BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

176 

Do People Living in Rural Areas Have Less Severe Depression 
Problems? Evidence from NHIS 2019 Survey Data  

Gaonan You 

Department of Economics, Dalhousie University, Halifax, Canada 

*Corresponding author: Gaonan.You@dal.ca 

Abstract. Although many papers confirm that people living in rural areas are less likely to suffer from 
depression than those living in urban areas, most of them employed a straightforward linear 
regression, which is not convincing. Using data from the 2019 National Health Interview Survey 
(NHIS). This paper first identifies a proxy for depression level through Pearson's Chi-Squared test. 
Then, using OLS, this paper determined that place of residence had an impact on depression 
prevalence. To reduce the bias of the estimated the effect of place, this paper employs a propensity 
score matching method. Finally, the matched sample was retested to see if residence increased the 
risk of depression. This study suggests that there is a correlation between where you live and 
depression. 

Keywords: depression level; Urban-rural; mental health; America. 

1. Introduction 

Mental health refers to individual’s emotional, psychological, and social well-being. It is as 

important as physical health to individual’s overall health. European countries have seen an increase 

in depression, which has become a very common condition [1]. Rates in Canada are lower than in the 

United States, but on par with those in European countries [2]. However, the situation in America is 

the worst. According to the Centers for Disease Control and Prevention [3], mental health has become 

one of the most common illnesses in the United States. More than 50 percent of Americans have been 

diagnosed with a mental illness or disorder at some point in their lives. The mental health is 

interrelated with physical health, and it also has impact on the economy as the worse mental health 

can lead to the loss of productivity and increased expenditure in medicare. Therefore, understanding 

mental health and which factors influence mental health are of great importance.  

The economic impact of mental illness includes lost productivity and increased use of treatment 

resources. If mental and physical illness coexist, the longer the disease lasts, the more economics cost 

to society will spend. According to Colorado, an American insurance company, people with mental 

illness spends more than doubled of health care service than healthy people. This is because some 

patients with mental illness need treatment for other chronic diseases, others have unexplained illness, 

they tend to get more medical tests. Mental illness can affect the course of an individual's life and 

even spread to families and communities [4].  

The importance of mental health is obvious, but many surveys are based only on people living in 

urban areas or discuss the mental health of rural and urban residents separately. For example, when 

examining the impact of COVID-19 on residents' well-being, the report focused only on urban 

residents. Because of the lack of well-being of people living in the countryside, the psychological 

state of these people is difficult to speculate. Compared with urban areas, residents in rural areas face 

more difficulties. Such as high poverty, low employment rates and poor health care [5]. What is more, 

the COVID-19 pandemic that began in 2020 has undoubtedly exacerbated the gap of social status and 

income. Due to the lack of comparison of mental health of residents in the two places, the factors 

affecting the difference in mental health of residents in the two places are ambiguous. Do cities cause 

mental illness? The answer in the literatures is yes. This reflects the shortcomings of cities, which can 

be persistent [6]. If this discrepancy is not addressed, it will lead to insufficient information and the 

policies proposed by the government will be ineffective for residents in rural areas.  



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

177 

The biological and genetic causes of depression have been studied by many researchers and there’s 

still much ongoing research. Nevertheless, the social characteristics are also considered as risk factors 

of depression and the urban or rural residence is one of them. According to Maslow’s hierarchy of 

needs [7], people are motivated by five basic categories of needs: psychological, safety, love and 

belonging, esteem, and self-actualization. The five categories are often depicted as a pyramid with 

the categories sorting from lower to upper level. Every person has the desire to move up toward the 

self-actualization. The rapid development of urbanization in the United States since the 1950s 

provides people more resources to satisfy their needs. People living in urban areas have greater access 

to job market, better education, social location and entertainment resources. It is reasonable to assume 

that living in urban area is both culturally and economically beneficial. However, many literatures 

have found living in urban area could increase the risk of depression. Crowell et al. [8] use the 

Piedmont Health Survey data from North Carolina and find that urban residents are more vulnerable 

to depression and the risk of major depression1 decreases for some specific demographic subgroups 

living in rural area. They conclude that rural residence can be a buffer against major depression. 

Similar results are also found in Scotland. Instead of focusing on individual level depression data, 

McKenzie, Murray, and Booth [9] utilize macro-level data and use drug prescriptions within certain 

data zones as a proxy of depression level in an area. More urban residents are found to be associated 

with higher usage of prescription.  

