Abstract
Substantial evidence suggests that economic hardship causes violence. However, a large majority of this research relies on observational studies that use traditional violence surveillance systems that suffer from selection bias and over-represent vulnerable populations, such as people of color. To overcome limitations of prior work, we employed a quasi-experimental design to assess the impact of the Great Recession on explicit violence diagnoses (injuries identified to be caused by a violent event) and proxy violence diagnoses (injuries highly correlated with violence) for child maltreatment, intimate partner violence, elder abuse, and their combination. We used Minnesota hospital data (2004–2014), conducting a difference-in-differences analysis at the county level (n = 86) using linear regression to compare changes in violence rates from before the recession (2004–2007) to after the recession (2008–2014) in counties most affected by the recession, versus changes over the same time period in counties less affected by the recession. The findings suggested that the Great Recession had little or no impact on explicitly identified violence; however, it affected proxy-identified violence. Counties that were more highly affected by the Great Recession saw a greater increase in the average rate of proxy-identified child abuse, elder abuse, intimate partner violence, and combined violence when compared with less-affected counties.
Abbreviations
- DID
difference-in-differences
- ICD-9
International Classification of Diseases, Ninth Revision
- MHA
Minnesota Hospital Association
Violence is defined by the World Health Organization as “the intentional use of physical force or power, threatened or actual, against oneself, another person, or against a group or community, that either results in or has a high likelihood of resulting in injury, death, psychological harm, maldevelopment or deprivation” (1, p. 5). Different types of violence are all highly correlated, with victims of one type of violence being likely to be victims of another type of violence in their lifetime (2–6). Different types of violence also appear to have shared risk factors, which suggests they may arise from similar causes (4). One risk factor for all types of violence appears to be economic hardship (1).
One major source of economic hardship is an economic recession, which results in widespread deprivation. An economic recession is defined as a significant decline in a variety of economic indicators but typically gauged in the gross domestic product (7, 8). In the United States, the “Great Recession”—the longest recession since World War II—began in December 2007 and ended in June 2009 (7). There is limited research specifically examining the association of economic crises such as the Great Recession with violence, despite evidence that personal or community level economic hardship is associated with violence (9–11). The Great Recession was characterized by 3 main components, and then resulted in the stock market crash: 1) the collapse of the housing market; 2) high unemployment in the labor market; and 3) the fall of several financial services such as Lehman Brothers. In the first quarter of the Great Recession alone, approximately $7 trillion dollars in home equity were lost and approximately 1 in every 4 homeowners had an outstanding mortgage balance that exceeded the value of their home (12). Outstanding mortgages and inability to pay led to a foreclosure rate that increased 4-fold during the recession (12). Many families lost their homes and were displaced.
Foreclosures were a central feature of the Great Recession, and evidence suggests they may affect violence rates (10, 13). Generally, studies that examine the association between foreclosure, such as those that occurred during the Great Recession, and health find that foreclosure predicts worse health outcomes (14–16). Theoretical understanding of violence suggests that there is a link between the Great Recession and violent victimization. On the macro level, there are multiple pathways by which foreclosures could have an impact on health. For example, economic spillovers such as a drop in local property values, physical environmental impacts such as neglected properties, or social environmental impacts such as residential turnover can elicit stress and create frustration-aggression (14, 17). More specifically, the economic spillovers may create structurally disadvantaged neighborhoods that inhibit relationships via high population turnover and thus lead to low levels of informal social control and high levels of crime (18). Further, the role of future economic uncertainty leads to strain and conflict such as increased aggression (19). Understanding the social links that drive violence is useful for informing violence prevention programs and policy.
Existing studies on links between the Great Recession and violence have several methodological limitations. These studies have separately examined child maltreatment (20–26); intimate partner violence/domestic violence (27, 28); and serious violent-crime victimization, which includes attempted or completed sexual violence, robberies, and aggravated assaults (29). First, these studies exclude elder abuse. Second, most studies have relied on national surveys with self-reported violence perpetration and/or victimization, which may suffer from social desirability bias leading to underreporting of violent perpetration or victimization (30). Other studies use official records of violence, which may be underreported and underinvestigated because of reduced staffing of Child Protective Services workers during recession-related cuts in public funding (30–32). Further, there is evidence that official records are biased toward identification of violence in more highly scrutinized communities, such as those that are poor or have marginalized racial/ethnic identities (33). Third, existing research on the associations between economic hardship and violence has been vulnerable to confounding, which limits causal inference. Specifically, it is possible that the factors that make someone poor (e.g., limited emotional regulation skills) may also make them more likely to commit or be victims of violence (34, 35). A quasi-experimental design with a tool such as the difference-in-differences estimator uses exogenous shocks to delink this relationship. To our knowledge, only one study has used quasi-experimental design to examine the association between child abuse and the Great Recession (26).
