Depressive Symptoms Among US Adults During the Great Recession and Economic Recovery

Abstract

Background and Aims:

We study the trajectory of depressive symptoms among US adults before, during, and after the 2008/2009 Great Recession.

Methods:

We use repeated cross-sectional surveys of the National Health and Nutrition Examination Survey (NHANES) between 2005 and 2018. Mental health is assessed with the Patient Health Questionnaire-9 (PHQ-9), with the following categorization for depressive symptoms: none or mild (score 0–9), moderate or severe (score 10–27). A parallel time series was calculated from the Behavioral Risk Factor Surveillance System (BRFSS) on self-reported number of days with poor mental health.

Results:

NHANES data show a statistically significant increase in depressive symptoms from 2005/2006 to 2007/2008 (the beginning of the Great Recession), but there were no significant or consistent changes after 2007/2008. In particular, the deterioration in the adjusted predicted PHQ-9 scores occurred prior to the large increase in unemployment rate (2009/2010). As the macroeconomic situations improved and unemployment rates recovered, mental health did not return to the previous level. In the latest wave of NHANES (2017/2018), unemployment rates were at the lowest level over the analysis period; however, the adjusted predicted PHQ-9 scores were higher than that at the beginning of the Great Recession. Trends of PHQ-9 scores were similar across income groups – all groups had an increase in depressive symptoms after 2005/2006 and PHQ-9 scores were still high in 2017/2018 after controlling for sociodemographic status. Group with the lowest income had higher levels of depressive symptoms at every time point. BRFSS data shows no consistent changes in the number of days with poor mental health that parallel economic conditions.

Discussion:

Depressive symptoms at the population level did not match the economic cycle before, during and after the Great Recession. Future research is needed to better understand the lack of correlation between population mental health and macroeconomic conditions.

Introduction

Do economic downturns worsen the mental health status of a population? Do recoveries lead to similar improvements in population mental health? This paper aims to answer these perennial questions for the US using a new data point in the National Health and Nutrition Examination Survey from 2005 to 2018, with additional data from the Behavioral Risk Factor Surveillance System from 2001 to 2019.

The Great Recession, which had a major economic impact worldwide, is now well in the past and consistent data that cover the period before, during, and after it are now available. In the US, the recession technically lasted from December 2007-June 2009 as measured by the nominal Gross Domestic Product (GDP) trough, although outcomes more relevant for well-being react with a lag.¹ Unemployment rates peaked at 10% in October 2009 – after the technical end of the recession – which was a level that had not been reached in the previous 25 years.² Household net worth regained its pre-recession level by 2012 and unemployment rates dropped to 2005 levels by 2015.² Economic conditions kept improving and by 2018, the seasonally adjusted unemployment rate had dropped under 4%, far below the historical average since 1948 and reached the lowest levels since 1969.²

The European Psychiatric Association published a comprehensive literature review in 2016 on mental health and economic crises and initially identified about 3000 records.³ The final review includes 350 publications up to 2015, excluding duplicates, opinion pieces, policy statements. A different systematic review, published at the same time, summarized 101 publications that satisfied its inclusion criteria.⁴ The review by the European Psychiatric Association noted that publications on health and economic recessions follow the business cycle – mostly during crises and their immediate aftermath in countries or regions that were most affected.³ However, few studies explored what happens during economic recovery and boom times. If economic crises were to have large deleterious consequences for mental health and mental health care, do economic boom times compensate for them? Or, as the European Psychiatric Association’s review warned, is it the case that “many of the adverse effects on mental health can be pervasive or manifest fully long after the onset of the crisis”?³

Olfson et al. recently reported on mental health need and service use in the US between 2004/2005 and 2014/2015 using the Medical Expenditure Panel Survey.⁵ The statistical tests only compared 2004/2005 and 2014/2015, which spans both the recession and recovery, and found that serious mental health distress declined overall and in all subgroups by gender, age, and race/ethnicity. Over the same period, use of mental health services increased.⁵ While Olfson et al. did not test whether need for mental health services increased during the economic downturn and then recovered, numbers shown for 2009/2010 did not indicate an increase in need compared to 2004/2005.⁵ Another study using the longitudinal Midlife in the United States series also found improving mental health trend between 2003 and 2013.⁶ In other words, data from the Medical Expenditure Panel Survey and the longitudinal Midlife in the United States study do not provide evidence that the biggest economic crisis since the 1930s had adverse effects on mental health that were either pervasive or appeared long after the onset of the crisis in the US.