Most of the literatures studying the urban and rural differences in depression level employ a 

straightforward qualitative analysis or linear regression framework or logistic regression. These 

models can barely be explained causally. This paper uses the 2019 National Health Interview Survey 

(NHIS) data to revisit this question by employing a propensity score match- ing method. I find that 

the depression becomes severer if individuals move to rural area. The probability of taking 

prescription for depression also increases by about 3% if individuals move to rural area.  

The next section describes the NHIS data and makes some descriptive analysis. Section 3 shows 

preliminary analysis for OLS regression in multiple specifications. It also introduces how this paper 

apply the propensity score matching method and estimate the effect of living in urban areas on the 

depression severity. Section 4 concludes the analysisand discusses the potential future work.  

Inevitably, the results of the survey are slightly biased. The difference in the results came from the 

fact that people in rural areas were less likely to be surveyed if they were depressed [10]. 

2. Data  

The NHIS data is a cross sectional household interview survey targeting at the residents in the 50 

states and the District of Columbia. The whole country is divided into 1,689 geographic areas. For 

some states, the geographic areas are divided into two strata while only one stratum is formed for the 

remaining states. Clusters of addresses are defined within each of these areas. Clusters are selected 

from the strata and the number of clusters selected is proportional to the number of clusters in the 

strata.  

This paper only uses the 2019 NHIS sample adult survey, which has 31,997 observations. The 

independent variable of interest is the urban-rural classification. This variable has four categories, 

and this paper only keeps those observations classified as living in large central metro or 

nonmetropolitan. Other variables related with social characteristics are also selected as control 

variables, including physical health status, family income, gender, education attainment, marital 

status, and age. Education attainment in the original survey has twelve categories. This paper re-

encodes the twelve categories to four categories: high school or less, college equivalent, bachelor, 

graduate2. Marital status has three categories. This paper combines the first two categories together 

and the new defined variable only tell whether the respondent lives with a partner or not. Because 

education attainment is included, this paper further restricts the observations to those whose age are 

from 24 to 65 (inclusive). The dependent variable of interest is severity of depression which is 

measured by the eight-item Patient Health Questionnaire depression scale (PHQ-8). The scale is 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

178 

recoded into four categories: none/minimal, mild, moderate, and severe3. Another variable which 

tells whether the respondent takes prescription for depression or not is also selected as a dependent 

variable following McKenzie, Murray, and Booth’s [7]work. This variable can be used as an indicator 

of the severity of depression. It is likely that individuals suffering from more severe depression tend 

to take prescription.  

Dropping observations with null values, the final sample size is 9398. Table 1 presents the basic 

summary statistics for the dependent variables and continuous independent variables.  

Table 1. Summary Statistics by Place of Residence 

 

3. Empirical results  

To study the effect of place of residence on the severity of depression, this paper first uses the 

linear regression  

Yi =α+δrurali +βXi +ei                           (1) 

Where Yi is dependent variables which are the PHQ-8 scale depression level of individual i, rurali 

is a dummy variable which equals one if the individual lives in rural area, Xi is a vector of control 

variables including age, age square, gender, marital status, physical health status, income class, and 

level of education. The parameter δ captures the effect of place of residence on the depression level. 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

179 

Another variable, whether the individual takes prescription for depression, is also used as a dependent 

variable for the regression. The prescription can be used as a proxy for depression level. Intuitively, 

individuals having severer depression problems are more likely to take prescription. Table 2 

summarizes the proportions of individuals taking prescription for depression at different depression 

levels. The proportion is increasing with the depression level. The Pearson’s Chi-squared test rejects 

the null hypothesis of equal proportion for the four groups at 5% level. Therefore, I conclude that the 

use of prescription can be used as a proxy for depression level. 

Table 2. Pearson’s Chi-squared Test of Equal Proportions 

 
Table 3 summarizes the regression results. Heteroskedasticity-robust standard errors are reported 

in the parentheses. Without including control variables, the place of residence has a statistically 

significant effect on the depression level at 5% level. Individuals living in rural area are more likely 

to have severer depression problems. But the effect becomes insignificant at 5% level if control 

variables are included. The magnitudes of the coefficient estimates from both regressions are quite 

small. It seems that the effect of place of residence on depression level is economically insignificant. 

However, the interpretation of the coefficient estimate is somewhat inappropriate. The linear 

regression model assumes that the dependent variable, PHQ-8 scale depression level, is continuous. 

But the numbers in the dataset only stands for different categories so they are not cardinal numbers. 

