Jobs and job quality between the eve of the Great Recession and the eve of COVID‐19

Abstract In 2019, the employment rate among 25‐ to 64‐year‐olds in the UK reached 80 per cent – the highest on record, and considerably higher than the 76 per cent rate recorded shortly before the Great Recession. In this paper, we investigate the growth in employment between the eve of the Great Recession and the eve of COVID‐19 across several dimensions. We analyse which sectors, demographic groups and regions accounted for the rise. We also investigate how job ‘quality’ – in both financial and non‐financial terms – has changed. We find that almost all demographic groups and regions saw a rise in employment, especially those with low pre‐existing employment rates and those near the bottom of the income distribution. Hourly pay growth was very weak over the period, with the median actually slightly falling. Other indicators of job quality show a more mixed picture: employees seem to have greater appreciation of their work and firm, but perceive less security and flexibility in their job.


I. Introduction
On the eve of the COVID-19 crisis, employment stood at its highest rate since records began, having increased strongly since the Great Recession. In this paper, we investigate what the inter-crisis period meant for the number and nature of jobs in the UK, and how this set the backdrop for what is happening now, including the economic vulnerabilities we face in the current crisis and the policy challenges that it will likely leave us with afterwards. The pre-COVID-19 labour market was certainly not without its shortcomings, and some of those have arguably been exposed further by the crisis. As the economy returns to something more normal, the same issues around pay and job quality will no doubt re-emerge. Meanwhile, there had also been some trends that were undoubtedly helpful in holding down povertymost notably the sheer number of people and households with some paid work. The COVID-19 crisis jeopardises these, and the immediate pre-crisis period provides a benchmark for how much work there may be to do post-crisis to recover ground that has been lost.
In this paper, we analyse which demographic groups, regions and industries saw the highest growth in employment (Section III) and what forces were behind this growth (Section IV). We then quantify the extent to which the increase in employment was associated with changes in job quality (Section V).
Throughout, we mainly focus on employment among those aged 25-64, as employment above these ages is relatively unusual (though becoming increasingly common), and trends in employment below 25 are complicated by more people staying in education for longer.

II. Data
We use four data sets in this paper, which we now describe in turn.

Labour Force Survey
The main data set that we use is the Labour Force Survey (LFS), for which the latest available data cover 2019. It is a quarterly survey of the UK population, with detailed information on labour market behaviour. The sample size in 2019 was around 53,000 individuals per quarter.

British Social Attitudes Survey
The British Social Attitudes Survey (BSAS) is a yearly survey of over 3,000 individuals collecting data on a wide variety of social, economic and political issues, includingin 2005 and 2015attitudes towards work and job satisfaction.

Family Resources Survey
For household income analysis, we use the Family Resources Survey (FRS), an annual survey of around 20,000 households with detailed information on incomes. We can simulate counterfactual incomes for FRS households using TAXBEN, the IFS tax and benefit microsimulation model (Waters, 2017). The latest data cover the financial year 2018-19.

Annual Survey of Hours and Earnings
The Annual Survey of Hours and Earnings (ASHE) is an employer survey of 1% of employee jobs (giving a sample of roughly 300,000) in April each year. It contains relatively precise data on earnings and hours worked, making it a useful source for understanding trends in hourly wages.

III. The rise in employment
In this section, we describe the magnitude and nature of the growth in employment that was observed between 2007 and 2019. Subsection A presents longer-term trends in different types of employment by sex for those aged 25-64 and Subsection B shows what this will have meant for poverty.
Subsection C then investigates employment trends across demographic subgroups over the period, while Subsection D shows how the distribution of workers across industries has changed over time.