On the other hand, data of suicide rates showed a different picture. Age-adjusted suicide rates have increased over the same time period, from 10.9 per 100,000 standard population in 2005 to 12.1 in 2010 to 13.3 in 2015.⁷ However, it does not provide evidence for a relationship between the business cycle and mental health. The time series for suicide seems unrelated to economic conditions, neither displaying a more rapid increase during the economic downturn nor a decline during recovery. These results from credible data sources that include a period of economic recovery were not consistent with each other nor with the conventional narrative about economic conditions and mental health.

Mental health is part of the broader question of how population health changes with macroeconomic conditions. Historically, economic downturns tended to be accompanied by lower overall mortality rates and improvements in some measures of physical health.^8–12 The relationship between economic downturns and reduced mortality appears to have weakened in more recent periods,⁸ possibly because deaths caused by air pollution and occupational or traffic accidents have become less important over time. However, evidence on causal pathways is not settled. One proposed pathway is that health behaviors improve during recessions, such as declines in smoking and excess weight and increases in leisure-time physical activity, improve cardiovascular health and reduce mortality.¹³ A contrasting study argues that 2008/2009 Great Recession negatively impacted blood pressure and blood glucose levels.¹⁴

For mental health, one would expect a less ambiguous relationship, but the evidence is weaker than commonly believed.^3,4 There is universal agreement that individuals experiencing financial difficulties or losing employment have worse mental health,^15,16 but not whether there are similar relationships between macroeconomic changes and population health. Moreover, if publications have focused on downturns,³ do economic recoveries affect mental health as well? In particular, what happened after the Great Recession ended?

We analyze data from the National Health and Nutrition Examination Survey (NHANES) from 2005 to 2018, measuring psychological distress and depressive symptoms with the Patient Health Questionnaire (PHQ-9), a self-administered version of the earlier clinician-administered Prime-MD.¹⁷ A second data set, the Behavioral Risk Factor Surveillance System (BRFSS), includes the number of days in poor mental health over the same time period. These data span the largest macroeconomic swings in the last 80 years with both a downturn and a recovery and offers more complete data of the recent relationship between macroeconomic conditions and population mental health.

Methods

Study Design and Population

The NHANES is a nationally representative cross-sectional survey conducted by the National Center of Health Statistics since 1999.¹⁸ The PHQ-9 is a 9-item instrument to screen for depressive symptoms over the past 2 weeks and incorporates DSM-IV depression diagnostic criteria.^17,19–21 The PHQ-9 has been included in NHANES since 2005, but not in earlier years. Questions were asked by trained interviewers in person using the Computer-Assisted Personal Interview system. This analysis is limited to adults age 18 and older who answer all 9 questions of the PHQ-9. Respondents who did not provide answers to educational level, poverty ratio, or any question in PHQ-9 are dropped from the study population, leaving our final study population to 31,389 (Table 1).

Table 1.

Exclusion Criteria.

	Number of respondents	Number of respondents dropped
Number of respondents aged 18+ during study period	42,143
After dropping respondents missing educational level	39,688	2,455
After dropping respondents missing poverty ratio	35,901	3,787
After dropping respondents missing one or more PHQ-9 responses	31,389	4,512

We explore three outcome variables in this study. The first outcome we look at is the total score of depressive symptoms calculated from the 9 item Patient Health Questionnaire. Each symptom question is scored from 0 to 3, corresponding to the response categories “not at all,” “several days,” “more than half the days,” and “nearly every day.” A total score is calculated for persons who have complete responses to the symptom questions and ranges from 0 to 27. In clinical setting, the diagnosis of depressive disorders also requires impairment of social, occupational, or other important areas of functioning, as indicated in an additional item (question 10), but this question is not used in calculating PHQ-9 scores. Scores are commonly classified into the following categories: 1–4 minimal depression; 5–9 mild depression; 10–14 moderate depression; 15–19 moderately severe depression; 20–27 severe depression. A PHQ-9 score greater or equal to10 had a sensitivity of 88% and a specificity of 88% for major depression.²⁰ Therefore, we use a threshold of 10 or higher to separate respondents with mild or no depressive symptoms from those with at least moderate levels of depressive symptoms as the second outcome variable.

In addition to NHANES, we also utilize the BRFSS data from 2001 to 2019. The BRFSS is the largest health-related telephone survey which interviews 400,000 adults each year.²² Since 2011, data collection and weighting methodology of BRFSS has been changed to allow the inclusion of data by cellphones. It only includes one question pertaining mental health across waves for all individuals, which asks respondents the number of days that their mental health was in a bad state. This includes stress, depression, and problems with emotions.²³ Our third outcome variable is thus the number of days interviewees felt that their mental health was not good.

Data Analytic Procedures

We apply an ordinary least squares (OLS) regression with total PHQ-9 scores as the dependent variable and the cycle indicator as the main independent variable of interest. Logistic regression is utilized to predict the probability of having PHQ-9 scores of 10 or higher for respondents in each NHANES cycle. For both regression models, we control for sociodemographic characteristics as the regressors to assure that results are not driven by compositional changes between waves, including sex, age groups, race/ethnicity, and educational levels. In addition, we stratify respondents into three groups based on the ratio of family income to poverty to see if the predicted PHQ-9 score trend varies for individuals with different incomes. Adjusted predicted PHQ-9 scores and probabilities of having moderate or severe depression are calculated for each NHANES wave using Stata’s margins command. All analyses incorporate the full sample 2-Year Examination Weights for national representativeness.

For the third outcome variable using the BRFSS, we report descriptive statistics that are weighted for national representativeness. Wald F statistics are used to examine whether the average number of days with bad mental health are statistically different between cycles. Data are analyzed using Stata 16 and statistical significance is set at a 2-tailed P < .05 for all analyses.

Results

NHANES Participant Characteristics

Between 2005 and 2018, 31,389 of the NHANES respondents aged 18 years or older completed the PHQ-9 questionnaire (Table 1). The sex composition is roughly constant across waves, but other demographic variables change noticeably (Table 2). Later waves include more respondents older than 60 years old, more people with some college education and fewer Non-Hispanic Whites (Table 2). The percentage of respondents with family income below 130 percent of the federal poverty level peaked in the 2013/2014 cycle and declined afterwards (Table 2).

Table 2.

Characteristics of US Adults Aged 18 or Older, 2005–2018 NHANES.

	2005–2006	2007–2008	2009–2010	2011–2012	2013–2014	2015–2016	2017–2018
Number of unique respondents	4139	4740	4840	4292	4707	4431	4240
Characteristics, n (weighted %)
Male	1,995 (48.42)	2,345 (48.30)	2,405 (49.26)	2,154 (48.97)	2,274 (48.85)	2,168 (48.45)	2,075 (48.56)
Female	2,144 (51.58)	2,395 (51.70)	2,435 (50.74)	2,138 (51.03)	2,433 (51.15)	2,263 (51.55)	2,165 (51.44)
Age group in years
18–39	1,597 (38.08)	1,541 (37.78)	1,594 (36.05)	1,513 (35.88)	1,595 (36.01)	1,502 (35.68)	1,287 (35.98)
40–59	1,286 (39.81)	1,521 (39.11)	1,636 (38.98)	1,381 (37.54)	1,585 (36.53)	1,472 (36.87)	1,355 (34.96)
60+	1,256 (22.11)	1,678 (23.12)	1,610 (24.97)	1,398 (26.58)	1,527 (27.46)	1,457 (27.45)	1,598 (29.06)
Race/ethnicity
Non-Hispanic White	2,127 (73.43)	2,305 (70.98)	2,482 (71.04)	1,695 (69.06)	2,123 (67.80)	1,566 (66.60)	1,585 (64.85)
Hispanic	933 (10.77)	1,296 (12.77)	1,285 (12.51)	819 (13.04)	998 (13.78)	1,326 (14.27)	903 (14.39)
Non-Hispanic Black	923 (10.86)	970 (10.93)	839 (10.54)	1,106 (10.68)	946 (10.87)	921 (10.70)	964 (10.70)
Other	156 (4.95)	169 (5.32)	234 (5.91)	672 (7.21)	640 (7.55)	618 (8.43)	788 (10.06)
Educational Level
Less than high school graduate	1,098 (16.73)	1,415 (19.85)	1,276 (17.59)	925 (14.94)	932 (14.06)	980 (13.15)	752 (10.08)
High school graduate/GED	986 (24.82)	1,164 (24.84)	1,125 (22.94)	903 (20.17)	1,054 (21.63)	986 (21.01)	1,036 (27.61)
Some college or more	2,055 (58.45)	2,161 (55.31)	2,439 (59.47)	2,464 (64.90)	2,721 (64.31)	2,465 (65.84)	2,452 (62.30)
Ratio of family income to poverty
<1.30	1,069 (16.87)	1,436 (20.30)	1,570 (20.89)	1,496 (23.71)	1,579 (23.87)	1,428 (20.73)	1,179 (19.62)
1.30–3.49	1,619 (37.38)	1,840 (34.72)	1,826 (36.07)	1,482 (35.03)	1,640 (34.61)	1,780 (37.11)	1,756 (36.10)
≥3.5	1,451 (45.75)	1,464 (44.99)	1,444 (43.05)	1,314 (41.26)	1,488 (41.51)	1,223 (42.16)	1,305 (44.28)
Outcome variables, n (weighted %)
PHQ-9 score=0	1,432 (35.18)	1,539 (30.61)	1,594 (33.06)	1,520 (35.21)	1,508 (32.29)	1,341 (29.38)	1,432 (31.50)
PHQ-9 score=l-4	1,848 (45.05)	2,004 (46.00)	2,015 (43.40)	1,731 (42.71)	1,998 (44.03)	1,954 (46.86)	1,729 (43.90)
PHQ-9 score=5–9	593 (14.11)	742 (15.38)	770 (15.70)	648 (14.07)	749 (14.91)	769 (16.31)	708 (16.30)
PHQ-9 score=10–14	179 (3.78)	289 (5.24)	286 (4.85)	229 (4.71)	279 (5.56)	228 (4.68)	240 (5.55)
PHQ-9 score=15–19	66 (1.45)	123 (2.07)	131 (2.30)	112 (2.25)	125 (2.25)	88 (1.84)	97 (2.08)
PHQ-9 score=20–27	21 (0.43)	43 (0.70)	44 (0.69)	52 (1.05)	48 (0.96)	51 (0.94)	34 (0.67)

The percentage of respondents with PHQ-9 score of zero is the highest in cycle 2005/2006 (35.18%) and cycle 2011/2012 (35.21%) (Table 2). Nearly half of the respondents in each cycle possess minimal depression, or PHQ-9 scores ranging 1 to 5, and in all waves, less than 9% of all respondents had a PHQ-9 score of 10 or more (Table 2).

Depressive Symptoms over the Business Cycle

Figure 1 shows the adjusted predictions of PHQ-9 scores using weighted OLS regression alongside unemployment rate and recession period. Depressive symptoms increased from 2005/2006 to the onset of the Great Recession in 2007/2008 (2005/2006: PHQ-9=2.54, 2007/2008: PHQ-9=3.03, p value<0.001). There was no significant additional change in depressive symptoms as unemployment rates kept rising and peaking in 2009/2010, nor was there an improvement as unemployment rates fell. In fact, even after unemployment rates reached their lowest levels over the study period in 2017/2018, the adjusted PHQ-9 scores were significantly higher than before the Great Recession in 2005/2006. Adjusted predicted probabilities of respondents having moderate distress, defined as having PHQ-9 scores greater or equal to 10, were in very similar trend as the adjusted predictions of PHQ-9 scores from OLS results (Figure 2).

