Design

Research in this field, according to Catalano and colleagues [10], can be classified into two groups characterised by design: i) risk factor studies; and ii) net effect studies. Risk factor studies are those which attempt to identify the effects of unfavourable economic circumstances among individuals, such as financial insecurity or involuntary job loss, on their health and behaviours. In contrast, net effect studies provide an insight into the sum of the economic effects on population health by using groups or geographical areas as the analytical units, and estimating temporal change in association between economic indicators (such as the rate of unemployment) and prevalence or incidence rates.

In this paper, we used individual-level data on health status, demographic and socioeconomic circumstances extracted from the UK Quarterly Labour Force Survey (QLFS) [25] to conduct research most akin to the ‘net effects’ study design. The QLFS is among the largest regular social surveys in the UK, collecting cross-sectional information every three months (‘quarters’) from over 100,000 people living in private households across all areas of England, Scotland, Wales and Northern Ireland. To capture circumstances before, during and since the economic recession which began in 2008, we pooled 60 months (20 quarters) of the QLFS from January 2006 to December 2010 to create a dataset of 2.38 million survey responses (1.15 million men, 1.23 million women). Our focus was on the health of the working-age population, so we kept all women aged 16 y–59 y and men aged 16 y–64 y in our data (reflecting gender differences in the national retirement age). This left a sample of 1.36 million survey responses, approximately 57% of the original dataset surveyed across 20 quarters from 2006 to 2010. This very large sample size and high frequency of update were key reasons for using the QLFS in our study, in comparison to alternative sources of data which are less frequent (e.g. the decennial UK Census) or have far smaller sample sizes (e.g. British Household Panel Survey). A narrow risk factor study (e.g. to test the effect of job loss on health) was clearly possible with the QLFS, however, our main objective was to use this rich source of individual-level data to estimate the net effects of the economic recession on population health over time, across regions, and between selected groups.

Measures

Health status in the QLFS was self-rated. The primary outcome variable was health status measured using responses to the question “Do you have any health problems or disabilities that you expect will last for more than a year?” 95% of all survey responses of working age in the QLFS indicated a response (yes or no). Non-responders were not asked this question on health status as they were too ill or distressed to continue with the survey; we omitted these survey responses from the analysis.

In secondary analyses, we investigated specific types of health problems reported in the QLFS that could have been influenced by the 2008 economic recession. In the interests of brevity, we assigned each group a name. The exact wording from the QLFS is given in parentheses.

1. Depression (“Depression, bad nerves or anxiety”);
2. Mental illness (“Mental illness, or suffer from phobia, panics or other nervous disorders”);
3. Cardiovascular (“Heart, blood pressure or blood circulation problems”);
4. Respiratory (“Chest or breathing problems, asthma, bronchitis”).

Individual economic status (employed, unemployed, and economically inactive – including early retirees, students, the long-term sick, and homemakers) was measured according to the International Labour Organization (ILO) definition [26]. Among those who were employed, their occupational class was identified using the UK National Statistics Socio-Economic Classification (NS-SEC) [27]. The NS-SEC groups occupations into classes to reflect working relations between employers and employees, the salaried and the temporary labour force, and the manual and non-manual workers. All occupations in the QLFS were classified into three recognized NS-SEC classes: ‘routine/manual’; ‘intermediate’; or ‘professional/managerial’.

We also utilised a number of conventional demographic measures which were surveyed in each quarter, including age, gender, country of birth, ethnicity (White, Mixed, Indian, Pakistani, Bangladeshi, Chinese, Other Asian, Black Caribbean or Black African, Other), educational qualifications (none, GCSE, A-Level, Degree, Other), couple status (married or living with partner, single never married, separated/divorced/widowed), household tenure (owned outright/with mortgage, rented, rent free, part rent/part mortgage), number of dependents (0 to 4+), and geographical region (n = 20).

Statistical analysis

To test whether overall levels of health worsened during the 2008 economic recession, we first assessed the prevalence of each health status across all other characteristics of the study sample and each of the 20 quarters using descriptive statistics. Association between health and each explanatory variable was investigated using logit regression. Coefficients were exponentiated to odds ratios (OR), indicating the likelihood of a person reporting a health problem compared to the likelihood of not reporting a health problem.