III.A Trends in employment rates
In 2019, around 28 million individuals aged 25-64 were in work, an employment rate of 80%the highest since records began in 1971. As Figure 1 demonstrates, employment had increased strongly over the previous eight or so years, leaving the rate 4.5 percentage points (ppts) higher than its 2007 level of 76%. Most of the employment growth was driven by women. Female employment in 2019 stood 7ppts above its 2007 level, while male employment was just 2ppts above. This served to close the gender employment gap from 14ppts to 10ppts. This can partly be explained by the huge change in working patterns at particular points in the life cycle observed for women over time, with far more women in their mid-to-late 20s and early 30s being in work in 2019. Women are having children both less frequently and later in life than they used to. When they do have children, they are also less likely to drop out of the labour market and tend to return to work faster (Roantree and Vira, 2018). In Section IV, we will explore two significant policy changes that especially affected women's labour supply.  1993  1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019 Male Female All Figure 2 demonstrates that much of the increase in the overall employment rate from 2007 to 2019 was driven by full-time employment. This is particularly true for female full-time work, which grew by 6ppts. In contrast, full-time male employment had only just returned to its pre-2007 level.
Roughly two-fifths of the growth in employment was seen in self-employmenta part of the workforce at which the government has found it especially difficult to precisely target insurance during the COVID crisis. While the share of the population working as employees dipped in the aftermath of the financial crisis, self-employment rates continued to rise. Cribb and Xu (2020) show that the rise in self-employment since 2007 was entirely driven by an increase in the 'solo selfemployed', who operate on their own without employees. The income gains from the employment growth between 2007 and 2019and particularly from falls in household worklessnesswere very much concentrated at the bottom. Though it is difficult to precisely estimate by how much employment growth will have kept poverty down, employment growth clearly was a key driver of income growth among low-income households. It is therefore likely to have been a significant factor in keeping poverty lower than it otherwise would have been.
The danger in the current crisis resulting from the coronavirus pandemic is that much of this may be undone, likely leaving many low-income households vulnerable. Looking at the types of people who were brought into work from 2007 to 2019 and the quality of their jobs can therefore help provide a picture of the pre-coronavirus-crisis circumstances of those below and around the poverty line.

III.C Change in employment rates for different groups
We now turn to exploring in more detail where the growth in employment from 2007 to 2019 occurred.
Overall, the largest increase in employment was experienced by population subgroups that historically had lower employment rates. These include women, ethnic minorities, lone parents, older people and immigrants. 1 This can be observed in Figure 4, which presents the changes in the employment rates for different demographic groups from 2007 to 2019. 2 Notably, employment rates increased for all of these groups 3 with the exception of individuals with a degree, for whom the employment rate decreased by 1ppt. As discussed in further detail in Section IV.B, this is likely at least in part due to the huge rise (4.7 million) in the number of 25-to 64-yearolds with a degree between 2007 and 2019. This means that degree holders are simply a different kind of group, on average, from the group they were in the past, which probably makes accurate like-forlike comparisons over time impossible (the same is likely true among different groups of those without a degree).
Three groups saw particularly large increases in employment. First, single mothers' employment increased by 12ppts, a rise partly caused by policy reforms incentivising paid work (as discussed in Section IV.A). 4 Most of this rise was in part-time employment (see Appendix Figures A1 and A2). It is worth noting that employment amongst lone parents, a central part of the Labour governments' 1 Note that we do not present the employment rate for people with and without disabilities, given that there have been multiple changes in the survey questionnaire with regards to the definition of being disabled and we thus are not able to construct a consistent measure of disability over time (see figure 5.1 of Cribb, Norris Keiller and Waters (2018)). 2 The extent to which changes in the employment rate by specific characteristics will have affected the overall employment rate will also depend on the number of people in each subgroup who were initially in work and the relative sizes of these groups. Appendix Table A1 shows the change in the number of people in each subgroup who are in work from 2007 to 2019, while Appendix Table A2 presents the change in the number of people in each subgroup as a share of the total population. 3 Excluding immigrants from the results presented does not change results substantially. 4 Single fathers saw an even larger rise, but are a very small group. child poverty strategy, had already increased from 47% to 57% in the decade leading up to the Great Recessiona large increase of over 20%.
Second, the employment rate among those aged 55-64 increased by 9ppts. Again, a particular policy reformthe rising female state pension ageplayed a key role and is discussed in Section IV.A.
Given this, it is not surprising that increasing employment among older workers was stronger among women (+13ppts) than men (+5ppts).
Third, immigrants saw a 10ppt rise in employment. Together with an increase in the number of immigrants in the UK, this led to around 2 million more immigrants in work. In Section IV.C, we explore whether the increase in the employment rate for immigrants can be explained by a change in their composition. The employment rate for non-immigrants also increased from 2007 to 2019, albeit to a lesser extent (4ppts). A common question in the policy debate is whether the substantial increase in the number of immigrants in the UK affected the employment of non-immigrants. While it is possible that the increase in the employment rate for non-immigrants would have been higher in the absence of immigrants, in general the empirical evidence suggests very little employment effects of that kind (Dustmann, Fabbri and Preston, 2005;Wadsworth, 2018).   Figure 5 presents changes in the employment rates between 2007 and 2019 by region. Every region shared in the employment growth over this period, though with considerable variation. On average, the lower-employment parts of the UK saw faster growth (with the exception of Northern Ireland and the North East).
The change in employment rates of non-immigrants by region was similar to that of the total population, with the exception of London, where the employment growth for non-immigrants was considerably (3ppts) lower. We do not find substantial differences in employment growth across regions by sex: for every region, female employment growth was around 2-7ppts higher than that of males (overall female employment growth was 4ppts higher). The preceding evidence has shown the largest growth in employment among women, immigrants and Londoners. One might ask whether the especially strong growth among these groups masked reduced 65% 67% 69% 71% 73% 75% 77% 79% 81% 83% 85%

