Introduction
Poverty is associated with violent crime. As illustration, consider the association between county poverty rates and homicide rates in the United States. The correlation is fairly strong (r = 0.52) and highly statistically significant (p < 0.001).
The association between county poverty rate and its homicide rate is substantial. But why?
To look more closely at how the average homicide rate increases with poverty rate, we can split the counties into ten bins with roughly equal total populations. That is shown below.
Poverty rate decile (from least to most poor) and homicide rate.
The association is even clearer now. The 10% of people living in the poorest counties have almost 6 times higher homicide rate than the 10% living in the least poor counties. That poverty correlates with violent crime is not in dispute.
What is in dispute is whether that association is due to poverty causing violent crime. This is not merely a reflexive and baseless “correlation doesn’t imply causation” quip. There are good reasons—supported by strong empirical evidence—to seriously cast doubt on the causal potency of poverty on violent crime.
There are many odd facts in need of explanation if one is to believe that economic conditions have a great causal impact on violent crime. Why do NFL players, with average salaries in the millions of dollars, not have lower violent crime arrest rates than the general population (Leal et al., 2015)? And why did the homicide rate fall during the Great Depression, one of the most severe economic crises? The answer, it turns out, is that poverty and violence are correlated for other reasons.
There are three (not mutually exclusive) potential explanations for the poverty-violence association:
Causal: poverty affects propensity for violence.
Reverse causal: crime harms economic conditions.
Selection/confounding: other variables affect both economic prospects and propensity for violence.
As I will argue, the causal effect of poverty on violence is weak, and potentially practically null. Instead, the association between poverty and violent crime is mostly the product of selection: traits like low cognitive ability, mental disorders and others, negatively impact economic success and are also risk factors for committing violent crime.
I will first review some evidence regarding reverse causality and selection, and establish their important role in the poverty-violence association. Afterwards, I will review a broad body of causally informative evidence and evaluate to what extent poverty affects violent crime.
Reverse causality
The poverty-crime association is clearly affected by reverse causality to some extent; that is, criminality harms economic prospects.
This is firstly true at the individual level. Though the typical criminal has poor labor market prospects prior to any convictions1, individuals see a significant drop in earnings and employment rates immediately following a criminal charge (Agan et al., 2023; Brown, 2018).
It is also true at the community level. Crime is costly for society. Even ignoring large intangible costs, the direct tangible costs of crime alone amount to substantial figures (Miller et al., 2021; Wickramasekera et al., 2015).
Crime worsens local economic conditions; it erodes productivity and well-being (Donovan et al., 2024). It makes a neighborhood less desirable to live in and to conduct business in, resulting in selective outmigration or “urban flight” (Cullen & Levitt, 1999). In that sense, crime not only has a direct economic costs (to the individual and other locals), it also has an indirect effect by contributing to selection — people who are productive and financially able tend to move away from high-crime areas. That reinforces the association between poverty and crime, regardless of whether poverty affects crime or not.
Selection
Wealth is not just randomly assigned. It is influenced by many factors, including personal characteristics. The same individual characteristics, personality traits and decision-making skills that tend to lead to worse economic prospects also increase the tendency to commit crime. This is selection, and there is ample evidence that it plays an important role in producing the poverty-crime association. That is, even absent any causal or reverse causal effects, you would expect poverty and crime to be correlated.
Criminals are different from non-criminals in many ways, and people sort strongly geographically and socioeconomically. Caspi et al. (2017) found that a “segment comprising 22% of the cohort accounted for 36% of the cohort’s injury insurance claims; 40% of excess obese kilograms; 54% of cigarettes smoked; 57% of hospital nights; 66% of welfare benefits; 77% of fatherless child-rearing; 78% of prescription fills; and 81% of criminal convictions.”
There are many characteristics that intuitively sound like they might influence both economic success and propensity for crime: impulsivity and lack of self-control, drug- and alcohol misuse, among others. For this remainder of the section, I will consider cognitive ability and mental disorders as example risk factors, though they are surely not exhaustive.