One possible alternative data source for surveillance and research of violence is hospital discharge data on injuries. Injuries in hospital discharge data can be identified as violence-related with explicit International Classification of Diseases (ICD) codes. However, these codes rely on patient disclosure, or provider subjective assessment, that the cause of the injury was violence, leading to underutilization and potential bias (36). One way to supplement this violence identification is through proxy ICD codes (37). Proxy codes are used to identify instances where someone experienced an injury previously demonstrated to be highly correlated with violence. Using these proxy codes for violence identification, in combination with explicit codes, could help better describe the true burden and distribution of violence.
In light of the limitations of the existing literature, and to improve rigor and causal inference over past research, this study used a quasi-experimental design to achieve improved causal inference of the impact of foreclosures during the Great Recession on county-level violence victimization rates. For the main analyses, explicit codes were used to identify child abuse, elder abuse, and intimate partner violence in the hospital discharge data. However, prior research (37–55) supported using proxy codes in addition to explicit codes to improve measurement of violence cases. We, therefore, also included additional analyses of child abuse, elder abuse, and intimate partner violence identified by proxy codes. We hypothesized that the impact of the Great Recession as measured by county-level foreclosure rate change would be positively associated with all types of violence.
METHODS
Data
Minnesota hospital discharge data.
Hospital administrative data from 2004–2014 were obtained through the Minnesota Hospital Association (MHA). All Minnesota hospitals submit inpatient, outpatient, and emergency department claims data to the MHA. The MHA collects these data into a statewide administrative claims database. This database includes a data point for each patient encounter with a health-care provider and any diagnosis during that encounter specified with International Classification of Diseases, Ninth Revision (ICD-9), codes.
There are 3 categories of diagnostic codes in hospital claims data. ICD-9 codes describe the diagnosis of the condition and/or the treatment and are required for billing. E-codes and V-codes are modifiers to ICD-9 codes that provide additional detail, but they are not required. In the case of injuries, E-codes describe when and where the injury happened, to whom or by whom, and how. V-codes, also known as history codes, provide information about the history of the diagnosis but also are not required. Cross-sectional data on ICD-9 codes, E-codes, and V-codes from 2004–2014, averaged over every 2 years (e.g., 2004/2005, 2006/2007, and so forth) to stabilize the estimate from the MHA, were used to measure cases of violence for this study.
Population denominator data.
Annual population counts from 2004–2014 by county, sex, and age were used to calculate yearly violence-related injury rates and were obtained from the Surveillance, Epidemiology, and End Results (SEER) Program (56). The following denominators were used for each violence subtype: age ranges of 0–17 years for child abuse, ≥65 years for elder abuse, and ≥16 years for intimate partner violence.
Economic data.
Foreclosure data were obtained from Minnesota HousingLink (57). Minnesota HousingLink is a 501(c)3 organization. They are a primary source for housing information and resources in Minnesota (58). Sheriff’s sale records are used to define “foreclosures.” A sheriff’s sale is when the lender forces a sale of the property after the borrower has defaulted on their loan (59). To compile foreclosure data, HousingLink, along with the University of Minnesota’s Center for Urban and Regional Affairs, contacted all Minnesota county sheriff departments (57). Further details about their process can be found elsewhere (57).
Variable construction
Main outcome: explicit operationalization of violence.
Several ICD-9 codes, E-codes, and V-codes are used to indicate diagnoses of violence, such as ICD-9 code 995.83, “child sexual abuse.” These are defined as “explicit” codes. The specific explicit codes used to identify each type of violence are listed in Web Table 1 (available at https://doi.org/10.1093/aje/kwac114). These codes are assigned when a medical provider ascertains, or when a patient discloses, that the injury that brought them into the hospital was due to violence.
Supplemental outcome: proxy operationalization of violence.
Proxy ICD-9, E-codes, and V-codes are injury codes that do not require an explicit diagnosis of violence but have been demonstrated to be strongly suggestive of violence (Web Tables 1 and 2). For example, a code of 362.81 for retinal hemorrhage, or bleeding in the retina, among children less than 3 years old has been found to be indicative of child physical abuse (44). The proxy operationalizations were based on past research identifying codes that are highly correlated with a “gold standard” of violence identification, using in-depth medical record review (41–43), predictive modeling (40, 41), common diagnoses of known violent encounters (31–39), and linkage of hospital records with known cases of violence identified in Child Protective Services or Elder Protection Services (34, 42, 60).
Treatment: counties more affected by the Great Recession.
County-level foreclosure rates were chosen to define the treatment and control counties because of previous work demonstrating their strong association with violence (61) and the large impact that the Great Recession had on the housing market (12). Foreclosure rate change from 2007 to 2008 was used to define the treatment and control counties in Minnesota to represent the impact of the beginning of the Great Recession during this timeframe. Foreclosure rates were calculated by Minnesota HousingLink by dividing total number of foreclosures by number of residential parcels (28). Residential parcels included homes, apartments, and farms. The difference in the foreclosure rates between 2007 and 2008 was taken to calculate foreclosure rate change during the height of the Great Recession. The treated counties for the entire study period were defined as those with greater than the median foreclosure rate change (1.3%) between 2007 and 2008, and control counties for the entire study period were defined as those with less than or equal to the median foreclosure rate change between 2007 and 2008.
