The polls and the pandemic: double trouble for Boris Johnson

A pincer movement threatens the government’s hopes of emerging from the Covid-19 crisis with its reputation intact. One half of the pincer is provided by the latest polls, showing that decline in support for the government is not just continuing, but accelerating. The other half of the pincer is evidence that the government is simply wrong when it draws on official statistics to say that only a tiny minority of people are defying the lockdown rules.

First, the latest polls over the weekend two told identical stories. Opinium’s polls for the Observer found that, for the first time, more people disapproved (42 per cent) than approved (39 per cent) of the government’s handling of the crisis. This net score of minus three compares with plus 12 just a week earlier, and plus 42 eight weeks ago at the start of the lockdown.

When a single poll shows a dramatic short-term shift, there is always the chance of a sampling glitch, and that the true story is more mundane. But the separate YouGov poll for Sky News confirms the Opinium results in every particular. Asking a similar question, YouGov also shows the government slipping into negative territory, with a net score of minus three, and a sharp drop from a net rating of plus 51 at the start of the lockdown.

In short, the first week of the new rules has been a disaster. Voters tend to think the Government has gone too far to relax the lockdown, and caused confusion with its new slogan, “stay alert, control the virus, save lives”.

Meanwhile, Keir Starmer’s rating continues to rise. Opinium’s latest poll gives him a net rating of plus 24 (approve 42 per cent, disapprove 18 per cent). This compares with Boris Johnson’s net rating of plus ten (46-36 per cent). A week earlier, Johnson’s net score of plus 20 put him narrowly ahead of Starmer’s plus 18. Now Starmer comfortably outranks Johnson in public approval.

Given the volatility in the polls, nobody can be certain that these trends will persist. Perhaps the new rules will be seen to work well. Perhaps parents’ fears of schools reopening will give way to relief if children return to their classrooms without the virus spreading. Perhaps employees will be able to travel to work without clogging up the roads, buses, tubes or trains, and the economy will revive. For the moment, the message from the polls is that the “rally round the flag” mood that swept Britain in late March has vanished. The government now needs to earn the public’s support, not assume it.

Which brings us to the second half of the pincer. The Office for National Statistics publishes regular survey results which show, among other things, that almost everyone is obeying the lockdown rules. Last week’s survey found that just 4.1 per cent of people were “non-key workers” who were breaking the rules by going to work. 1.8 per cent (and there may be some overlap between the two groups) broke another rule by “meeting up with people in a personal space” in the previous seven days.

If true, these are extremely reassuring figures. They support the slides that are shown occasionally at the afternoon press conferences in Downing to demonstrate law-abiding national harmony.

However, research by Decision Technology casts serious doubt on these data. The company has used a technique called “unmatched count” to test whether the answers people give to pollsters on sensitive questions are always truthful.

This is what Decision Technology did. It divided a large sample into ten similar subgroups. The first group was asked eight conventional questions about their behaviour during the lockdown — had they washed their hands for less than 20 seconds; had they failed to keep two metres distant from other people outside their household; had they left their house for non-essential reasons; and so on.  Three different samples are questioned in three different ways.

The second group was given a list of four items and asked how many applied to them:

* I drank a can/bottle of a carbonated soft drink in the past 24 hours (e.g. Coke, 7Up)
* I have gone running/jogging in the past week
* I have read a novel cover to cover in the past six months
* I have been to America

The key thing is that respondents were asked just to give a number, not to identify which particular items applied to them.

Each of the remaining eight groups were shown the same list of four items — and an extra item: one of the eight lockdown actions in the questionnaire given to the first group.

The idea behind “unmatched count” polling is that respondents will be more honest if they don’t have to admit to (or boast about) particular actions. Statistically, the research company is unable to find out which respondents have, say, left their home for a non-essential reason, but can deduce how many have done so. They do this by comparing the number of actions declared by the second, four-item, group with those of each of the five-item groups.

For example, if 100 people average 2.5 responses to the four-item list, that adds up to 250 in total. If another 100 people give an average of 3 responses to the five-item list, that’s 300 in all. The difference — 300 minus 250 — indicates that the fifth item applies to 50 of them; in other words, the 50 per cent have undertaken the sensitive action.

These responses can then be compared with the conventional results obtained from the first sample.

For example, Decision Technology’s, normal, questionnaire found that 3 per cent admitted to leaving their home for non-essential reasons — a figure broadly in line with the ONS’s slightly different way of exploring the same issue. But the “unmatched count” method indicated that the true figure was 29 per cent. Other questions also produced large discrepancies: 16 per cent admitted failing to keep two metres away “from other people outside my household”; the “unmatched count” suggests the true figure is 35 per cent. And while a substantial 40 per cent are happy to admit washing their hands for less than 20 seconds, the more anonymous approach puts the figure at 64 per cent.

Does the “unmatched count” technique provide precisely accurate numbers? I don’t know. But I reckon it gets us nearer the truth than conventional polls on sensitive issues. In the past, this method has indicated that more people have cheated on their partners than they admit in a normal poll — and that conventional polls significantly overstate the extent to which people take the environment into account when they go shopping. These seem entirely plausible findings.

What’s more, the truth matters. If ministers are taking decisions on the basis of false information, they are liable to take the wrong decisions. Knowing what is actually going on is not just a democratic virtue; it is a practical imperative.

This blog was forst published by The Article: