1. Population

For age group classification, we used preferred terms for age groups from Academic Search Premier & Soc-Index, which provides life stage groups. As per the index, people 18~25 years of age are classified as young adults, 25~40 years of age as adulthood, 40~60 as middle age, and 60+ as older people. We assumed that unemployment would not be relevant to individuals who are retired; we therefore considered people ages 65+ as retired and excluded them from our analysis. Of the total population from eligible states (N=79,629), we extracted data on individuals aged 18~65 years (inclusive) for our sample (N=53, 406) and categorized them into the four age groups as defined above (Figure 1).

Figure 1 Selection of sample in analyses of unemployment and depression in 12 states, Behavioral Risk Factor Surveillance System (BRFSS), 2010. Click for larger image

2. Measures

The dependent variable was depression. We drew upon the data from the optional module, which uses eight variables to measure depression, to create a depression score based on the Patient Health Questionnaire (PHQ-8) [11, 12]. The questions ask about depressive symptoms over the previous 2 weeks. For example, “Over the last 2 weeks, how many days have you had little interest or pleasure in doing things?” For each question, individuals received a score between 0 and 3 depending on the number of days they reported having the specific depressive symptom. Those who responded that they had had a depressive symptom for 0 days or 1 day received a score of 0. Those who responded that they had had a depressive symptom for 2 to 6, 7 to 11, or 12 to 14 days, inclusive, received scores of 1, 2, or 3, respectively. The scores for each question were summed. To construct a binary dependent variable, subjects were classified as depressed if the depression score was greater than or equal to 10[12].

The main independent variable of interest was employment status. Those individuals who responded to the survey that they had been out of work for either less than 1 year or more than 1 year were classified as unemployed. Respondents who indicated they were “employed for wages” or “self-employed” were classified as employed.

Potential confounding variables were included in the multivariable logistic model because of their association with both depression and employment status. These were disability status, smoking status, health insurance status, body mass index (BMI), and sociodemographic status (sex, race/ethnicity, marital status, education, and income level). Most of the potential confounding variables were selected based on previous work by Brown and colleages [13] and McGee and Thompson [2], which examined relationships between unemployment and mental health among adults.

For the disability status variable, we used two variables (QLACTLM and USEEQUIP) to classify three disability statuses setting ‘no disability’ as a reference. Respondents who answered ‘yes’ to the questions regarding whether they have limitations in performing activities and use special equipment were categorized into the ‘Yes, equipment needed’ group. Individuals who reported having limitations in performing activities but did not report using special equipment were categorized into the ‘Yes, no equipment needed’ group and individuals who answered ‘no’ for both questions were coded as the reference group (no disability). Individuals who responded that they have no limitations but use equipment were treated as missing data.

For smoking status, individuals who answered as current non-smokers were set as the reference group.

BMI respondents were classified into 3 BMI categories. Those with BMIs less than 25.0 were categorized as normal weight or underweight, those with BMIs between 25.0 and 29.9 were categorized as overweight, and those with BMIs greater than or equal to 30.0 were categorized as obese. Value 9 (Don't know/Refused/Missing, Notes: _BMI4=9999) were coded as 0.

For health insurance status, respondents answered the question “Do you have any kind of health care coverage, including health insurance, prepaid plans such as HMOs, or government plans such as Medicare?” Those who responded ‘yes’ were classified as individuals with health insurance and those who responded ‘no’ were classified as individuals without health insurance. A ‘don't know/not sure’ response was considered a nonresponse. Subjects who classified as having health insurance were used as the reference.

Selected socioeconomic variables were adjusted in the models including race, education, sex, and marital status. Individuals who responded ‘don't know/not sure/refused’ to the race questions were classified as ‘other’ in our analysis. We identified individuals as having a high school degree or higher as the reference group. For sex, ‘male’ was used as the reference group. For marital status, we assigned only those who responded as ‘married’ to the married group and the rest of the respondents (divorced, widowed, separated, never married, and a member of an unmarried couple) to the unmarried group. Unmarried was used as the reference group. Respondents who refused to disclose their marital status were treated the same as missing. All variable, except income, operationalization was performed following the classification scheme of McGee and Thompson [2] for the purpose of comparing the results for individuals in the same age group.

Annual household income was classified into 5 categories, including less than $15,000, $15,000 to less than $25,000, $25,000 to less than $35,000, $35,000 to less than $50,000, and $50,000 or more. Since this variable is excluded from the final model due to multicollinearity issue (more details explained in the discussion part), the missing value of income variable is not deleted from the whole sample.

3. Analyses

Analysis of sample frequencies and weights was performed to compare the distribution of selected variables in each group. Then bivariate analyses were conducted to identify the statistical significance of each variable related to the outcome variable of interest. Multivariable logistic regression was used to calculate odd ratios. Three logistic regression models were fitted to the data to examine the relationship between depression and unemployment while adjusting for age and other confounding variables. First, we performed a logistic regression including age categories (18~25; 26~41; 41~60; 61~65) to examine whether age is associated with depression risk when controlling for socioeconomic and health insurance variables. In the second model, we added disability status to the analysis to see whether disability status affects the relationship. The last model adjusted for smoking and BMI in addition to the sociodemographic, health insurance, and disability variables. Finally, we compared the odds of depression between unemployed and employed people in different age groups. The analyses were completed using weighted data to account for the complex sample with STATA (version 14.2).