Statistical analysis
where Y represents the rate of violence (explicit or proxy) in county i at time t. Timet is the time trend for all counties. To reduce the impact of random noise in the data, the time period used is biennial (2-year averages), rather than annual. Evertreati is the indicator variable for counties that experienced a change in foreclosure rate over the course of the recession that was greater than the median foreclosure rate change (1), or less than or equal to the median foreclosure rate change (0), for county i. The interaction between Timet and Evertreati represents the difference in the time trend between the treated and control counties. The variable of interest is Treatmentposti. The corresponding parameter of interest, β3, represents the difference between treated and control counties in the change in violence due to the Great Recession. Treatmentposti “flips on” from 0 to 1 only for treated counties starting at the beginning of the recession in 2008 while control counties remain 0. The variable Treatmentposti encompasses both treatment condition and time. The model also includes county fixed effects, f, to account for time-invariant county attributes that may influence health-care utilization, such as percentage of women in each county, since women utilize health care more than men (63). Standard errors are adjusted to account for clustering at the county level to account for nonindependence over time within counties. Traditional DID design assumes that before-recession trends in violence are parallel in treated and untreated counties. To examine parallel trends, we plotted trends of the outcome in the years before the recession and visually compared treated and untreated trends (Web Figures 1–8) as well as testing the time trend regression. Before-recession trends were approximately parallel for all outcomes, except for explicitly identified elder abuse. Thus, the model specified above was chosen to allow independent time trends for treated and control counties. This model also assumes no unmeasured time-varying confounders (i.e., no factors that varied within counties over time affecting violence that also affected which counties were treated), and that there were no other changes at the time of the recession that differentially affected the outcome in counties by higher or lower foreclosure rate. One county in Minnesota was excluded from the analysis because of no foreclosure data.
Four main sensitivity analyses were performed. First, analyses were restricted to low-mobility counties to address concerns that changes in population distributions in a county, rather than the recession, may be driving changes in violence. Low-mobility counties were defined by ranking counties by the proportion of “nonmovers” (people in the county who had lived there for at least the past year), as reported on the American Community Survey, and identifying the 75% of counties with the highest proportion of nonmovers. A second set of analyses were performed excluding counties with populations under 5,000 residents, due to concerns that they may not have a large enough population to reliably estimate injury rates (64, 65), as well as excluding the 2 largest counties in Minnesota (Hennepin and Ramsey, the location of Minneapolis and St. Paul, respectively), due to concerns that the results might be driven by these large influential counties. Third, analyses were restricted so that the “post” time period included only what was defined as the Great Recession time period (2008/2009). Fourth, a falsification test was run using data from 2004–2006. Specifically, the main analysis was repeated but with shifting the beginning of the simulated Great Recession to occur in 2006 and excluding 2007 and beyond from this analysis to remove the actual Great Recession timeframe from this test.
An additional 4 supplemental sensitivity analyses were performed. The fifth sensitivity analysis repeated the main analysis but excluded those counties that were between the 40th percentile and the 60th percentile of foreclosure rate change. The sixth sensitivity analysis reclassified the treatment and control counties to compare the foreclosure rate changes of the top tertile (treated) versus the lower 2 tertiles (control). Both of these supplemental sensitivity analyses were performed due to concerns of small random fluctuations that could have led to a county’s classification in either the treatment or control group. A seventh and eighth supplemental sensitivity analyses were performed removing the interaction term for the fifth and sixth models, to investigate attenuation in the proxy-identified combined effect estimates in the fifth and sixth supplemental sensitivity models.
No null hypothesis significance testing was conducted (acceptable rate of type I and type II errors), and results instead focused on estimation (66). Data management and analysis were performed using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina), and STATA, version 14.2 (StataCorp LP, College Station, Texas), respectively.