III.D Change in distribution of workers across industries
So far, we have investigated the sorts of people who saw increases in employment over the period; we now turn to the industries they worked in. Table 1 shows the change in the number of workers in each industry in absolute terms and as a share of the total workforce. Two public-sector-dominated industrieshealth and educationboth saw significant increases over the period. An ageing population is likely to have boosted work in the former, with the strongest increases among those working in residential and social care. The rise in the number of people working in education (0.5 million) appears to have been driven by those working in the private sector or universities, with the increase there considerably greater than the rise in the number employed in public sector education (0.1 million (Cribb, Davenport and Zaranko, 2020)).
There were also significant increases in the number of workers in the hospitality sector, in particular catering, as well as in professional activities, with more people working in management consultancies, head offices, engineering and architecture. There were falls in the share of workers working in wholesale, retail and transportation, and in manufacturingwith the latter being the continuation of a long-run decline in the sector.
Overall, these trends show increases in typically lower-paying industries (such as accommodation and food services or human health) as well as higher-paying ones (such as professional services). Section V looks at this issue further, investigating whether the increase in the employment rate from 2007 to 2019 was accompanied by a change in the pay and quality of jobs.

III.E Younger workers
Thus far, our focus has been on workers aged 25-64. We now briefly investigate employment patterns among younger workers (16-24). While almost all demographic groups saw an increase in employment from 2007 to 2019, those aged 16-24 saw a 3ppt decline. We examine this decline further in Table 2, which lists the economic activity of 16-to 24-year-olds in 2007 and 2019. There are two things to note from the table. First, because unemployment fell across the period, the decline in labour force participation (employment or unemployment) was considerably larger than the fall in employment alone. Second, this decline was entirely accounted for by an increase in the share of 16-to 24-year-olds in full-time education (likely driven in part by the raising of the school (or training) leaving age). Among those not in full-time education, the 16-24 employment rate increased by 2ppts.

