Estimating the UK population during the pandemic

By Michael O’Connor and Jonathan Portes

British jobs for British workers? Over the last year, the number of people with a job and living in London who were born in the UK has increased by more than a quarter of a million. That’s according to the Labour Force Survey (LFS), the main data source for UK statistics on employment and unemployment, published by the independent Office for National Statistics (ONS). Does it sound implausible that over the course of a pandemic that has devastated the UK economy, and which has hit London particularly hard, with its high concentration of jobs in hospitality, arts and entertainment, and tourism-related businesses, the number of UK born workers living here should have soared by 10%? Indeed it does. So why are perhaps the most important and widely used set of statistics we have telling us this?

The answer is to be found not in the data about UK-born workers, but rather in the immigration statistics, or more precisely our lack of knowledge about what is happening to immigration. This was already a serious issue before the pandemic, with the quarterly international migration statistics losing their “National Statistics” designation (after one of us  pointed out, in a previous article for ESCoE, their inconsistency with the Labour Force Survey). But it has, unsurprisingly, been exacerbated over the last year, with the (understandable) suspension of the International Passenger Survey because of the disruption resulting from the pandemic.

So for information about what’s happening to immigration and to migrant workers in the UK we have to look to other data sources, such as the LFS. ONS are doing a considerable amount of work, looking in particular at administrative data, to improve migration statistics. But this will take time to come to fruition, and requires triangulation across a number of different sources. Meanwhile, it is still informative to examine what the LFS is implying about migration flows.

And this gives a clear – and astonishing – message. The number of non-UK born workers resident here has fallen by more than half a million. Nor has this fed through into a rise in unemployment and activity – overall, the number of non-UK born residents has fallen by even more. And while the LFS does not and is not intended to measure directly migration to or from the UK, there is no doubt that as far as the LFS is concerned, there has been an unprecedented fall in the number of foreign-born residents in the UK, from which an unprecedented exodus could be inferred. But conversely, the data suggest that the UK born population has considerably increased such that despite the economic disruption resulting from the pandemic, the number of UK born workers (and non-workers) hit new record levels, as the chart (covering essentially the working age population as a whole) shows.

And while some of this may well be due to the difficulties of conducting surveys during the pandemic, which might make it even harder than usual to contact people who haven’t been here very long, and who live in private rented accommodation (as many recent migrants do) that doesn’t, at least so far, appear to explain the fall. It seems more plausible that people actually are leaving. And given the nature of the pandemic and its economic and social impacts, that makes sense. 

Migrants, especially from Europe, are disproportionately likely to be employed in the hospitality sector, and other service sectors that require face-to-face contact, so are more likely to have been furloughed or lose their jobs. With many universities moving wholly or largely to on-line teaching, many foreign students may have decided not to come to the UK or to return here. But most of all, the UK has (alongside a few other western European countries such as Spain and Italy), performed relatively badly in both economic and health terms during the first wave of the pandemic. For many migrants, especially those from  eastern, central and south-eastern Europe and especially those who have arrived recently or have family back home, the choice would have been to stay here, with no job, less or no money, and pay for relatively expensive rented accommodation – or return home to family, with lower costs and most likely less risk of catching Covid. Not a hard choice.

So it seems that much of the burden of job losses during the pandemic has fallen on non-UK workers and has manifested itself in return migration, rather than unemployment. This is in itself an important insight, and, alongside the furlough and other business support schemes, helps explain why, despite the very large and continuing hit to GDP and output, unemployment has not as yet soared to the levels some predicted.

But how does that explain the *increase* in UK-born workers? It’s hardly plausible that the returning migrants are leaving vacancies for them to move into, since it’s the lack of jobs that is causing the exodus in the first place. The answer instead lies in the way the LFS estimates are constructed. It’s a survey, not a count. By definition, then, it can’t tell us anything directly about how many people are employed or unemployed in the whole population – only about the proportions. That, then, needs to be scaled up to the whole population. This is done using “grossing factors” – each person who responds to the LFS is given a “weight”. These weights are constructed so that not only do they sum up to the whole population, but that the sum matches up to the population by age, sex and region. So if, for example, fewer people than “expected” respond in London, those who do are assigned a larger weight. 