Cognitive ability is a well-known contributor to economic success, and low cognitive ability is also a risk factor for violent crime. In large population studies, low cognitive ability is associated with increased probability of criminal offending (Schwartz et al., 2015), and violent criminals have lower average cognitive ability than the general population (e.g., Khoshnood et al., 2020). It is even reflected in self-reported criminal participation, showing that it’s not merely about not getting caught (Jacob et al., 2019).
Importantly, the association between low cognitive ability and crime is not due to confounding with, say, socioeconomic status. Sibling studies reveal that most of the association remains within-family; that is, of siblings reared in the same household, the sibling with lower cognitive ability offends more often, and the association is only slightly lower than the cross-sectional association (Frisell et al., 2012).
The relationship between cognitive ability and proportion convicted of violent crime. Men in the lowest cognitive ability category had 20-30% chance of being convicted of a violent crime, whereas those in the highest category had less than 2.5% chance. Source: Frisell et al. (2012).
Cognitive ability is therefore a clear example of a confounding variable contributing to differences in both economic success and criminal offending rates. But it is not the only relevant variable in selection.
Mental disorders are also known to predict labor market and educational outcomes, even using sibling fixed-effects (Mousteri et al., 2019; Nordmo et al., 2022). Mental disorders are more common in low-SES environments (Kinge et al., 2021), which is in large part due to such selective movements (Sariaslan et al., 2016).
Mental disorders are also associated with higher rates violent offending and reoffending (Dean et al., 2024; Whiting et al., 2020; Whiting et al., 2021), a relationship that again persists when controlled for unmeasured familial confounding (Sariaslan et al., 2020; Chang et al., 2015; Yukhnenko et al., 2023).
Association between number of psychiatric disorders and violent reoffending in male released prisoners in Sweden. Source: Chang et al. (2015).
Because these associations exist when familial confounding is controlled for, these associations are not merely recapturing the effect of poverty itself. At least some selective factors contribute to the poverty-crime association, without poverty being the cause of it.
These characteristics are also heritable (e.g., Panizzon et al., 2014; Athanasiadis et al., 2022), which makes them contribute to intergenerational transmission and genetic confounding in the poverty-crime association.
The causal effect of poverty
It is clear that reverse causality and selection together play a major role in producing the association between poverty and crime. This means that the existence of an association between poverty and crime cannot be interpreted as poverty causing crime. It does not matter how many times the association is replicated, because a systematic association is expected even if poverty has no causal effect on violent crime. The existence of the association is not what’s in question; its etiology is.
How then do we estimate the causal effect? Several approaches have been used to eliminate most of the contamination from selection bias and reverse causality.
Sibling studies
A clever approach of disentangling the effects comes from exposure-discordant sibling studies. The idea is to compare siblings with the same parents who nevertheless were exposed to different economic conditions growing up — for example, the parents earned much more money while raising one of the children, or they had moved to a better neighborhood. The utility of comparing siblings is that by design it “controls” for anything the siblings share, which includes things that might otherwise difficult to capture (half of their genes, subtle environmental similarities, etc).
Two huge studies from Scandinavia are particularly noteworthy. Sariaslan et al. (2013) looked at the effect of neighborhood deprivation on violent crime convictions in Sweden (180k siblings). Sariaslan et al. (2021) considered the effect of parental income in Finland instead (427k siblings). The results of these two such studies are summarized below.
Neighborhood deprivation and lower income is associated with higher violent conviction rates. However, for siblings exposed to different socioeconomic conditions, there is no effect.
The crude association shows a strong “effect” of lower socioeconomic status on violent crime convictions. This is merely restating what we already knew, that the two are correlated. When adjusted for measured confounders, the associations of low income or neighborhood deprivation with violent crime convictions become small, but remain statistically significant. However, when using sibling comparison, which adjusts for unobserved familial confounding, the relationship becomes practically null and non-significant. Sariaslan et al. (2021), in particular, establishes a tight bound around no effect of income on violent crime conviction rate. They also tested alternative measures of socioeconomic status, and all results were consistent with this central finding.
A different study also considered sibling differences in exposure to paternal conviction (Sivertsson et al., 2023). Again, for violent crime convictions the pattern was the same. The crude effect was substantially reduced with measured controls, but remained significant. But in the sibling comparison, there was no effect of paternal conviction on male violent crime conviction probability.