Instead, they are ordinal numbers. The differences between the categories are unknown and it is 

unreasonable to assume they are equal, so neither the estimate 0.096 nor 0.028 tells us the actual 

effect of place of residence on the depression severity. Nevertheless, the regression results still 

provide some insights into the difference of depression levels between individuals in rural and urban 

area. For regressions in the third and fourth column in Table 3, the dependent variable prescription is 

a dummy variable. Therefore, the linear regression model is in fact the linear probability model. 

Without including control variables, the coefficient estimate is statistically significant at 5% level. It 

indicates that the probability of taking prescription for depression increases 4.7% when individuals 

moving from urban area to rural area. Although there’s a small decrease in magnitude after including 

control variables, the coefficient estimate is still statistically significant at 5% level and indicates a 

3% probability increase of taking prescription.  

Although the estimates indicate that living in rural area has a negative impact on in- dividuals’ 

depression level, the model can hardly be interpreted causally. The problem of self-selection may 

weaken the result concluded from OLS estimates. In other words, unob- served factors, such as 

increase in crime rates and environmental degradation, may cause individuals move to rural 

communities and these factors may also contribute to the worse Mental health status. In the past, 

people moved to cities to look for job opportunities and better resources for education and so on. But 

the concentration of huge population in cities also brought the problem of scarcity of resources. 

People face with increasing living cost and crime rates. The environment in urban area is also 

deteriorating. The past years have witnessed redistribution of population not only in the United States 

but also many other countries. The development of technology in internet and communication also 

makes it pos- sible to work from home. People begin to reconsider the costs and gains of living in 

cities and many of them have chosen to move to rural area. This trend becomes even clearer since the 

outbreak of COVID-19.  

 

 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

180 

Table 3: OLS: Estimated Effects of living in Rural Area 

 
To address the potential issues, this paper employs a propensity score matching method to reduce 

the bias of the estimated effect of place of residence on the depression level. The propensity score 

matching method is typically used when the treated units and untreated units are not comparable or 

the selection of comparable units is difficult. Table 1 shows that the mean differences of most 

variables are significant and this indicates that the covariates are not balanced. The goal of propensity 

score matching is to produce covariate balance so that the distributions of covariates in the treated 

and untreated group are approximately equal. Then the difference in means can be used to estimate 

the treatment effect. Ho et al. [7] introduce a propensity score matching method which can reduce 

model dependence and produce more accurate estimated causal effects. This paper will employ their 

method to preprocess the observational data and then apply the linear regression model again with 

the matched sample.  

A probit model  

Pr(rurali = 1|Xi) = Φ(α + βXi)                    (2) 

Is used to estimate the propensity score. Φ(·) is the standard normal cumulative distribution 

function and Xi is the vector of control variables. Propensity scores are the fitted values from the 

probit regression and they are used to calculate the conditional probability of living in rural area 

regardless of actual place of residence. Figure 1 shows the distributions of propensity scores for the 

two groups. We can visually check the common support condition which is required to calculate the 

treatment effect. It’s apparent that the distributions overlap between the two groups and no units are 

outside the range of the common support. Therefore, it’s ensured that there will be adequate matches 

between treated and untreated units. 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

181 

 

Figure 1. Histogram of Propensity Score 

Next, a full matching algorithm is used so that every treated unit is matched to at least one untreated 

unit and every untreated unit is matched to at least one treated unit. Figure 2 visualizes the distribution 

of propensity score of the matched sample. Because full matching is used, there’s no unmatched 

treated or untreated unit.  

 

Figure 2. Distribution of Propensity Scores for the Matched Sample 

Table 4 summarizes the standard mean differences of characteristics between the un- treated and 

treated group. A value of standard mean difference which is close to zero indicates good balance of 

the characteristic. Comparing the standard mean differences of characteristics before and after 

matching, it’s apparent that the balance is far better after matching as the values are all closer to zero. 

Figure 3 visualizes the covariates balance before and after matching. The dotted line in the plot 

represents the threshold 0.1. It’s easy to see that the balance is quite poor before matching. The full 

matching improves the balance on all characteristics as almost all of them are within the threshold.  

 

 

 

 

 

 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

182 

Table 4. Standard Mean Difference of Characteristics Before and After Matching 

 
 

 

Figure 3. Love Plot of Covariate Balance 

Now this paper use the linear regression model again to estimate the effect of living in rural area 

on depression but with the matched sample. The matching weights are included to estimate the 

coefficients. Column 2 and Column 4 in Table 5 summarize the estimations with the matched sample. 