For assessing the extent to which health varied across time, we used an explanatory variable containing 20 categories. Each category represented a single quarter (i.e. three months of data). We fitted logit regression models of each outcome variable, adjusted for this categorical variable denoting time, with the reference category set as quarter 1 (January to March 2006). An advantage of adopting this categorical approach was that we did not impose any presupposition of the distribution of health status across time. This approach was more favourable in comparison to using a combination of linear and polynomial functions of time (e.g. square and cubic functions), which were found to have a smoothing effect, resulting in underestimation of rapid changes in health occurring from one quarter to the next.

We constructed multivariate models in several steps. First, our models were adjusted by age group and gender. An interaction term between age and gender was fitted to account for differences in health between males and females at different ages. Second, we added dummy variables denoting economic status, occupational class, and geographic region into the models separately, and then simultaneously. Third, all other explanatory variables were added sequentially to the models. For each model, 95% confidence intervals (95% CI) were used to assess whether change in the unemployment rate and prevalence of health problems between each quarter was statistically significant (p<0.05).

With a view towards addressing whether changes in each of the outcome variables coincided with the 2008 economic recession, we fitted an age, gender and time adjusted logistic regression model of unemployment (vs. employment) separately. Only economically active people were included in these models, as these people were either in work or actively looking for jobs, in line with the ILO definition [26]. Indirect comparisons between trajectories in unemployment and health status were made in this way, rather than adjusting for a regional measure of unemployment, as our use of a fixed effect for region precluded the analysis of any explanatory variables at this level. However, individual economic status was included in the health models. Indirect comparisons also helped us to avoid imposing any assumptions regarding the length of time required for the economic recession to have observable influences on population health.

To investigate influence on health inequalities, we extended our multivariate logit regression of health status with interactions between time, economic status, occupational class, and geographic region. The model interacting time with occupational class omitted all people who were not employed. All of these models identified the extent of change in health status across time between people in different geographical and socioeconomic circumstances during the economic recession. All analyses were conducted in Stata v.12 (StataCorp, TX, USA).

Key findings

Using a large UK-based survey repeated every three months for 20 quarters from January 2006, our study demonstrated a significant increase in the prevalence of poor health status shortly after the rise in unemployment at the start of 2008. This was reflected in similarly timed increases in cardiovascular and respiratory health problems, though not for depression and mental illness. Overall, this suggests that economic recession poses a substantively negative effect on population health.

Counterintuitively, the prevalence of poor health status among the unemployed decreased and remained lower than pre-2008 levels between Jul–Sep '08 and Oct–Dec '09, shortly after the rise in unemployment. After this period, however, the prevalence of poor health status increased significantly across the board, regardless of geographical and socioeconomic circumstances. Socioeconomic inequalities in poor health status observed prior to 2008 were not exacerbated, as slightly higher increases were experienced by people who remained employed (vs unemployed) and among those in professional and managerial occupations (vs routine and manual). Moreover, the greatest increase in poor health prevalence was not concentrated in regions with the highest pre-2008 prevalence.

Interpretation

Previous research has suggested that adverse effects of recessions on health are most likely to occur when economic declines are rapid [4], as was the case in 2008 [22]. The substantial rise in poor health during the economic recession is an important finding. While previous research demonstrates a two-fold increase in the risk of developing long-term illness among people who become unemployed [2], controlling for economic status did not explain the increasing risk of poor health in our study. Nor did occupational class, or any of the other numerous explanatory variables we fitted in our models. This rise only appeared to occur following a period in which the prevalence of poor health among the unemployed decreased. Although this may appear counterintuitive, it is likely due to a mixing of those who were recently laid off with people who had suffered health problems associated with long-term unemployment.

The subsequent increase in poor health experienced for all groups, but particularly those who remained employed (and regardless of their occupational class), is an important finding which falls in line with previous literature. Although many people did not experience redundancy, Catalano and colleagues [10] suggest that stress associated with job-insecurity is a key mechanism for increasing poor health status during economic recession. Fear of unemployment also helps to explain why there was a time-lag between the start of the recession and the rise in poor health, as changes in feelings of job security and declining confidence in the economy are unlikely to be instantaneous across the board. Optimism among some people who felt secure at the start of the recession is likely to have diminished over time, as unemployment grew, and although it appeared to plateau, it remained high. A persistently bleak economic outlook and with hopes of a quick rebound to pre-recession prosperity looking increasingly unlikely, it would not be surprising if financial difficulties had accumulated more slowly among some people who were not instantly effected (e.g. by unemployment), to a point where it did have a significant impact on their health; even among those in more affluent circumstances. To the best of our knowledge, the only other published empirical findings (so far) of this recession on health have been reported by studies of suicide within countries of the European Union [17], [18]. They reported increasing rates of suicide immediately preceding and then during the economic recession which began in 2008. Together, these findings and ours suggest that the anticipation of potential job loss is likely to have been an important mechanism linking the 2008 economic recession with health [4], [23], [24].