Overall, sibling studies provide strong evidence against the idea that poverty has any substantial causal effect on violent offending.2
Within-individual studies
The within-individual design asks whether a change in exposure within a person is associated with a change in outcome (in that same person). Instead of using siblings as controls, it uses the person itself. Anything that is fixed within the person over the analyzed period is controlled for by design.
Airaksinen et al. (2021) used this research design to analyze the effect of neighborhood deprivation on violent crime (n > 500,000). As shown below, the between-individual association was far larger than the within-individual association, indicative of tremendous selection and/or reverse causality. That is, while people in more deprived neighborhoods tend to have much higher violent crime rates, within-individual changes in neighborhood deprivation have little effect on violent crime rates.
Neighborhood deprivation is cross-sectionally associated with more violent crime. But within-individual changes in neighborhood deprivation have little meaningful effect.
The within-individual effect is statistically significant, but practically tiny. In fact, the effect from the least to the most deprived decile is only a measly 3% as large as the cross-sectional association [100% × (1.21 - 1) / (7.98 - 1)]. This suggests that the causal effect of poverty is very marginal. This is possibly even an overestimate of the causal effect, given that it plausibly could be confounded by variables not time-invariant within the individual.3
Sariaslan et al. (2017) used a similar research design for released prisoners with psychotic disorders. They analyzed whether placing the released prisoners in more or less deprived neighborhoods affected their reoffending risk. At this point it should be unsurprising that none of the socioeconomic neighborhood characteristics had any significant effect on reoffending rates in the within-individual comparisons.
In summary, within-individual studies also provide strong evidence against the idea that poverty has a strong causal effect on violent crime.
Income-transfer and jobs programs
In a recent review, Ludwig & Schnepel (2024) review the literature pertaining to whether jobs and income-transfer programs affect crime. The overall conclusion was clear. Property crimes, directly motivated by profit, are potentially responsive to such programs, but “such policies have little, if any, systematic effect on violent crime.”
The authors note in their review: “Taken together, the best available data and evidence suggest that economic conditions contribute importantly to property crime but are not the key driver of the crime problem itself–that is, of violent crime. The things that matter for violence seem to be correlated with income poverty but are not the same thing as income poverty.” This is, of course, consistent with the idea that it’s largely driven by selection.
Review of the economic impact on (a) property crime, and (b) violent crime. There is no systematic effect on violent crime. Source: Ludwig & Schnepel (2024)
Macroeconomic evidence
If economic conditions affect crime, an obvious prediction is that trends in national crime rates will coincide with the state of the national economy. It would be expected that when the economy improves, violent crime drops; that when the economy deteriorates, violent crime increases. But this expectation is clearly not always borne out. For example, during the Great Depression, homicide rates fell, contrary to what’s expected by the poverty-as-cause hypothesis.
Numerous studies have tested whether changes in the national economy over time are systematically associated with changes in crime rates. As reviewed by Ellis, Farrington & Hoskin in the Handbook of Crime Correlates4, the findings of such studies do not point in any systematic direction. The symmetry in findings is consistent with there being little to no causal effect. Similarly, Ramos (2014) also finds no effect of changes in poverty on violent crime, and concludes “results confirm prior research that concludes that poverty does not have a significant effect on violent crime.” Ludwig & Schnepel (2024) reviewed the macroeconomic evidence and came to the same conclusion: macroeconomic downturns have no consistent effect on violent crime. Even during the recent global economic depression following the COVID epidemic, violent crime rates also did not show a systematic increase globally, despite the large negative economic shock and increased unemployment rates (Hoeboer et al., 2024).5
Misc studies
To evaluate the economic effect on crime, Stam et al. (2024) used a different approach. They looked at crime over the welfare payment cycle, and exploited the timing of payment. If crime rates are higher later into the welfare cycle (when resources have been spent), it is evidence in favor of the poverty-as-cause hypothesis. This was found to be the case for financially motivated crimes (e.g., theft), but it was not the case for other crimes (including violent crimes). In fact, there the pattern was slightly in the opposite direction. Perhaps violent crime increases slightly due to them spending the money on alcohol or illicit substances. In conclusion, this line of research does not support the hypothesis that lack of money causes violent crime.