RESULTS
Table 1 displays the socioeconomic and demographic characteristics of 86 counties in Minnesota by treatment or control condition. Generally, the treatment and control counties were roughly similar, with the exception that control counties tend to have a higher proportion of White people.
| Characteristic | Control, % (SD) | Treated, % (SD) |
|---|---|---|
| Median age, yearsc | 41.7 (4.9) | 40.0 (4.1) |
| <18 years old | 23.1 (2.0) | 24.4 (2.9) |
| ≥65 years old | 18.0 (4.1) | 15.2 (4.6) |
| Male sex | 50.0 (1.2) | 50.0 (.7) |
| Below the poverty line | 11.1 (2.8) | 9.9 (3.7) |
| Bachelor’s degree | 9.1 (1.9) | 10.1 (3.5) |
| Married | 23.0 (2.1) | 22.6 (2.2) |
| Race/ethnicity | ||
| White | 94.8 (3.8) | 92.4 (7.2) |
| Black | 0.7 (.8) | 1.3 (2.1) |
| Native American | 1.5 (3.0) | 2.26 (5.4) |
| Hispanic | 3 (3.0) | 3.3 (3.3) |
| Characteristic | Control, % (SD) | Treated, % (SD) |
|---|---|---|
| Median age, yearsc | 41.7 (4.9) | 40.0 (4.1) |
| <18 years old | 23.1 (2.0) | 24.4 (2.9) |
| ≥65 years old | 18.0 (4.1) | 15.2 (4.6) |
| Male sex | 50.0 (1.2) | 50.0 (.7) |
| Below the poverty line | 11.1 (2.8) | 9.9 (3.7) |
| Bachelor’s degree | 9.1 (1.9) | 10.1 (3.5) |
| Married | 23.0 (2.1) | 22.6 (2.2) |
| Race/ethnicity | ||
| White | 94.8 (3.8) | 92.4 (7.2) |
| Black | 0.7 (.8) | 1.3 (2.1) |
| Native American | 1.5 (3.0) | 2.26 (5.4) |
| Hispanic | 3 (3.0) | 3.3 (3.3) |
Abbreviation: SD, standard deviation.
a 2009 5-year American Community Survey estimate.
b The treated counties (n = 43) for the entire study period were defined as those with greater than the median foreclosure rate change (1.3%) between 2007 and 2008, and control counties (n = 43) for the entire study period were defined as those with less than or equal to the median foreclosure rate change between 2007 and 2008.
c Data expressed as median (standard deviation) number of years.
| Characteristic | Control, % (SD) | Treated, % (SD) |
|---|---|---|
| Median age, yearsc | 41.7 (4.9) | 40.0 (4.1) |
| <18 years old | 23.1 (2.0) | 24.4 (2.9) |
| ≥65 years old | 18.0 (4.1) | 15.2 (4.6) |
| Male sex | 50.0 (1.2) | 50.0 (.7) |
| Below the poverty line | 11.1 (2.8) | 9.9 (3.7) |
| Bachelor’s degree | 9.1 (1.9) | 10.1 (3.5) |
| Married | 23.0 (2.1) | 22.6 (2.2) |
| Race/ethnicity | ||
| White | 94.8 (3.8) | 92.4 (7.2) |
| Black | 0.7 (.8) | 1.3 (2.1) |
| Native American | 1.5 (3.0) | 2.26 (5.4) |
| Hispanic | 3 (3.0) | 3.3 (3.3) |
| Characteristic | Control, % (SD) | Treated, % (SD) |
|---|---|---|
| Median age, yearsc | 41.7 (4.9) | 40.0 (4.1) |
| <18 years old | 23.1 (2.0) | 24.4 (2.9) |
| ≥65 years old | 18.0 (4.1) | 15.2 (4.6) |
| Male sex | 50.0 (1.2) | 50.0 (.7) |
| Below the poverty line | 11.1 (2.8) | 9.9 (3.7) |
| Bachelor’s degree | 9.1 (1.9) | 10.1 (3.5) |
| Married | 23.0 (2.1) | 22.6 (2.2) |
| Race/ethnicity | ||
| White | 94.8 (3.8) | 92.4 (7.2) |
| Black | 0.7 (.8) | 1.3 (2.1) |
| Native American | 1.5 (3.0) | 2.26 (5.4) |
| Hispanic | 3 (3.0) | 3.3 (3.3) |
Abbreviation: SD, standard deviation.
a 2009 5-year American Community Survey estimate.
b The treated counties (n = 43) for the entire study period were defined as those with greater than the median foreclosure rate change (1.3%) between 2007 and 2008, and control counties (n = 43) for the entire study period were defined as those with less than or equal to the median foreclosure rate change between 2007 and 2008.
c Data expressed as median (standard deviation) number of years.
Overall, the foreclosure rate increased from 2005 to 2008 and peaked around 2008. From 2008 to 2010 there was some variation, but from 2010 onward foreclosure rates decreased (Figure 1).
The difference-in-differences model results (Table 2) showed relatively modest or close to null results for all types of explicitly identified violence. For example, when examining combined explicitly identified child abuse, elder abuse, and intimate partner violence, more highly affected (treated) counties saw an average decrease of 0.05 violence cases per 1,000 people (95% confidence interval (CI): −0.11, 0.01) over the course of the Great Recession compared with the change over the same period in less-affected (control) counties. This estimate had wide confidence intervals that crossed the null. All the individual violence subtypes (child abuse, elder abuse, and intimate partner violence) had similar direction and magnitude as the combined measure.