III.F How did labour market trends affect economic exposure to the COVID-19 crisis?
At the time of writing, it had become clear that the strong employment growth the UK experienced over recent years is likely to haveat least partially, and at least temporarilyreversed as a result of the COVID-19 crisis. Firm shutdowns, economic uncertainty, and temporary falls and possibly permanent changes in demand for particular goods and services are all factors that will have led to job losses and could lead to more (Costa . In this subsection, we look at the extent to which the growth in employment between 2007 and 2019 was concentrated among jobs and individuals that are more or less vulnerable to the employment impacts of COVID-19.
We identify three types of workers who are particularly at risk of being unable to work as a result of the crisis: first, those working in a sector that has been largely or entirely shut down because of the lockdown measures, including air travel, hospitality and non-food retail; 5 second, workers with young children and no non-working adult in the household, who may struggle to find childcare; 6 and third, those whose job makes it difficult or impossible to work from home. 7 Conversely, one groupthose classified by the government as key workersare less exposed financially to the COVID-19 crisis, but often more exposed to health risks.
Previous research has shown that certain types of individuals are particularly likely to be in these sorts of groups. Younger workers, low earners and women are more likely to work in shut-down sectors (Joyce and Xu, 2020). Low earners are also less likely to be able to work from home, whereas those living in the South are more likely (Costa . Key workers are predominantly female and lower-earning Sibieta, 2020a and2020b). We add to this evidence in Table 3, which shows the growth in employment from 2007 to 2019, split into the four categories described above.
All of the growth in (16-64) employment from 2007 to 2019 can be accounted for by additional jobs that can be done from home. The workforce has also shifted towards key workers. These changes make workers as a whole more resilient to the COVID-19 shock than they would have been in 2007.
However, both in absolute terms and as a fraction of the workforce, more people are at risk of being unable to work because of childcare responsibilities. This is unsurprising given the large increase in employment amongst parents. Furthermore, there was a slight increase in the number of people working in shut-down sectors. 7 We define an individual as being unable to work from home if they (a) have an occupation where, according to Costa , fewer than half of workers can work from home; (b) are not a key worker; and (c) do not have a vulnerable person in their family. When we zoom in on younger workers not in full-time education, the patterns are a little different. As shown by Joyce and Xu (2020), they were particularly likely to have been working in a shut-down sector in 2019. That had only become truer since 2007people had increasingly been starting their careers in occupations such as hospitality (Costa Dias, Joyce and Norris Keiller, 2020). On the other hand, in 2019, fewer younger workers had childcare responsibilities, and they were more likely to be able to do their work from home, than in 2007. They were also more likely to be key workers than they used to be, with the greater financial resilience but greater health risks that this brings in the current crisis.
The increase in the employment rate from 2007 to 2019 was relatively widespread, with almost all demographic groups seeing a rise. Increases were typically larger for those groups and regions with lower employment rates to begin with. While some of this growth was in work that is relatively shielded from the current crisis, the increase in female participation that has driven much of the overall increase in employment means that more workers are at risk of being unable to work because of childcare responsibilities. To what extent and for how long the COVID-19 crisis will hinder their career progression and employment prospects remains to be seen, although the immediate evidence on how much women's work is being disrupted during the crisis is not encouraging (Andrew et al., 2020).

IV. Understanding the rise in employment
In the previous section, we reported the magnitude and nature of the growth in employment seen between the Great Recession and the COVID-19 crisis. We now turn to examining the causes behind that growth. We investigate two causes that can be analysed fairly reliably: specific policy reforms and the changing composition of the population. After discussing these, we look specifically at understanding the rising employment rate of immigrants.

IV.A Policy reforms
There have been a number of reforms since 2007 that could affect employment. Among these are reductions in income tax, increases in VAT, cuts to both in-and out-of-work benefits, changes to work search requirements for some benefit recipients, expansions of childcare subsidies, and sharp increases in the minimum wage for those over the age of 25.
A complete assessment of the employment impacts of these policies is beyond the scope of this paper.
Instead, we focus on two reforms that target specific groups and appear to have had a significant effect: increases in work search requirements for lone parents with young children and the increase in the female state pension age. We examine these in turn.
Prior to 2008, lone parents with a child under the age of 16 were eligible for income supporta means-tested out-of-work benefit. Income support is paid at the same rate as jobseeker's allowance, the UK's unemployment benefit. The difference between the two is that while the recipient of jobseeker's allowance must look for work and meet regularly with a 'work coach' to be eligible, the same is not true for recipients of income support. Between 2008 and 2012, the government implemented the lone parent obligation (LPO), which restricted entitlement to income support in four stages: limiting it first to lone parents with a child aged under 12, then under 10, then 7, then 5. The figure shows that the employment rates of the two groups who were unaffected by the policy (youngest child aged 0-4 and aged 16-18) were fairly constant over the period studied. 9 By contrast, the employment rates of the four groups affected by the LPO persistently increase significantly followingbut not beforethe loss in entitlement to income support. Larger increases are seen for mothers with younger children. For example, the employment rate of single mothers with a youngest child aged 12-15 was about 3ppts higher 18 months after the implementation of the LPO on them, whereas those with a child aged 5-6 saw a rise of 14ppts over the equivalent period. 9 There is a small amount of variation with the recession and recovery, and a modest increase in the employment rates of those with the youngest children in 2013.

Figure 6: Employment rates and entitlement to income support among single mothers, by age of youngest child
Note: The figure shows entitlement or otherwise to income support on the grounds of caring for a young child. Individuals could also be entitled on other grounds, such as incapacity. The figure shows when entitlement to income support was removed for new claimants; existing claimants could continue to claim for a period (determined by their child's exact age and not longer than 14 months) after that point. We can study this policy more formally with a difference-in-difference approach. Specifically, we categorise single women in the LFS into five treatment groups, split by the age of the youngest child as in Figure 6. We run the following regression: where , , is a dummy indicating the employment status of individual in treatment group in quarter . and are treatment group and time fixed effects respectively. , , is a series of four dummies indicating whether is before the LPO was applied to group , 0-5 quarters after, 6-10 quarters after, or 11+ quarters after (these periods are indexed by ). holds a series of controls. 10 We find 6-10 quarters after the LPO was implemented, it had increased employment rates among affected lone mothers (those with a youngest child aged 5-15) by around 5.3ppts. 11 That translates to an increase in employment among (25-to 64-year-old) lone mothers as a whole of 3.5ppts, accounting for about a third of the total increase in employment among lone mothers between 2007 and 2019.
This is a relatively large increase in employment compared with what one can normally expect from a welfare reform. That certainly does not mean that the policy is unambiguously advantageous, however. Single mothers who remain out of work after the LPO was implemented clearly lose outas they either look for work (time which they presumably would rather spend doing something else), or they do not look for work and so cease to be eligible for out-of-work benefits. And at least some of those single mothers who went into employment as a result of the reform are likely to be worse off than they would have been, if we look more broadly than just at their total incomethey could, after all, have worked under the old (more generous) welfare regime but did not.
The second policy we examine is the rise in the female state pension age (SPA). Before 2010, women could start to receive their state pension upon turning 60. The Pensions Acts 1995 and 2011 legislated for that age to steadily rise to 66 between April 2010 and September 2020, affecting women born after April 1950 (those born later seeing a larger rise in their SPA). Figure 7 shows employment rates for women aged 59-65. The dashed lines indicate the period over which the female SPA was rising from below to above the corresponding age. In other words, it indicates the period over which state pension entitlement was being removed for women of a given age. The figure generally shows a slow increase in employment rates for all age groups over most of the period examined, but a much sharper increase specifically when state pension entitlement was being removed. Although our focus in this paper is on those aged 16-64, the female SPA is currently in the process of increasing to 66, and the figure indicates a similar increase in employment for women aged 65.

Figure 7: Female employment rates by age
Note: Dashed lines indicate the period when the female SPA was rising from below to above the corresponding age.

IV.B Changing composition of the population
The previous subsection showed that two policies have had a significant effect on the employment rates of particular groups. We now examine changes in the composition of the population, which have a weaker relationship to specific reforms but which may account for some of the change in overall employment. Appendix Tables A2 and A3 give an indication of some of these compositional changes.
For example, between 2007 and 2019, the share of the population aged 45-54a relatively highemployment groupincreased by 1.5ppts, while the share of immigrantsa lower-employment groupincreased by 5.4ppts. These compositional changes can affect the headline employment rate.
We can more systematically estimate the contribution of changes in composition to changes in employment rates by using a Oaxaca-Blinder decomposition (Oaxaca, 1973;Blinder, 1973). The intuition here is that we measure the relationship between various individual characteristics and  Taking these results at face value, the compositional changes in the population with regards to age, race, immigration status, family structure and region served to reduce employment by about 1pptmaking the actual increase in employment all the more surprising. But if we also account for the growth in qualifications, the sum effect of these compositional changes is to increase employment by 2.3pptsenough to account for about half of the total increase in employment.
What should we make of these results? Whether or not to include qualifications depends on exactly why those with better qualifications are more likely to be employed. We can think of the decompositions with and without qualifications as representing two extreme scenarios. On one extreme, the entire reason for the correlation between qualifications and employment is causal: the only reason that better-qualified people are more likely to be employed is their qualifications. On the other extreme, having better qualifications has no causal effect on employment rates, and the observed correlation is exclusively due to better-qualified people being more likely to have other characteristics which are themselves the reason for their higher employment rates; and the composition of the population has not changed with regards to these other characteristics.
If the former view is correct, then the right decomposition is the one that includes qualificationssince the increase in average qualifications across the population represents a compositional 'improvement' in the likelihood of employment. If the latter view is correct, then the right decomposition is the one that excludes qualificationsmore people getting better qualifications has no causal impact on their employment prospects and so we do not want to count that change as a change in the composition of the population.
In reality, the correct view is likely to be somewhere in betweenthe increase in average qualifications probably did have a causal impact on employment rates, but part of the reason that those with better qualifications are more likely to be employed is because such people have other characteristics which improve employment prospects. Because we do not know precisely where on this spectrum the correct view lies, it is difficult to say whether the compositional changes in the population served to reduce employment or increase it. What we can rule out is that these compositional changes explain all of the growth in employment. At least half remains unexplained by these factors, even if the correlation between qualifications and employment is entirely causal.

IV.C Employment among immigrants
As shown in Figure 4, the employment rate of immigrants increased substantially from 2007 to 2019.
To what extent can this be explained by compositional changes in the immigrant population? Table 4 shows the composition and employment rates of immigrants in 2007 and 2019 by their country of birth and the age they left full-time education. There was a clear shift from low-educated to more highly educated immigrants. At the same time, the immigrant population tilted towards those from the (mainly eastern European) 'rest of Europe' group and away from Africa, the Americas and Oceania.
Both of these effects represent a movement towards immigrant groups that are more likely to be employed. The change in composition with regards to country of birth may have been driven in part by the loosening of restrictions on immigration from eastern Europe and tighter controls on immigration from outside the European Economic Area seen over the period. Again we can more formally assess the contribution of the compositional change in immigrants to the change in the immigrant employment rate with a Oaxaca-Blinder decomposition. We examine the effect of changes in the population along the same dimensions as in the decomposition described above, plus country of birth (grouped as in Table 4). This shows that, of the 10ppt increase in immigrant employment, about 2ppts can be explained by compositional changes 14a meaningful contribution, but still only a fifth of the overall rise. It is possible that the UK's relatively low unemployment rate compared with many European nations following the Great Recession increased the frequency of immigration to the UK specifically for work. This could have increased the noncompositional effect.