1. Bivariate Analysis

The bivariate relationships between potential confounders and depression as well as unemployment in sample of all ages together were calculated with χ² analyses. In bivariate analyses, depression was significantly associated with other independent variables (Table 2). Females were more likely to report depression (9.2%) compared to males (5.9%) (p<.001). Black or African Americans (9.9%) were more likely to be depressed compared to White (7.4%), Hispanic (9.2%), and other races (7.1%) (p<.001). About 17.8% of adults without a high school degree reported depression compared to 7.3% of those with at least a high school degree (p<.001). Those unmarried (11.8%), without health insurance (15.5%), or reporting a disability and needing equipment (28.4%) were more likely to report depression compared to their respective reference groups (p<.001). Smokers (16.3%) were more likely to be depressed than nonsmokers (5.7%) (p<.001). BMI was also related to depression. 10.6% of respondents classified as obese reported depression, which was significantly more than those who were categorized as overweight (6.6%) and normal weight or underweight (6.5%) (p<.001). Finally, 12.3% of respondents aged 18 to 25 reported depression, 8.8% of 26 to 40 year-old respondents reported depression, 7.7% of 41 to 60 years old respondents reported depression, while 4.4% of respondents aged between 61 to 65 reported depression.

Table 2 Sample Characteristics by Depression and Unemployment Status Click for larger image

Being unemployed was also associated with other independent variables. Blacks or African Americans (18.4 %) were significantly more likely to report being unemployed compared to Hispanics or Latinos (13.6%), other races/ethnicities (12.1%), and Whites (8.5%) in all age groups (p<.001). Those without a high school degree (27.4%), unmarried (16.2%), or without health insurance (30.0%) were more likely to report being unemployed compared to their respective reference groups (p<.001). Additionally, 19.3% of smokers reported being unemployed compared to 8.4% of nonsmokers (p<.001). Those reporting a disability and needing equipment (27.6%) were more likely to be unemployed compared to those with a disability but not needing equipment (18.5%) and those without a disability (8.8%) (p<.001). 12.1 % of obese respondents reported unemployment, which was significantly more than those who were categorized as overweight (9.4 %) and normal weight or underweight (10.3%) (p<.001). Finally, 22.5% of respondents aged 18 to 25 years reported being unemployed, 10.7% of those aged 26 to 40 years reported being unemployed, and 9.7% of those aged 41 to 60 years reported being unemployed, while only 9.9% of respondents aged 61 to 65 years reported being unemployed.

2. Multivariable Analysis

Multivariable logistic regression was used to calculate odd ratios and three logistic regression models were fitted to the data to examine the relationship between depression and unemployment while adjusting for age and other confounding variables (Table 3). There was a clear trend in the odds of depression in all age groups when adjusted for unemployment and other variables. In all three models tested, we found that individuals in the older age group (aged 61~65) have a lower depression risk compared to the younger age group (aged 18~25) after adjusting for other variables including being unemployed. This can be interpreted as older people being less likely to be depressed than younger people when controlling for risk factors like unemployment, together with other demographic and health variables.

Table 3 Odds of Depression among Adults Adjusted for Age Group, Behavioral Risk Factor Surveillance System, 2010 Click for larger image

We then compared the relationship between depression and unemployment in different age groups (Table 4). We found that unemployment is associated with depression in all age groups, but different age groups have different scales of association. In model 1, the odds of depression were 3.14 (95% CI: 2.58~3.82) times higher for unemployed respondents aged 26~40 compared to employed people in the same age group. In models 2 and 3, the odds of depression were 2.5 times higher for unemployed respondents compared to those employed in this age group. For the age group of 41 to 60 years old, the odds of depression were 3.16 times higher (95% CI: 2.74~3.65) for the unemployed compared to those employed in model 1, whereas in model 2 and 3 the odds were 2.34 and 2.16 times higher, respectively. For respondents aged 61~65, the odds of depression were 3.95 (95% CI: 2.48~6.28) times higher for unemployed people in model 1, and 2.81 times higher in models 2 and 3. In other words, respondents in the 61~65 years old age group exhibited the highest increased risk of depression when unemployed.

Table 4 Odds of Depression Comparing Unemployed and Employed People among Adults in Different Age Group, Behavioral Risk Factor Surveillance System, 2010 Click for larger image

To summarize, the initial analysis indicates that the older the age group, the lower the odds of depression after adjusting for employment and other confounders (Table 3). However, when comparing models by age group, we found that individuals in the older age group had higher odds of depression if they were unemployed compared to employed individuals (Table 4). This could suggest that older people who are unemployed have a higher risk of depression than individuals who are unemployed in younger age groups.

1. Limitations

There are other limitations of the BRFSS to be considered in interpreting and generalizing the findings. First, as discussed by McGee and Thompson [2], the current study results might be skewed because of the potential undercoverage of emerging adults in the BRFSS, as many of them use cellular telephones instead of landlines [29]. Second, as this study was designed to extend the earlier findings of McGee and Thompson [2] by identifying possible age group differences, there were some limitations in variable selections. Lastly, the PHQ-8 questions were an optional module and only 12 states asked these questions in 2010. Therefore, the results are limited to respondents from those 12 states.

In addition to the limitations from BRFSS, there are other possible limitations from using survey data. Like most survey data, this study has limitations related to the selfreported symptoms of depression as well as other covariates. Additional social and clinical information such as income, behavioral observation, medication history, and a formal depression diagnosis would increase the precision of the study.

How age groups are divided could also yield different magnitudes of the associations between unemployment and depression in each age group. We used preferred terms for age groups from Academic Search Premier & SocIndex to divide age groups as the groups are based on developmental tasks, not merely on 5- or 10-year age brackets. In addition, the younger adult age range exactly corresponds to the range used in McGee and Thompson [2] thus enabling comparison of their results with the results of our study. Further research exploring the relationship using variations in age groups is recommended.