But how large is the whole population? The LFS survey data doesn’t tell us, because it’s just a sample. Instead, the ONS assumes a number for the current population, based on what we know about births, deaths and population ageing – and on some assumptions, based on historical data, about levels of immigration. Crucially, the ONS has not made any changes to its population projections to take account of Covid-induced emigration – even though, as we’ve just explained, it has fairly clearly reduced the resident population by some hundreds of thousands.

So what does that mean in practice? As set out above, the assumption underlying the estimates is that the total population hasn’t fallen, but instead that it has increased on the previously forecast trajectory. And the LFS data also say that the number of non-UK born people has fallen by nearly a million. It follows, arithmetically, that the number of UK born people must have increased – even if there’s no plausible way that could actually have happened in reality to anything like this extent. But that’s precisely what the published statistics show. And moreover, because of the way the weights work, that applies at a regional level as well.

And that’s what explains the striking London statistic we quoted above. As this chart shows, what this methodology does is, in each region, to “create” (using the weightings) new UK-born people to “replace” the non UK-born ones who have left the country and to grow the population to the previously forecast level. And because London has by far the largest share of foreign-born residents, by far the largest share of those who appear to have left, and was – under the pre-pandemic projections – expecting to see significant population growth. London has therefore been assigned the largest share of the new Britons: more than 400,000 in total.

So what would the data look like if instead of increasing the weights to maintain the size of the measured population regardless of any exodus, we assume that the reduction in non-UK born residents was real? Such calculations can only be rough and ready. We can, however, approximate, by adjusting the grossing up factors (the weights) in such a way as to allow the observed reduction in the non-EU born population to translate into a reduction in the overall population, rather than an increase in the UK-born population.

Technically (see Box), we do this by assuming (for each age-region-gender cell) that the UK-born population has remained constant, and that the changes in the ratio of UK and non-UK born population in the survey reflect actual changes in the resident population. The former assumption seems reasonable: the original, pre-pandemic ONS projections assumed a small increase of about 100,000, reflecting natural change (births and deaths) and some net emigration. In fact, given the pandemic (which has led to about 75,000 more deaths than expected) there seems to be no reason to expect any significant change to the UK-born population from natural change. It is possible that – just as non-UK born people appear to have responded to the pandemic by returning to their countries of origin, so UK-born people abroad have returned to the UK. But there does not seem to be any strong evidence suggesting this impact is material. It seems implausible that if hundreds of thousands of emigrant Britons had returned, this would not be evident from news reports. Moreover, any such effect has not been seen in the number of state pensions in payment, which has, sadly, fallen, in part for obvious reasons.

The second assumption is obviously more questionable – it is plausible to imagine that the pandemic, and associated changes to the way the survey data is collected, might have affected non-UK born people still resident here more than others, so that the differential reduction in the non-UK born that we see in the data in part reflects differential increases in non-response. However, we see no evidence of this in the data, so our assumption seems a reasonable baseline. To the extent that it is in fact nevertheless the case, our calculations will over-correct.

Our results are shown below for the change in the total population, by region, split between UK born and the non-UK born. We show the difference between published LFS figures for the third quarter of 2020 and third quarter of 2019, and our own adjusted estimates. The results are dramatic. Instead of the UK-born population rising by over a million, it is essentially unchanged. The non-UK born population falls by even more than in the published estimates, reflecting in particular what the data suggest about trends in London, where a large proportion of non-UK born people live.

