We use data from the Vida en Pandemia survey in Chile [42]. The survey was implemented 13–17 July 2020, through phone calls (from a large database of phone numbers obtained from the merge of administrative and private databases), with 2,545 adult respondents. The survey was designed to be balanced in terms of age, gender, and municipality of residence, to render representativeness across age, gender, and socioeconomic status. Because the data are deidentified, this study was not considered human-subjects research.

The survey contains information on basic demographic variables of respondents, including age, gender, educational level, socioeconomic characteristics, and municipality of residence. Further, it contains detailed information on employment, economic hardship, financial health, living arrangements, and a series of measures on self-reported well-being.

The main variables in this study are divided into five groups: financial distress, economic fragility, psychological well-being and emotions, utilization of mental health-care services, and conflict within the household. The measures of financial distress include an indicator variable for difficulty paying mortgage loans (“During the pandemic, have you had difficulties paying your mortgage monthly payment? 1 = yes, 0 = no) and an indicator variable for difficulty paying consumer debt (“During the pandemic, have you had difficulties to pay your consumer debt on time? 1 = yes, 0 = no). We also construct a measure from 0 to 4 regarding whether the individual has problems paying for basic goods and services: 1) basic goods, 2) medicine, 3) rent, and/or 4) school.

Savings are measured as the number of months the respondent believes that basic expenses can be afforded with savings if the main income source is lost. We create a lack of savings variable, that is an indicator that equals 1 if the individual reports that the household’s savings are below the sample median (3 months), and 0 otherwise. We measure economic fragility with indicator variables for unemployment (own or family member, separately) and income loss. The former is equal to 1 if the individual (or a family member) reports unemployment after March 2020 as a direct consequence of the pandemic, and 0 otherwise. The indicator for income loss is equal to 1 if the income reported in May 2020 is lower than the income reported for February 2020, and 0 otherwise.

Measures of psychological subjective well-being include sleep problems (“During the last week, did you have sleeping problems?”), and negative emotions, such as distress, frustration, worry, and restlessness (“During the last 2 weeks, how often have you felt … ?”). Responses to these variables are categorized by frequency in a 4-point Likert scale (4 = very frequently, 3 = frequently, 2 = sometimes, 1 = never). We construct a dichotomous variable for each measure, that is equal to 1 if the individual reports experiencing the abovementioned condition or feeling frequently or very frequently, and 0 otherwise. We create an indicator variable for poor well-being, based on a measure (“How would you describe your current well-being?” 5 = very good, 4 = good, 3 = neither good nor poor, 2 = poor, 1 = very poor), where the indicator is equal to 1 if the individual reports feeling poor or very poor well-being, and 0 otherwise. We also include an indicator for well-being deterioration that is equal to 1 if the individual reports that her well-being or mental health has worsened relative to February (before the pandemic), and 0 otherwise [“Would you say your well-being or mental health has (3 = improved, 2 = is the same, 1 = worsened) compared to before March 2020”]. Note that although the survey design is cross-sectional, this variable allows us to capture the change in well-being as reported by individuals.

Utilization of mental health-care services is measured with a set of indicator variables that capture a new diagnosis, new treatment, and/or new medication (for a mental health condition), after the start of the pandemic (1 = yes, 0 = no). Finally, we consider two variables that measure conflict within the household: an indicator for higher frequency of conflicts relative to before the pandemic, and a variable that measures severity of conflicts. Severity of the conflicts is a variable that goes from 0 to 5, where 0 indicates there is not an increase in conflicts, 1 indicates an increase in conflicts of low intensity, and 5 indicates an increase in conflicts of high intensity.