Health inequalities, by economic status, occupational class or geographical region, were not exacerbated by the 2008 economic recession during the time period our study was able to observe. Despite layoffs being concentrated in sectors such as manufacturing [22], people were affected regardless of whether they were in professional/managerial or routine/manual occupations. Rising prevalence of poor health across all groups may suggest a common pathway affecting employed and unemployed people. Studies of behavioural responses to previous recessions, however, suggest the possibility for several mechanisms operating simultaneously on people in different social positions. Indeed, the effect may not always be adverse for health. For example, psychological distress [9] and suicide [17], [18] tend to increase, but mortality from motor vehicle accidents and cardiovascular disease are known to decrease [15], [16]. Alcohol consumption and tobacco smoking reportedly decrease, especially among heavy drinkers and smokers [28], [29], [30]. An increase in physical activity and decrease in bodyweight has been found among those who were previously sedentary and obese in some contexts [28], though not all [31]. Declining working hours during economic recession [22], according to some commentators [28], could free up time for investing in health-related behaviours. Meanwhile, the threat of job loss is known to have had a strong ‘disciplinary effect’ [32] not only among individuals whose sole source of income is their job, but also for those who fear social stigma, potential loss of networks and diminished status [33]. Consequently, some individuals may well have had more time to invest in healthier activities, but it is highly probable that many others will have spent extra hours worrying about how to budget loss of income and lifestyle [10], potentially increasing the risk of substance misuse as a short-term coping remedy [13]. Behavioural responses to the economic recession and their influence on health may therefore play an important role in explaining the time-lag between the start of the recession and the increase in poor health – particularly among those who remained employed. It also suggests that there may be further impacts of the recession on health which are yet to fully emerge. Therefore, an avenue for further research will be the effect of the 2008 economic recession on health-related behaviours, and how responses might have differed across the social gradient.

Strengths and limitations

Our study benefits from the use of a very large sample of individual-level data, collected very frequently and at regular intervals before, during and since the economic recession which began in 2008. This enabled our study to detect important within-year changes in all variables, which would otherwise have been hidden by studies relying upon data collected less frequently (e.g. annually). We were able to compare the extent that the 2008 recession coincided with the reporting of a variety of health problems – some were more subject to change than others. The use of individual-level data on health, economic status, occupational class and all of our confounding variables, avoids the fallacy of inferring the experiences of individuals from combining sources of aggregated data. Furthermore, the large number of explanatory variables helped to avoid methodological problems suffered by some previous studies (e.g. omitted variables bias). We were unfortunately not able to examine the influence of neighbourhood characteristics, as this information could not be linked to the QLFS. Thus, although our study has shown that regional inequalities were not exacerbated, this does not discount the possibility that inequalities within those regions at local levels (e.g. between affluent and deprived neighbourhoods) could have widened.

Self-reported measures of health strongly correlate with mortality [34] and are also associated with important clinical information, such as inflammation in otherwise healthy adults [35], but they will not reflect the full extent of health conditions in the population and future studies might consider the use of objectively measured health status if data is available. The small effect of the recession on mental health in our analyses was unexpected, and may be due to the sensitivity of the question contained within the QLFS wherein participants were required to report health problems or disabilities that they expected to last for more than a year. This stipulation of 12 months or longer may lead to an under-reporting of mental health problems in the QLFS and it may be advisable for future surveys to include short screening instruments (e.g. [36]) to monitor trends in psychological distress at the population-level. The time-lag and rapid within-year changes identified in our study illustrate the importance of having a high frequency of follow-up surveys, but also points towards an important qualification; there may yet be further distal influences of the 2008 recession on health status, with the full effects manifesting over a much longer period of time. Longitudinal studies are warranted in the years ahead.