Average foreclosure rate trend in Minnesota from 2005–2012. ”Treated” counties for the entire study period were defined as those with greater than the median foreclosure rate change (1.3%) between 2007 and 2008, and control counties for the entire study period were defined as those with less than or equal to the median foreclosure rate change between 2007 and 2008.
Estimated Effects of the Great Recession on Violence Rates From Difference-in-Differences Models (All 86 Counties), Minnesota, 2004–2014
| Violence Type | Explicit | Proxy | ||
|---|---|---|---|---|
| Estimate | 95% CI | Estimate | 95% CI | |
| Violence subtypes combined | −0.05 | −0.11, 0.01 | 0.64 | 0.06, 1.21 |
| Child abuse | −0.10 | −0.32, 0.13 | 0.43 | 0.14, 0.72 |
| Elder abuse | 0.00 | −0.08, 0.07 | 2.04 | 0.00, 4.08 |
| Intimate partner violence | −0.04 | −0.11, 0.03 | 0.96 | 0.06, 1.86 |
| Violence Type | Explicit | Proxy | ||
|---|---|---|---|---|
| Estimate | 95% CI | Estimate | 95% CI | |
| Violence subtypes combined | −0.05 | −0.11, 0.01 | 0.64 | 0.06, 1.21 |
| Child abuse | −0.10 | −0.32, 0.13 | 0.43 | 0.14, 0.72 |
| Elder abuse | 0.00 | −0.08, 0.07 | 2.04 | 0.00, 4.08 |
| Intimate partner violence | −0.04 | −0.11, 0.03 | 0.96 | 0.06, 1.86 |
Abbreviation: CI, confidence interval.
Estimated Effects of the Great Recession on Violence Rates From Difference-in-Differences Models (All 86 Counties), Minnesota, 2004–2014
| Violence Type | Explicit | Proxy | ||
|---|---|---|---|---|
| Estimate | 95% CI | Estimate | 95% CI | |
| Violence subtypes combined | −0.05 | −0.11, 0.01 | 0.64 | 0.06, 1.21 |
| Child abuse | −0.10 | −0.32, 0.13 | 0.43 | 0.14, 0.72 |
| Elder abuse | 0.00 | −0.08, 0.07 | 2.04 | 0.00, 4.08 |
| Intimate partner violence | −0.04 | −0.11, 0.03 | 0.96 | 0.06, 1.86 |
| Violence Type | Explicit | Proxy | ||
|---|---|---|---|---|
| Estimate | 95% CI | Estimate | 95% CI | |
| Violence subtypes combined | −0.05 | −0.11, 0.01 | 0.64 | 0.06, 1.21 |
| Child abuse | −0.10 | −0.32, 0.13 | 0.43 | 0.14, 0.72 |
| Elder abuse | 0.00 | −0.08, 0.07 | 2.04 | 0.00, 4.08 |
| Intimate partner violence | −0.04 | −0.11, 0.03 | 0.96 | 0.06, 1.86 |
Abbreviation: CI, confidence interval.
The difference-in-differences model results for the proxy-identified violence are stronger in magnitude and opposite in direction compared with the explicitly identified violence measures. For example, when examining proxy-identified child abuse, elder abuse, and intimate partner violence combined, more highly-affected counties showed an average increase of 0.64 violence cases per 1,000 (95% CI: 0.06, 1.21) from before-recession to recession, over and above the changes over this time period in less-affected counties. However, this estimate also had wide confidence intervals that crossed the null. All violence subtypes had positive coefficients in line with our hypothesis, with the greatest impacts appearing to be on elder abuse (2.04 per 1,000, 95% CI: 0.00, 4.08) and intimate partner violence (0.96 per 1,000, 95% CI: 0.06, 1.86), with proxy-identified child abuse (0.43 per 1,000, 95% CI: 0.14, 0.72) showing less impact.
The main sensitivity analyses (Table 3) estimating these associations among counties with low mobility and medium population size found results very similar to the main analyses, suggesting that population change and size are not important. The sensitivity analysis estimating these associations including only the defined recession period was also similar to the main analysis except for results for proxy-identified elder abuse and proxy-identified combined violence rates. These estimates changed directionality as compared with the main analysis. Finally, the falsification sensitivity analysis found that the association between the impact of the Great Recession and explicitly and proxy-identified violence were close to the null, as expected. One exception was for proxy-identified elder abuse, which indicated a positive effect of a 2006 event on rates, although the magnitude of association was muted in comparison with the main analysis.
The supplemental sensitivity results are available in the Web Material (Web Table 3). Briefly, these supplemental sensitivity analyses produced results very similar to those from the main analysis. The most notable difference was the proxy-identified elder-abuse results. In general, proxy-identified elder-abuse results appear to be the least robust of the outcomes, thus there should be a cautious interpretation of findings for this outcome.
