were ignored too. In 2019, New Zealand's chiefsStatistician resigned after the census had a return rate of 83 per cent. Response was particularly low among the Māori population. A review said the failure led to a ‘significant data gap’ due to ‘too much emphasis on the online census.’ Those warnings were ignored in Scotland.

The NRS CEO, Janet Egdell replaced Paul Lowe (my old boss) who resigned just weeks before Audit Scotland published a report into the handling of the census. The inquiry found that 35 per cent were ‘too busy’ to fill it in, 16 per cent were ‘not aware of the census’ and another 14 per cent ‘didn’t realise they had to complete it’. That could have been ameliorated had the census been conducted a year earlier – and piggybacked onto a UK wide marketing campaign.

Although I did not work directly on the census, my role at the NRS heavily relied on census data and I shared offices with census staff. It was clear in the run-up to the collection period there were going to be problems. Internal messages talked of problems recruiting for the ‘contact centre’ that would field calls for those with difficulties filling in the online questionnaire. IT staff spoke of problems testing and setting up the call centre’s systems too. Political pressure played its part in question design too. 

One day a group of senior civil servants returned from a select committee grilling, crestfallen that the MSPs questioning them had failed to understand the statistician’s points about the importance of biological sex in certain questions. But politicians were far more interested in gender self-ID (this was years before Isla Bryson and the Gender Recognition Bill). Work my team was involved in – creating anonymised datasets for researchers – could be affected by too much freedom in the sex questions. The census was often used to create cohorts for large studies and it’s easy to see how medical research looking at cervical cancer – for example – falls down if someone who has never and could never have a cervix ends up in the dataset.

On the morning of the extended deadline – which cost the taxpayer an extra £6 million – the return rate had crept up to 86 per cent: still far short of the 94 per cent target. Nicola Sturgeon had admitted that the results could be ‘unreliable’ whilst academics went further warning the data could be ‘useless’. Meanwhile, Angus Robertson, the minister in charge, jetted off to Brussels. 

Robertson had previously claimed that media coverage of the census had inadvertently ‘undermined’ the census project whilst some SNP commentators suggested unionists were boycotting it to embarrass the government. The opposite was in fact true. As the below graph shows at the point of the census deadline, the higher the Yes vote in 2014 the lower the census return rate. At that point Glasgow had the lowest census return rate with a fifth of Glaswegians failing to respond. Dundee, the city that had the strongest Yes vote had the third worst return rate.

But of course, what this data showed was not that there had been any boycott to any meaningful extent, but that deprivation made an online census challenging. Access to internet is poorer in some areas than others, meaning that door-knocking headcounts was always going to be crucial. According to Audit Scotland though, one of the main reasons for delaying the census a year – and losing out on the UK marketing campaign – was because the estimated 3,500 strong field staff required wouldn’t be able to work during Covid. This was half the number required in 2021. In the end, only 1,250 were recruited.

Perhaps the real reason for the delay and lack of door-knockers was about image. Sturgeon presided over the strictest lockdown controls anywhere in Britain. Census collectors would not have been able to move between local authorities. Despite all this, Scotland’s Covid record was no better than England’s.

Considering those mistakes, it suited the SNP government under Nicola Sturgeon and now Humza Yousaf to point to the notional independence of NRS. Anything wrong was therefore down to independent civil servants. But on the inside there was clearly lots more political involvement. In the early stages of the pandemic for example, before I left to become a journalist, I remember a meeting where it was announced a statistician was to be taken off duty one-day a week to produce a Scotland-England Covid comparison that Sturgeon could use in her daily TV briefings. Those briefings proved crucial to the SNP’s re-election in 2021 and caused great anger in the First Minister’s office when the BBC eventually bumped them for Bargain Hunt.

In the end, extensions and imputed data had to be used resulting in the census operation costing nearly twice as much per head of the population (if we really know how many people that is) as it did in England and Wales. The census outputs narrowly achieved National Statistics status – the gold stamp for official stats but doubt remains over the quality of guesswork data.

Some regard the transgender prisoner Isla Bryson being placed in a women’s prison as the moment that led to Nicola Sturgeon’s downfall. But the botched census was the beginning. It’s hard to govern a country if you’re not sure who’s in it, what they do and how they live. If more than 10 per cent of the population has been computer-modelled from other sources, i.e guessed, then how reliable will Scotland’s data be? The SNP have gotten away with scandals and policy failures that would have brought down most Westminster governments. But their inability to produce an accurate count of the country they govern was the moment that the electorate woke up to the government’s woeful record on delivery.