These estimates are crude and illustrative, and will certainly not be accurate. In particular, as noted above, to the extent that the pandemic has reduced response rates among non-UK born residents more than others, they will overstate the decline in that group and overall. However, given both what we know about the impacts of the pandemic and what the raw survey data actually say, we would argue that they are clearly more plausible than the published statistics for the UK-born, which, as explained above, simply don’t make any sense at all. And of course the UK born are the great majority of the population.

So what do our estimates imply? They give us a very different picture of recent population change in the UK, but we would highlight these points (although there are many other aspects of interest)

• Overall, instead of an increase of about 350,000 over the year, the total population falls by more than 1.3 million. While total population was almost flat in the late 1970s, if this is even close to being accurate, this is the largest fall in the UK resident population since World War 2.

• The impact has been largest in London. While the published statistics say that the resident population has grown over the year, our estimate implies that the resident population of London might have fallen by nearly 700,000. Much of this may be temporary, if non-UK born people return to London after the pandemic; but it may not. Big shifts in population trends in London, driven by economic changes and events, are by no means historically unprecedented – Inner London’s population shrank by fully 20% in the 1970s, so the recent picture of sustained growth driven by international migration is relatively recent. If this has now reversed, the medium to long-term implications for London will be profound.

• Our estimates do not make much difference to estimates of unemployment or employment rates, as you would expect, since in each cell we are scaling up all those in the survey, employed or unemployed, by the same amount. But they do make big differences to levels.

• In particular, the rise in the number of employees seen in the LFS, of around 200,000 in the year becomes a fall of around 750,000. This is a huge change, and would profoundly alter our perception of recent labour market developments. Is it plausible? In our view it is indeed more plausible than the published figures, corresponding as it does much more closely to the changes shown in the HMRC’s data on the number of people known as employees to the PAYE RTI system (which represents a full count of employees reported to HMRC, rather than just a survey) and which shows a year on year fall of about 650,000. As the chart below shows, our analysis also reflects changes seen at a regional level in the RTI data reasonably well for employees.[1] This suggests that while our approach may somewhat overstate population falls, it does not do so by a huge amount. A possible explanation for the somewhat greater falls in our adjusted series, especially in London, is that the pandemic might have particularly affected workers in the informal economy whose employers might not have been reporting them as they should to HMRC and hence were never recorded in the RTI data in the first place. While this analysis applies only to employees (not the self-employed or those not in employment), it does suggest that our methodology is producing broadly plausible results.

Our work leaves many questions unanswered. And we would stress that these are illustrative calculations rather than a fully worked-through attempt to “correct” the published LFS statistics. They can perhaps be viewed as putting an upper bound on the likely fall in the non-UK born population and that of the population as a whole, both nationally and on a regional basis. Nevertheless, they paint a very different – and in our view much more plausible picture of recent developments in the UK population and labour markets than the published statistics. And they appear to us to be much closer to what common sense and what both qualitative and other quantitative data tell us.

Where next? We should emphasise that this work is not intended as a criticism of the ONS, who face unprecedented difficulties in maintaining the continuity, comparability and reliability of these vitally important statistics. However, we do think it is urgent that ONS should take into account likely population changes resulting from the pandemic in the LFS statistics, and caveat published estimates accordingly. In addition, other data sources could shed further light on these issues. The HMRC data discussed above also includes National Insurance numbers, which in turn can be linked to nationality at time of registration. It would be possible to examine a sample of the data to determine the extent to which the reduction in employees was indeed disproportionately concentrated among non-UK nationals. Similarly, the longitudinal element of the LFS might shed further light onto who is dropping out of the survey and why.

[1] Workers may be employed in a different region to the one they live in; both the RTI and LFS data here refer to region of residence.

Featured image photo by Garry Knight, Creative Commons licensed CC BY 2.0

Michael O’Connor is an independent consultant and an ESCoE Research Associate
Jonathan Portes is Professor of Economics and Public Policy at King’s College London and an ESCoE Research Associate

ESCoE blogs are published to further debate.  Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.