Another attempt at identifying economic variation that is independent from important confounders is with lottery studies. The best studies of this kind don’t just compare lottery winners to non-winners, as participation may itself be selective, but instead they exploit the randomness of lottery prize sizes. Cesarini et al. (2023), the best study of its kind, found no significant effect of lottery wealth on criminal behavior. The confidence interval was such that effects larger than one fifth of the cross-sectional association between wealth and crime could be ruled out.
Conclusion
In this piece I have scrutinized the association between poverty and violent crime. I have documented that this association is greatly influenced by reverse causality and selection. That is, (a) committing crime negatively harms personal and local economic conditions; and (b) there are shared traits — cognitive ability, mental disorders, and many others — which influence both economic prospects and the likelihood of committing violent crime, and leading them to cluster together.
Because the association is strongly affected by reverse causality and selection, its existence cannot be taken as evidence for a causal effect of poverty on crime. To assess the causal effect of poverty on violence, we need to consider more causally informative research designs.
The most causally informative studies consistently establish that the effect of poverty on violent crime is minimal. Sibling studies, within-person studies, income-transfer and jobs program studies, macroeconomic evidence, as well as lottery and other miscellaneous studies together paint a clear picture: poverty has little systematic effect on violent crime. While no single study by itself is conclusive, and any one study can be critiqued on plausibly sounding grounds, the convergence of results is difficult to deny.
In conclusion, a correlation between poverty and violent crime exists, but its existence can largely or perhaps entirely be explained by selection and reverse causality; and not to the effects of poverty itself.
Footnotes
1
After all, criminals tend to be different from criminals in numerous ways — which is elaborated upon in the section about selection.
2
There are two common types of criticisms of these sibling studies. The first criticism is that these results cannot speak to the American context, given that they are based on data from Scandinavia — a place with a better social safety net. I don’t find this response particularly convincing for two reasons. First, as replicated in the cited sibling studies, there is a strong crude association between poverty and violent crime in Scandinavia, just as there is in the United States; and the exact same causal hypothesis is routinely proposed to explain the SES-crime correlation in Scandinavia. It is only when we use research designs more appropriate for estimating the causal effect that it goes away, even in Scandinavia. Second, people tend to exaggerate the differences between the United States and Scandinavia. The differences are a matter of moderate degree, they are not qualitatively distinct societies. If a causal effect exists in the United States, you'd expect it to exist in Scandinavia as well, perhaps just a bit smaller there. Given that the Finnish study gave a tightly estimated null effect, an effect a bit larger would still be close to zero.The second criticism leveled at these sibling studies is the possible concern that the between-sibling measures do not truly capture stable differences in rearing conditions between the siblings; or that effects downstream of parental SES are adjusted away. For the latter, consider if parental income affects neighborhood quality, and you compare siblings who live in the same neighborhood, you’re controlling away a component of income’s effect. This is, in my opinion, a more serious concern. However, the fact that the studies shows that the non-effect is robust to many socioeconomic indicators, and that it's true for both neighborhood and familial conditions suggests to me that this probably also cannot explain the lack of an effect. Still I regard this possibility more seriously than the criticism that this finding is particular to Scandinavia.
3
For example, as illustration, suppose a person becomes an alcoholic at some point in their lifetime. That could plausibly cause them to become both downwardly economically mobile, ending up in a more deprived neighborhood, and become more violent.
4
See Table 3.5.9 in the 2nd edition of the Handbook of Crime Correlates, or Table 3.6.8 in the 1st edition.
5
Unlike most countries, the United States experienced a large increase in homicide rates in 2020. However, a careful analysis shows that this homicide spike was not primarily due to COVID. When homicide counts are tallied in fine-grained time intervals, CDC data shows that the large sudden shock in homicides happened right after the killing of George Floyd, instead of gradually increasing over course of the pandemic. Immediately following the killing, there were many protests and even riots. There was also significant de-policing over an extended period, which contributed to the homicide increase. See also Kim (2023) and Campbell (2023).
Comments