First, we analyze the relationship between economic fragility and financial distress, following a linear probability model (LPM): financial _ distres s ir = α + β 1 e c o n _ fragilit y ir + X ir ′ β 2 + A r + ε ir , where financial _ distress i r is a dependent variable of interest of individual i in region r (difficulty paying mortgage loans, difficulty paying consumer loans, difficulty paying for basic goods and services, and lack of savings), and e c o n _ fragility i represents own (or family) unemployment, or income loss. We include a set of control variables in the vector X i . Control variables include the following: an indicator for female, age-category dummies (18–24, 25–34, 35–44, 45–54, 55–64, and over 65), an indicator for household head, education-achievement indicators (complete or incomplete primary, secondary, technical, bachelor’s, and master’s level or more), indicators for the presence of young children (under the age of 12) in the household, migrant status, and pre-pandemic household income (expressed in natural logarithm). All regressions include a set of region indicators ( A r ) to account for fixed characteristics at the region level. Standard errors are robust to heteroskedasticity.

We then analyze the relationship between financial distress and psychological well-being, according to the following equation: y i r = α + β 1 financial _ distres s ir + β 2 unemployed ir + X ir ′ β 3 + A r + ε ir , where y i r is an outcome variable of individual i in region r (poor well-being, sleep problems, well-being deterioration, new diagnosis of mental health condition, new treatment, and new medication for mental health condition). We use the same set of controls and include own unemployment. We conduct linear probability models because estimating logit (probit) models with many dummy variables (e.g., a set of indicators by region) may introduce an incidental parameter problem. However, all our findings are robust to estimations via logit/probit models (results available upon request).

Supplementary Table A1 in the appendix reports descriptive statistics for all outcome variables, and Supplementary Table A2 reports descriptive statistics for individual characteristics. In the final sample employed in our baseline regression, 47% of the respondents are female, 7% are migrants, 57% have children, 47% have a college degree, and 22% experienced unemployment (own or family member) during the pandemic.

Supplementary Table A3 in the appendix reports correlations between mental health measures and individual characteristics. Women and younger adults experience on average worse mental health, facing higher rates of poor well-being, sleep problems and mental health deterioration. Being unemployed has a negative effect on mental health. Indicator variables for household head and for education levels appear not to be associated with worse mental health in this sample and income has a weak negative correlation with poor well-being.

Figure 1 shows the proportion of individuals facing financial problems by their labor status, we compare employed workers with workers who became unemployed, and workers with stable income with workers who experienced an income loss, during the pandemic. As expected, the figure shows that workers who became unemployed or suffered an income loss were more likely to face problems paying their debts and bills.

Table 1 explores the relationship between economic hardship and financial distress, which capture individuals reporting not being able to meet financial obligations during the pandemic period, after controlling for a comprehensive set of variables. Individuals who became unemployed or who experienced an income drop are more likely to experience financial problems, such as problems paying their mortgages [columns (1) and (2)], consumer debt [columns (3) and (4)], and difficulty paying basic goods and services [columns (5) and (6)]. They also have less savings to face future financial obligations.

The relationship between financial distress and unemployment or income drop is highly statistically significant and economically meaningful regardless of the model specification. Our estimates indicate unemployment increases the probability of having problems paying a mortgage loan or a consumer loan by roughly 23 and 24 percentage points, respectively, after controlling for pre-pandemic income, a comprehensive set of variables, education, and region fixed effects. Our estimates also suggest that, on average, unemployment increases roughly by 0.6 the number of different basic goods or services one may have problems paying for (i.e., basic goods, medicine, rent, and/or school). Finally, unemployment increases by 5 percentage points the probability of having insufficient savings to pay for basic needs for at least 3 months. Overall, results are qualitatively similar when we use an indicator of income drop as a measure of economic hardship.

Most of the coefficients associated with the control variables have the expected sign and are statistically significant. Older individuals have fewer financial problems and are less likely to have a lack of savings. Households with more children have more financial problems and less savings. Individuals with a higher pre-pandemic income are less likely to face any financial problem or lack savings. Finally, individuals with a family member who became unemployed also face a more difficult financial situation during the pandemic.