V. How did the quality of jobs change between the Great Recession and the COVID-19
crisis?
Thus far, we have described the rise in employment between 2007 and 2019 and we have examined its causes. But the relationship between families' living standards and paid work is dependent upon not just the number of workers, but also the nature of the jobs they do. In this section, we investigate how job quality changed over the period. By 'job quality' we mean the value that workers might get out of the job they have, rather than a broader notion such as how well the job contributes to societal welfare.
The most straightforward indicator of job quality is hourly pay, since it measures the financial reward a worker receives for an hour of their time. Figure 9 shows changes in real hourly earnings among employees (aged 25-64) across the wage distribution between 2007 and 2019, split by sex. Women saw faster growth than men, with female hourly pay rising almost across the board and all but the bottom 10% of male wages actually falling. Wage growth has also been a little stronger at the middle than the top, and much stronger at the bottom than the middlea consequence partly of rises in the minimum wage (as discussed in Cribb, Norris Keiller and Waters (2018)). (Note that if we examine household total earnings, rather than individual hourly pay, the opposite trend emerges, with growth weaker further down the distribution than further up.) However, these differences by sex and across the distribution should not distract from the key result from Figure 9: by historical standards, the decade or so following the recession was a very bad one indeed for pay growth. At the median, overall hourly pay fell by about 2%; even at the 10 th percentile, it only grew by 9%. In comparison, in the decade before the recession, median hourly earnings grew by 24%. 15 In other words, whereas we would usually expect job quality as measured by wages to improve over time, between 2007 and 2019 there was, for most jobs, no improvement at alland even among lower-paid jobs the improvement was fairly meagre. Another way of measuring job quality is analysing workers' attitudes to and perceptions of their job. relatively small, and so we indicate statistically significant differences with asterisks.
The figure shows several dimensions along which job qualities have improved, and several along which they have worsened. Workers were more likely to consider their job interesting and valuable in 2015 than they were in 2005. There is some evidence that their relationship with the firm they work for improved. And the fraction reporting that 'opportunities for advancement are high' in their job increased from 25% to 34%. However, workers were more likely to report difficulties at work, including stress and (perhaps surprisingly) hard physical work. There is also some evidence that in 2015 they considered their job less secure than they did in 2005. Though it might be thought that greater flexibility can be the flipside of less security (for example, because some gig economy workers have reduced employment rights but more control over their hours of work from one week to the next), in fact perceptions of flexibility also appear to have, if anything, worsened on average over the period. Investigating differences in these trends across different subgroups is hampered by BSAS's small sample sizes. Insofar as we can detect any differences across subgroups, it appears that the increase in 'difficulties at work' was driven almost entirely by those in the bottom half of the earnings distribution, while the improvement in 'interest in and value of work' was more concentrated among women.

Figure 11: Attitudes to and perceptions of job, 2005 and 2015, workers aged 25-64
Note: * indicates a statistically significant difference at the 10% level; ** indicates a statistically significant difference at the 5% level. Taken together, we see a mixed picture for changes in job quality between the Great Recession and the current crisis. By the end of the period, workers appeared to be more interested in their work, have a better view of the firm they work for, and perceive better opportunities for advancement. There was relatively little change in dissatisfaction with hours worked, and in the frequency of people on a temporary contract or looking for a different job. But on some dimensions, job quality declined: workers reported greater difficulties at work such as stress, less flexibility and less security.
Moreover, hourly paythe aspect of job quality that we would usually expect to steadily improve over timefell across three-quarters of the distribution, a very poor showing by historical standards.

VI. Conclusion
The two key characteristics of the labour market over the period bookended by the Great Recession and the onset of the COVID-19 crisis were the strong employment growth and the weak pay growth.
The former was widely shared, and was strongest for those demographic groups that started out with low employment ratesincluding immigrants, lone parents and older workers. Specific policy reforms account for part of the rise in the latter two groups, and the steadily increasing educational qualifications among the population may also have contributed to the overall increase in employment.
Some of the rise in employment was in sectors such as hospitality which are vulnerable to the current crisis, particularly among younger workersalthough overall the workforce shifted slightly away from sectors that have been shut down during the pandemicand much of the increase was driven by the self-employed, who have in some respects been a relatively vulnerable group during this crisis too, whom the government has struggled to comprehensively insure. There was also a shift towards the occupations now classed as key workers. Fortunately from the point of the view of the current crisis, all of the employment growth since the Great Recession had been in jobs that can be done from home.
The weak pay growth probably stands out as the worst attribute of the labour market over the period: at the median there was no growth at all, and though wages grew faster at the bottom, the pace was fairly meagre by historical standards. Other characteristics of job quality give a more mixed picture.
While employees seem to have greater attachment to their work and their firm, they also perceive less security and flexibility in their job.
In terms of living standards and poverty, there were certainly plenty of challenges before the COVID-19 crisisthe weakness in earnings growth and benefit cuts had been putting a lot of pressure on incomes at the bottom. But there is no question that large falls in unemployment, and particularly in household worklessness, had been a significant factor in keeping poverty lower than it would otherwise have been. The danger in the current crisis is that so much of that will be undone, and there is nothing in its place to prevent more vulnerable households from falling into hardship. One of the huge challenges going forward will be trying to ensure that, by the time the temporary increases in support are unwound, employment is bouncing back quickly.