The reported standard errors in the parentheses are clustered- robust standard errors. Column 1 and 

Column 3 are OLS estimates with unmatched sample and they are included as comparisons. When 

the dependent variable is the depression level, the estimated coefficient of treatment variable now 

becomes significant at 5% level with the matched sample and there’s a slight increase in the 

magnitude. This indicates that living in rural area lead to severer depression problems. When the 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

183 

dependent variable is whether the individual takes prescription for depression, the significance and 

the magnitude have not changed. The estimation obtained from matched sample still predicts a 3% 

increase in probability of taking prescription when individuals move to rural area.  

Table 5. Estimated Effects of Living in Rural Areas With Matched Sample 

 

4. Conclusion  

This paper uses the NHIS 2019 survey data to identify the effect of living in rural area on the 

depression level. Although previous literatures find that living in urban area is associated with higher 

risk of depression and higher usage of prescription, I find that living in rural area is associated with 

severer depression and higher usage of prescription by employing a propensity score matching 

method.  

However, several concerns exist with this analysis. First, comparing with previous liter- atures, 

this paper uses cross sectional data from a specific year instead of panel data over several years. The 

number of observations is not that large as those literatures studying the same question. Panel data 

also contains more information than cross sectional data, such as the time effect. The estimates might 

be more accurate if panel data is used. Second, the two dependent variables of interest are categorical 

variables, and one of them has ordinal information. But this paper only use linear regression model 

to study the effect of place of residence on the dependent variables. For the ordinal dependent variable, 

it’s hard to interpret the results. For the binary dependent variable, the linear probability model might 

predict probability outside the range between zero and one. A logit/probit framework should be 

considered in the future work. Another concern is the potential failure of conditional in- dependence 

assumption. Although the propensity score matching method can help reduce the bias, it still requires 

the conditional independence assumption to hold in the applica- tion. But this assumption is often 

untestable and it is likely that there still exist omitted factors that simultaneously affect the treatment 

variable and the outcome variable in the application. Therefore, the structural model for the risk of 

depression should be studied in the future work, or use a different framework which does not rely 

heavily on conditional independence assumption.  

References  

[1] Ayuso-Mateos, J. L., Vázquez-Barquero, J. L., Dowrick, C., Lehtinen, V., Dalgard, O. S., Casey, P., 

Wilkinson, C., Lasa, L., Page, H., Dunn, G., Wilkinson, G., & ODIN Group (2001). Depressive disorders 



BCP Education & Psychology CEPC 2022 

Volume 8 (2023)  

 

184 

in Europe: prevalence figures from the ODIN study. The British journal of psychiatry : the journal of 

mental science, 179, 308–316. https://doi.org/10.1192/bjp.179.4.308 

[2] Patten SB, Wang JL, Williams JV, et al. Descriptive Epidemiology of Major Depression in Canada. The 

Canadian Journal of Psychiatry. 2006;51(2):84-90. doi:10.1177/070674370605100204 

[3] CDC.. About Mental Health. Retrieved from https://www.cdc.gov/mentalhealth/learn/index. (2021, June 

28) 

[4] Layard, R.. The Economics of Mental Health. (2017, January 1). Retrieved from 

https://wol.iza.org/uploads/articles/321/pdfs/economics-of-mental-health.pdf 

[5] Knapp M, Wong G. Economics and mental health: the current scenario. World Psychiatry. 2020 

Feb;19(1):3-14. doi: 10.1002/wps.20692. PMID: 31922693; PMCID: PMC6953559 

[6] J. Tom Mueller, Kathryn McConnell, Berne Burow, et al. Impacts of the COVID-19 pandemic on rural 

America. PNAS. 2020  

[7] Van Os, J. (2004). Does the urban environment cause psychosis? British Journal of Psychiatry, 184(4), 

287-288. doi:10.1192/bjp.184.4.287 

[8] Maslow, A., & Lewis, K. J. (1987). Maslow's hierarchy of needs. Salenger Incorporated, 14(17), 987-990.  

[9] Crowell, Bradford A., Linda K. George, Dan Blazer, and Richard Landerman. 1986. “Psy- chosocial Risk 

Factors and Urban/Rural Differences in the Prevalence of Major Depres- sion.” British Journal of 

Psychiatry 149 (3):307–314 

[10] McKenzie, Karen, Aja Murray, and Tom Booth. 2013. “Do urban environments increase the risk of 

anxiety, depression and psychosis? An epidemiological study.” Journal of Affective Disorders 150 

(3):1019–1024 

[11] Ho, Daniel E., Kosuke Imai, Gary King, and Elizabeth A. Stuart. 2007. “Matching as Nonparametric 

Preprocessing for Reducing Model Dependence in Parametric Causal In- ference.” Political Analysis 15 

(3):199–236 

. 
 

https://www.pnas.org/doi/10.1073/pnas.2019378118#con3