In Table 2 we explore the relationship between mental health and financial distress, controlling for a comprehensive set of variables. Results indicate individuals who experienced financial problems or had insufficient savings were more likely to experience poor well-being [columns (1) to (4)], sleep problems [columns (5) to (8)], and a deterioration of their well-being [columns (9) to (12)]. All our coefficients of interest are highly statistically significant and economically meaningful regardless of the model specification. For example, results in columns (9) and (10) indicate that having difficulty paying off a mortgage loan or consumer debt leads to a roughly 14-percentage-point increase in the probability of experiencing a deterioration of well-being. Column (11) indicates that having difficulty paying for one additional basic good or service leads to an 8-percentage-point increase in the probability of experiencing a deterioration of well-being. Finally, the results in column (12) suggest the lack of savings implies a 12-percentage-point higher probability of experiencing a deterioration of well-being.

Relatedly, in Supplementary Table A4 we examine whether financial problems are also associated with adults experiencing negative emotions such as distress, frustration, worry, and restlessness. All our measures of financial distress, including lack of savings, are associated with an increase of these negative emotions, ranging from 8 to 18 percentage points.

Table 3 examines whether financial distress affected utilization of mental health-care services during the COVID-19 pandemic, by analyzing whether individuals report having a new diagnosis and/or a new treatment of a mental health problem. On the one hand, the results reported in columns (2), (3), and (4) indicate individuals with fewer savings and financial problems (paying consumer debt and basic needs) are more likely to be newly diagnosed a mental health problem, but we do not find a statistically significant relationship between financial distress and mental-health treatment. Finally, columns (9) and (10) indicate individuals with problems paying off their mortgage and consumer debt are more likely to use mental health medication than individuals without those problems. We present a robustness check exercise in the Supplementary Material, where we re-estimate our baseline regressions using instrumental variables.

We examine conflict within the household as one potential mechanism linking financial distress and mental health outcomes. Specifically, Supplementary Table A5 investigates the relationship between financial distress and increases in the frequency and severity of conflict within the household. Columns (1) to (4) show our financial-distress measures are positively associated with a higher frequency of family conflicts. Columns (5) to (8) show a positive relationship between our measures of financial distress and the intensity of the family conflicts.

In view of our previous results, we augment in Table 4 our baseline regressions reported in Table 2 to control for the frequency of conflicts within the household. The relationship between financial distress and conflicts is highly significant, and the relationship between financial distress and mental health remains highly significant and economically meaningful. Thus, these results confirm financial distress is likely to affect mental well-being both directly and indirectly from a conflict channel.

An important limitation in our study is that the survey was not designed around a validated instrument to screen for anxiety and depression such as the Patient Health Questionnaire 4 (PHQ-4). We argue that the variables used in our study do a good job in assessing subjective well-being, as modeled by OECD Guidelines on Measuring Subjective Well-being [69]. Another limitation of our study is the representativeness of the sample, which was not constructed from a random sample from the entire population. Our sample covers a higher proportion of professionals and men, and therefore results cannot be extrapolated to the entire population. However, we expect the segments of the population analyzed to be less financially constrained [46]; if this is the case our estimates should be biased towards zero, and are therefore conservative estimates of the overall relationship between financial fragility and mental health. At the time of our study no nationally-representative surveys from the government or other institutions allowed to explore the link between financial fragility and mental health during the pandemic. A final limitation is that the survey data are cross-sectional and limit the possibility of analyzing dynamics, except some of the main outcome variables that are expressed as changes with respect to the pre-pandemic period.

Understanding the effects on mental health from the pandemic, as well as the effects of the policies implemented, is vital. Relief policies targeting economically fragile and vulnerable individuals are necessary. One key lesson of our findings is that financial assistance in the form of more loans increases debt and may worsen mental health and psychological well-being. Interventions designed with the objective of improving financial education are necessary to increase levels of precautionary savings, build financial resilience, and alleviate the psychological burden of debt in the future.