Jailbirds of a Feather Flock Together
From DNA to DOJ, Cremieux Recueil traces the genetic pathways of crime, explaining why poverty is not a good causal explanation.
Written by Cremieux Recueil.
The Most Serene Republic of Venice lasted for 1,100 years before it was divided between the Austrians and the French in 1797. It would ultimately become incorporated into Italy during the Third Italian War of Independence almost seventy years later. The life of the Serene Republic is a storied one. Many of Europe’s great explorers and artists would claim it as their home. Some accounts call it the birthplace of capitalism; others cite it as an example of the follies of protectionism. Regardless of which are true, no one seriously doubts this small maritime republic’s contribution to European and, indeed, world history.
Much of that history still has relevance today.
One of the most notable features of the Serene Republic is that it kept excellent records. Thanks to institutions developed by and for seafaring merchants, the republic’s record-keeping was among the best in the premodern world. One particular set of records is exceptionally worthy of attention: Venetian accounts of crime.
In the modern world we live in, crime and poverty go hand-in-hand. In the Serene Republic, that was not the case. In fact, the relationship was reversed. As recorded by historian Guido Ruggiero in Violence in Early Renaissance Venice, the people most likely to be involved in speech offenses, assaults, rapes, and, well, all crimes other than murder, were noblemen.
Despite comprising only 4% of the population of early Renaissance Venice, nobles were responsible for nearly a quarter of the recorded assaults! In the same paper from which those statistics were derived, sociological criminologist Manuel Eisner noted that Venice wasn’t exceptional in this regard:
In thirteenth-century Bologna, 10 percent of 521 banishment cases for major crimes were urban magistrates and nobles… In fourteenth-century Lyon, Gonthier (1993) observes a recurrent involvement of nobles in violent behavior, including the organization of gangs to revenge failures to comply with their interests. Also, tax returns of offenders in fifteenth-century Constance reveal that wealthy groups were at least as likely to engage in violent offending as the poor (Schuster, 2000, p. 137). Hanawalt (1979, p. 131) found that members of the oligarchy in a small fourteenth-century English rural area committed about one-third of all homicides recorded in the Gaol Delivery Rolls (Hanawalt 1979, p. 131).
Throughout much of European history, noblemen were at least as likely as the poor to be involved in criminal offending because of their tendency to get into “conflicts over rights and goods, which, in the absence of reliable state control, often escalated into violent conflicts” (Eisner, 2003, p. 117).
Returning to Venice, the above-chart seems to suggest that “Important People” —from among the Cittadini, the broad middle-class of Venetian society—were wealthier than nobles, given their lower rates of criminality. But the data we have suggests otherwise. Using Venetian probates, Isabella Cecchini quantified the wealth and the different types of possessions held by members of different classes between 1510 and 1615.1
The noblemen, the members of the patriciate, were clearly wealthier than the members of other classes. Yet this conceals something important: Venetian nobles were less concerned with literacy or the arts. One can see this by looking at the proportions of probates containing books:
Or, using an alternative marker of high culture, instruments:
Despite their far greater wealth, noblemen were simply less interested in intellectual pursuits like reading and playing instruments. They preferred foreign goods and other status symbols that served to impress, rather than edify. Based on the historical record of their demographic crisis after the plagues of 1575 and 1630 and the Cretan (1645-1669) and Morean Wars (1684-1699), they also had a penchant for violence outside of Venice.
However, we know noble violence was something that did not last (aristocratic families today are not known for their criminality). Like the nobility itself, violent behavior declined with modernity. Returning to Eisner, he noted:
[Some] evidence suggests that upper classes in northern Europe may have become more pacified and less prone to physical aggression from the sixteenth century onward. Spierenburg (1998), for example, argues that homicide rates declined in Amsterdam after 1620 because wealthy, churchgoing citizens renounced violence, while lower-class violence in the form of knife fighting remained undiminished. Similarly, Sharp (1984, p. 95) assumes a gradual decline in the involvement of the upper class in criminal violence in the century after 1550.