Figure 1.

The Adjusted Predicted PHQ-9 Score Trend and 95% Confidence Interval^a with Unemployment Rate and Recession^b, 2005–2018 NHANES.

^aPredicted values and their 95% confidence intervals are from weighted OLS regression using margins command in Stata, controlling for sex, age bins (24–39,40–59,60 or above), race (Hispanic, Non-Hispanic White, Non-Hispanic Black, and other race), and educational level (less than high school, high school graduate/GED, some college/associates (AA) degree or higher).

^bData source: Bureau of Labor Statistics via the Federal Reserve Bank of St. Louis.

Figure 2.

The Adjusted Probability of Having PHQ-9 Score ≥10 and 95% Confidence Interval^a with Unemployment Rate and Recession^b, 2005–2018 NHANES.

^aPredicted values and their 95% confidence intervals are from weighted logistic regression using margins command in Stata, controlling for sex, age bins (24–39,40–59,60 or above), race (Hispanic, Non-Hispanic White, Non-Hispanic Black, and other race), and educational level (less than high school, high school graduate/GED, some college/associates (AA) degree or higher).

^bData source: Bureau of Labor Statistics via the Federal Reserve Bank of St. Louis.

To examine differential impacts by economic status, we stratified the study population by the ratio of family income to poverty in each wave. Lower income groups always experience higher levels of psychological distress (higher PHQ-9 scores), but the time trends are largely similar. All groups see an increase in depressive symptoms and by a comparable amount from 2005/2006 to 2007/2008 and remain elevated in 2017/2018. There was a significant dip in the 2011/2012 cycle for the group with the highest income, or respondents with a poverty ratio of greater or equal to 3.5, but this deviation was only observed for a single wave (Figure 3).

Figure 3.

The Adjusted Predicted PHQ-9 Score Trend and 95% Confidence Interval^a, Stratified by Ratio of Family Income to Poverty, 2005–2018 NHANES.

^aPredicted values and their 95% confidence intervals are from weighted OLS regression using margins command in Stata, controlling for sex, age bins (24–39,40–59,60 or above), race (Hispanic, Non-Hispanic White, Non-Hispanic Black, and other race), and educational level (less than high school, high school graduate/GED, some college/associates (AA) degree or higher).

Regression coefficients on individual characteristics were as expected (Table 3). Women had significantly higher scores than men, and respondents aged 60 or above had lower psychological distress than those younger than 40. Groups with higher educational levels had lower scores than those with only a high school degree. Hispanics showed fewer depressive symptoms than non-Hispanic White respondents (p-value=0.007), although the differences between non-Hispanic Black and other races compared with non-Hispanic Whites were not statistically significant (p-value=0.152 and 0.574 respectively).

Table 3.

Result of Multivariable-adjusted OLS Regression, Using PHQ-9 Score as the Dependent Variable, 2005–2018 NHANES.

	Coefficient	p-value	95% Confidence	Interval
Year (reference:2005/2006)
2007/2008	0.491	<0.001	0.305	0.676
2009/2010	0.509	<0.001	0.325	0.693
2011/2012	0.449	<0.001	0.241	0.657
2013/2014	0.657	<0.001	0.461	0.853
2015/2016	0.624	<0.001	0.421	0.827
2017/2018	0.659	<0.001	0.448	0.870
Gender (reference: male)
Female	1.020	<0.001	0.908	1.132
Age (reference: 18–39)
40–59	0.168	0.014	0.034	0.303
60 or above	−0.436	<0.001	−0.570	−0.301
Race/ethnicity (reference: non-Hispanic White)
Hispanic	−0.201	0.007	−0.348	−0.055
Non-Hispanic Black	0.101	0.152	−0.037	0.238
Others	−0.059	0.574	−0.264	0.146
Educational level (reference: less than high school graduate)
High school graduate/GED	−0.669	<0.001	−0.864	−0.475
Some college or more	−1.372	<0.001	−1.544	−1.201
Constant	3.089	<0.001	2.872	3.307

The trend of mental health in BRFSS shows a similar picture. Although there was an increase in the average number of days with poor mental health from 2007 to 2008, the magnitude was small and not statistically significant (2007:3.38 days, 2008:3.44 days, p=0.096). When unemployment rates fell back to pre-recession level in 2015 and kept declining until the end of 2019, the average days respondents reported with poor mental health in the past 30 days increased from 3.71 days in 2015 to 4.33 days in 2019 (p<0.001) (Figure 4).

Figure 4.

Number of Days With Poor Mental Health in the Past 30 days and 95% Confidence Interval with Unemployment Rate and Recession^a, 2001–2019 BRFSS.

^aData source: Bureau of Labor Statistics via the Federal Reserve Bank of St. Louis.

Discussion

This paper studied the trajectory of depressive symptoms before, during, and after the Great Recession. It provides additional data as other recent publications showed conflicting results: The Medical Expenditure Panel Survey suggests improved population mental health since 2005 with no evidence of a deterioration in 2010,⁵ the longitudinal Midlife in the United States study found improving mental health trend between 2003 and 2013,⁶ while suicide statistics suggest worsening health with no evidence of an improvement since 2010.⁷ Our analysis of the NHANES and BRFSS showed an increase in depressive symptoms around the beginning of the Great Recession, but there was no improvement in population mental health when economic recovery was observed.

There were numerous prior publications examining the relationship between economic downturns and mental health.^3,4 The consistent narrative is that economic crises are negatively associated with population mental health.²⁴ The evidence is far more nuanced and qualified, as noted by the European Psychiatric Association’s guidance paper on the relationship between economic crisis and mental health follow the business cycle.³ To date, we know little about whether mental health improves during economic boom times or if the negative impact on population health from recessions is long-term. The Great Recession had a major economic impact worldwide, but eventually there was a full recovery and seasonally adjusted unemployment rate had dropped under 4% in 2018, the lowest levels since 1969.² Recent studies conducted in England showed different mental health trajectories among subgroups during economic recovery,²⁵ and that recession-related fluctuations in suicide rates were only found in men but not women.²⁶ Our analysis of stratifying NHANES participants by ratio of family income to poverty also demonstrated slightly different trends of the adjusted predicted PHQ-9 scores and no significant recession-related trends were found.

Our paper provides evidence that the changes in economic conditions since 2007/2008 are not paralleled by similar changes in depressive symptoms. However, the interpretation of our results is constrained by several limitations. First, we only have one wave of NHANES data (2005/2006) before the onset of recession, and ideally more waves would help illustrate the pre-recession trend of mental health. We cannot rule out the possibility that the 0.49 increase in adjusted predicted PHQ-9 score from 2005/2006 to 2007/2008 is due to changes in the NHANES sample that we did not control for. Nevertheless, the main modification between the two periods was switching oversampling of Mexican-Americans to entire the Hispanic population; otherwise, the survey and sample design of NHANES were stable.²⁷ Second, we refrain from comparing BRFSS data before and after 2011, the year data collected by cellphones were included. According to Centers for Disease Control and Prevention (CDC), prevalence estimates for some indicators of poor health will increase in most states due to the adoption of iterative proportional fitting and/or the addition of cellular phone households. Therefore, observed changes in prevalence between BRFSS 2010 and 2011 are more likely to reflect the new data collection and weighting methodology rather than trued trends.²⁸ Finally, although we do not attempt to establish causality in this study, the lack of correlation of population mental health and economic cycles sheds light to the need for future research into possible mechanisms.

Currently, there is another economic cycle underway. The COVID-19 epidemic led to temporary spikes in unemployment, even exceeding the 2010 peak briefly. The political response, however, was different in that an initial stimulus bill stabilized average household incomes, additional unemployment payments provided a buffer for more vulnerable groups, and unemployment durations were on average much shorter. Nevertheless, the epidemic creates stress and anxiety unrelated to economic factors and it will be important for future studies to compare those with the primarily economic effects of the Great Recession.