Estimated Effects of the Great Recession on Violence Rates From Difference-in-Differences Sensitivity Analysis, Minnesota, 2004–2014
| Explicit | Proxy | ||||
|---|---|---|---|---|---|
| Sensitivity Analysis | No. of Counties | Estimate | 95% CI | Estimate | 95% CI |
| Low mobility | 65 | ||||
| Violence subtypes combined | −0.08 | −0.15, 0.00 | 0.58 | −0.14, 1.3 | |
| Child abuse | −0.16 | −0.43, 0.12 | 0.32 | 0.09, 0.56 | |
| Elder abuse | −0.03 | −0.12, 0.05 | 1.92 | −0.54, 4.38 | |
| Intimate partner violence | −0.05 | −0.14, 0.03 | 1.19 | 0.04, 2.35 | |
| Medium population | 81 | ||||
| Violence subtypes combined | −0.05 | −0.11, 0.02 | 0.53 | −0.07, 1.13 | |
| Child abuse | −0.11 | −0.35, 0.13 | 0.46 | 0.14, 0.77 | |
| Elder abuse | −0.01 | −0.08, 0.07 | 2.10 | −0.08, 4.28 | |
| Intimate partner violence | −0.03 | −0.1, 0.05 | 1.03 | 0.08, 1.99 | |
| 2008–2009 | 81 | ||||
| Violence subtypes combined | −0.08 | −0.19, 0.03 | −1.45 | −2.33, −0.56 | |
| Child abuse | −0.31 | −0.69, 0.08 | 0.28 | −0.11, 0.67 | |
| Elder abuse | −0.02 | −0.12, 0.09 | −10.23 | −13.38, −7.08 | |
| Intimate partner violence | 0.01 | −0.07, 0.09 | 0.72 | −0.54, 1.97 | |
| Falsification | 86 | ||||
| Violence subtypes combined | 0.05 | −0.05, 0.14 | 0.03 | −0.76, 0.83 | |
| Child abuse | 0.11 | −0.23, 0.45 | 0.13 | −0.03, 0.29 | |
| Elder abuse | 0.08 | −0.02, 0.18 | 1.60 | −1.29, 4.49 | |
| Intimate partner violence | 0.00 | −0.07, 0.06 | −0.10 | −1.19, 0.99 | |
| Explicit | Proxy | ||||
|---|---|---|---|---|---|
| Sensitivity Analysis | No. of Counties | Estimate | 95% CI | Estimate | 95% CI |
| Low mobility | 65 | ||||
| Violence subtypes combined | −0.08 | −0.15, 0.00 | 0.58 | −0.14, 1.3 | |
| Child abuse | −0.16 | −0.43, 0.12 | 0.32 | 0.09, 0.56 | |
| Elder abuse | −0.03 | −0.12, 0.05 | 1.92 | −0.54, 4.38 | |
| Intimate partner violence | −0.05 | −0.14, 0.03 | 1.19 | 0.04, 2.35 | |
| Medium population | 81 | ||||
| Violence subtypes combined | −0.05 | −0.11, 0.02 | 0.53 | −0.07, 1.13 | |
| Child abuse | −0.11 | −0.35, 0.13 | 0.46 | 0.14, 0.77 | |
| Elder abuse | −0.01 | −0.08, 0.07 | 2.10 | −0.08, 4.28 | |
| Intimate partner violence | −0.03 | −0.1, 0.05 | 1.03 | 0.08, 1.99 | |
| 2008–2009 | 81 | ||||
| Violence subtypes combined | −0.08 | −0.19, 0.03 | −1.45 | −2.33, −0.56 | |
| Child abuse | −0.31 | −0.69, 0.08 | 0.28 | −0.11, 0.67 | |
| Elder abuse | −0.02 | −0.12, 0.09 | −10.23 | −13.38, −7.08 | |
| Intimate partner violence | 0.01 | −0.07, 0.09 | 0.72 | −0.54, 1.97 | |
| Falsification | 86 | ||||
| Violence subtypes combined | 0.05 | −0.05, 0.14 | 0.03 | −0.76, 0.83 | |
| Child abuse | 0.11 | −0.23, 0.45 | 0.13 | −0.03, 0.29 | |
| Elder abuse | 0.08 | −0.02, 0.18 | 1.60 | −1.29, 4.49 | |
| Intimate partner violence | 0.00 | −0.07, 0.06 | −0.10 | −1.19, 0.99 | |
Abbreviation: CI, confidence interval.
Estimated Effects of the Great Recession on Violence Rates From Difference-in-Differences Sensitivity Analysis, Minnesota, 2004–2014
| Explicit | Proxy | ||||
|---|---|---|---|---|---|
| Sensitivity Analysis | No. of Counties | Estimate | 95% CI | Estimate | 95% CI |
| Low mobility | 65 | ||||
| Violence subtypes combined | −0.08 | −0.15, 0.00 | 0.58 | −0.14, 1.3 | |
| Child abuse | −0.16 | −0.43, 0.12 | 0.32 | 0.09, 0.56 | |
| Elder abuse | −0.03 | −0.12, 0.05 | 1.92 | −0.54, 4.38 | |
| Intimate partner violence | −0.05 | −0.14, 0.03 | 1.19 | 0.04, 2.35 | |
| Medium population | 81 | ||||
| Violence subtypes combined | −0.05 | −0.11, 0.02 | 0.53 | −0.07, 1.13 | |
| Child abuse | −0.11 | −0.35, 0.13 | 0.46 | 0.14, 0.77 | |
| Elder abuse | −0.01 | −0.08, 0.07 | 2.10 | −0.08, 4.28 | |
| Intimate partner violence | −0.03 | −0.1, 0.05 | 1.03 | 0.08, 1.99 | |
| 2008–2009 | 81 | ||||
| Violence subtypes combined | −0.08 | −0.19, 0.03 | −1.45 | −2.33, −0.56 | |
| Child abuse | −0.31 | −0.69, 0.08 | 0.28 | −0.11, 0.67 | |
| Elder abuse | −0.02 | −0.12, 0.09 | −10.23 | −13.38, −7.08 | |
| Intimate partner violence | 0.01 | −0.07, 0.09 | 0.72 | −0.54, 1.97 | |
| Falsification | 86 | ||||
| Violence subtypes combined | 0.05 | −0.05, 0.14 | 0.03 | −0.76, 0.83 | |
| Child abuse | 0.11 | −0.23, 0.45 | 0.13 | −0.03, 0.29 | |
| Elder abuse | 0.08 | −0.02, 0.18 | 1.60 | −1.29, 4.49 | |
| Intimate partner violence | 0.00 | −0.07, 0.06 | −0.10 | −1.19, 0.99 | |
| Explicit | Proxy | ||||
|---|---|---|---|---|---|
| Sensitivity Analysis | No. of Counties | Estimate | 95% CI | Estimate | 95% CI |
| Low mobility | 65 | ||||
| Violence subtypes combined | −0.08 | −0.15, 0.00 | 0.58 | −0.14, 1.3 | |
| Child abuse | −0.16 | −0.43, 0.12 | 0.32 | 0.09, 0.56 | |
| Elder abuse | −0.03 | −0.12, 0.05 | 1.92 | −0.54, 4.38 | |
| Intimate partner violence | −0.05 | −0.14, 0.03 | 1.19 | 0.04, 2.35 | |
| Medium population | 81 | ||||
| Violence subtypes combined | −0.05 | −0.11, 0.02 | 0.53 | −0.07, 1.13 | |
| Child abuse | −0.11 | −0.35, 0.13 | 0.46 | 0.14, 0.77 | |
| Elder abuse | −0.01 | −0.08, 0.07 | 2.10 | −0.08, 4.28 | |
| Intimate partner violence | −0.03 | −0.1, 0.05 | 1.03 | 0.08, 1.99 | |
| 2008–2009 | 81 | ||||
| Violence subtypes combined | −0.08 | −0.19, 0.03 | −1.45 | −2.33, −0.56 | |
| Child abuse | −0.31 | −0.69, 0.08 | 0.28 | −0.11, 0.67 | |
| Elder abuse | −0.02 | −0.12, 0.09 | −10.23 | −13.38, −7.08 | |
| Intimate partner violence | 0.01 | −0.07, 0.09 | 0.72 | −0.54, 1.97 | |
| Falsification | 86 | ||||
| Violence subtypes combined | 0.05 | −0.05, 0.14 | 0.03 | −0.76, 0.83 | |
| Child abuse | 0.11 | −0.23, 0.45 | 0.13 | −0.03, 0.29 | |
| Elder abuse | 0.08 | −0.02, 0.18 | 1.60 | −1.29, 4.49 | |
| Intimate partner violence | 0.00 | −0.07, 0.06 | −0.10 | −1.19, 0.99 | |
Abbreviation: CI, confidence interval.
DISCUSSION
The aim of this study was to assess the impact of the Great Recession on violence victimization. We used Minnesota hospital discharge and foreclosure data with a quasi-experimental design to estimate changes to violence rates from before the recession to after the recession in counties most affected by the recession versus changes over the same time period in counties less affected by the recession. To our knowledge, this study is unique in its assessment of explicit and proxy coding of violence using a quasi-experimental design. Our findings suggest that the change in foreclosure rate during the Great Recession had little or no impact on explicitly identified violence (injuries directly identified as violent) but increased risk for proxy-identified violence (injuries associated with violence). Specifically, over the course of the Great Recession, counties with higher foreclosure rates saw a greater increase in the average rate of proxy-identified child abuse, elder abuse, intimate partner abuse, and their combination when compared with less–foreclosure-affected counties.
This analysis did not find evidence that the impact of Great Recession influenced rates of explicitly identified violence. These results contrast with findings from prior studies assessing how changes to individuals’ perceptions of economic hardship during the Great Recession are associated with violence perpetration or victimization (20, 21, 24). A portion of this observational literature focuses on consumer sentiment, a subjective measure of the economy during the Great Recession. These studies found that decreasing consumer confidence during the Great Recession was associated with increased self-reported psychological aggression from mother to child (21), self-reported increases in serious violent victimization (including attempted or completed sexual violence, robberies, and aggravated assaults) (29), and self-reported increased risk of maternal spanking (20), which is correlated with other types of child maltreatment (67, 68). It is plausible that it is the perception of economic uncertainty, as measured by consumer sentiment, rather than objective measures of economic change, such as foreclosure changes, that are important for violence risk (69). Of note, however, the null results from the present study were found specifically for explicitly identified violence, whereas proxy-identified violence increased in response to foreclosure rate change. The reasons for these divergences by type of identification are not clear. It may be due to selection bias. Further, it is known that explicit codes of violence are often underutilized and may not be representative of the true distribution of violence-related injuries (37, 70–73). Specifically, explicitly identified violence cases may comprise more marginalized people, such as people of color (74–76) and people of lower socioeconomic status (33, 77), whereas proxy codes may yield a more representative distribution of violence and less prone to bias. The communities that are composed of more marginalized people may have more social cohesion and resilience in times of hardship, thus explaining the divergences by identification type (78).
This study has important limitations. First, this study does not include those who experience violent events but do not go to the hospital; therefore, it may be an oversample of those with health insurance in the population. That said, more severe or urgent injuries likely bring people in for care despite the lack of health insurance coverage (79). Second, the results may not be generalizable outside of the region. However, Minnesota is similar to other midwestern states in the United States (80). Third, analyses were performed at the county level. This limits inference regarding the individual level (81, 82). Although ecological studies have value for studying a variety of structural-level factors impacting health, the ideal measurement of the association between the Great Recession and violence would be multilevel with data available on both the individual and community level for both exposure and outcome. Fourth, the foreclosure process can take up to 290 days from initial payment default in Minnesota, and the sheriff sales measure foreclosures at one time point during this process (57). Therefore, foreclosures defined by sheriff sales may be missing individuals that began the foreclosure process but were about to redeem their mortgage payment. However, foreclosures defined using sheriff sales represent someone further down the path in the process, who thus may have experienced a higher level of financial duress. Fifth, the DID approach has several assumptions that are not testable (e.g., no unmeasured time-varying confounders). If these assumptions are not met, the results may be biased. Nevertheless, sensitivity analyses were performed to try and address possible confounders. Sixth, this study does not include death-certificate data. Therefore, those that died due to violence outside the hospital were not included in the study. However, the proportion of those that die due to violence is approximately 1% (83), and thus the addition of these data would likely not change the results of the study. Seventh, the denominator for foreclosure rate, residential parcels, includes those who may not have a mortgage on their house. In spite of this limitation, there is existing literature supporting this type of population denominator to study the association between foreclosure rates and violence (13, 84, 85).
These results suggest several areas for future research. Future validation studies on the utility of proxy-identified violence may help to illuminate the true burden of violence in the population through expanding the tools to do rigorous surveillance and research on violence. Most violence surveillance systems suffer from inherent bias such as racism (73–75) or classism (33, 76). Last, this analysis is, to our knowledge, the first to test the association between the Great Recession and elder abuse. Future research understanding this association is needed.
In conclusion, this study suggests that the Great Recession, measured by foreclosure rates, led to an increase in certain types of violence. These findings have implications as researchers begin to investigate the potential long-lasting effect of the COVID-19 pandemic and its resulting recession (86).
ACKNOWLEDGMENTS
Author affiliations: Division of Epidemiology and Community Health, University of Minnesota School of Public Health, Minneapolis, Minnesota, United States (N. Jeanie Santaularia, Theresa L. Osypuk, Susan M. Mason); Minnesota Population Center, University of Minnesota, Minneapolis, Minnesota, United States (N. Jeanie Santaularia, Theresa L. Osypuk); and Division of Environmental Health Sciences, University of Minnesota School of Public Health, Minneapolis, Minnesota, United States (Marizen R. Ramirez).
This work was supported by the Minnesota Population Center (grant P2C HD041023) and the Interdisciplinary Population Health Science Training Program (grant T32HD095134), both of which are funded by the Eunice Kennedy Shriver National Institute for Child Health and Human Development.
The majority of the data that support the findings of this study are available from Minnesota Hospital Association, but restrictions apply to the availability of these data, which were used under license for the present study, and so are not publicly available. The remaining data are from HousingLink and are available from their website https://www.housinglink.org/.
The views expressed in this article are those of the authors and do not reflect those of the National Institutes of Health.
Conflict of interest: none declared.