He continued:
In the south of Europe, the retreat of the nobility from aggressive behavior appears to have occurred later. In the French Auvergne, a mountainous and very poor area, the decisive shift occurred between the beginning and the end of the eighteenth century, when the upper classes increasingly withdrew from violent behavior. This transformation was paralleled by the increasing acceptance, among the nobility, of merit and competence as core social values, to the detriment of honor and the military ethic (Cameron 1981, p. 202). Further south, late eighteenth-century Sardinia offers striking evidence (Doneddu 1991). Ridden by chronic banditry as well as the vendetta, Sardinia had an overall homicide rate of thirty-five to forty per 100,000 in the years 1767-89. During that period, members of the nobility were recorded for committing fifty-one homicides. Since the island’s nobility counted some 6,000 members at that time, this puts their homicide rate at thirty-seven per 100,000, the same as the approximate rate in the total population.
Most accounts suggest it was a change in values that caused the nobility’s shift away from violence. But what if the violence itself was what caused the change in values? In other words, noble violence led to noble non-violence. How? The mechanism is simple: violent nobles died violently, leaving behind relatively more non-violent nobles.
We have some good evidence for this because nobles frequently died on the same exact days. And they did so because they died in combat with one another. In a brilliant paper by Neil Cummins, we’re presented with this table:
Notice how many nobles died in combat? Notice how many died in combat relative to total noble deaths each year? In Cummins’ data, we see fewer of these mass death events over time, and subsequently, noble lifespans increased. They did not increase solely because of a decline in violence, nor solely in periods with notable economic development (although in such periods, they did increase). As Cummins documented, much of the increase preceded economic development.
The decline in elite violence2 is particularly remarkable when you consider that it began right at the top of the social hierarchy: prior to the middle of the eighteenth century, royals had shorter lifespans than commoners.
Subsequent changes in violent behavior coincided with other elite changes, like rising numeracy levels, but they also coincided with high execution rates and extraordinarily high homicide rates (which were trending down for the rest of society too).
Premodern elite violence has substantial contemporary relevance: the fact that for large swathes of history the most privileged members of society were disproportionately responsible for violence (in all its varieties—petty, familial, and martial) suggests that criminality is not a product of deprivation.3
Deprivation is certainly the dominant explanation for why people commit crimes today, and yet in the not-so-distant past, it was the least deprived who committed the most crimes. One could attribute the reversal to all sorts of things, but such an argument must ultimately concede that a nontrivial part of the relationship between deprivation and crime in the present day is due to selection (rather than the effect of deprivation itself). This fact is inescapable, but we needn’t take it for granted.
To determine whether deprivation—or poverty, low socioeconomic status or some other term for the same class of thing—impacts criminality, we need a way to overcome selection into deprivation. By selection I mean the tendency for individuals with a predisposition for crime to be found in deprived environments.
Take schizophrenia, for example. Sufferers’ labor market outcomes are exceptionally bad. They are also considerably more likely than their siblings to commit violent crimes. Schizophrenia is not caused by neighborhood deprivation, but it can lead to neighborhood deprivation being associated with violence insofar as it predisposes people to be both violent and poor.
Let’s consider neighborhood deprivation in more detail. Siblings can be exposed to different neighborhoods growing up. Families might move from a worse place to a better one, or from a better place to a worse one, and they might do so multiple times throughout their children’s childhoods. If the siblings aren’t twins, they’ll spend different amounts of time living in those places, and we can exploit the differences in exposures to assess the impacts of things like deprivation on subsequent criminality. Sariaslan et al. (2013) and 2014, as well as Sivertsson, Carlsson & Hoherz (2021) did exactly that. Here’s what they found:
When looking between families (left-hand circles), children exposed to paternal conviction, neighborhood deprivation, or low income are more likely to become violent criminals. Making a few statistical adjustments (middle circles) reduces the strength of these relationships, but residual confounding remains. When looking within families (right-hand circles), siblings exposed to paternal conviction, neighborhood deprivation, or low income are not more likely to become violent criminals; all the associations disappear.4
There are other clever designs out there, such as within-person models. These involve following the same individuals over time to see whether their outcomes change when they move into or out of different environments, such as deprived neighbourhoods. Airaksinen et al. (2021) carried out one such study:
When we compare individuals living in more versus less deprived neighborhoods (red line), we see that those living in more deprived neighbourhoods are much more likely to commit violent crimes. But when we compare the same individuals over time (blue line), we see that they don’t become more likely to commit violent crimes when they move into more deprived neighborhoods. There’s no effect of neighborhood deprivation once we’ve accounted for this selection and allowed time to pass. (Similar results were found for property crimes.)
All sorts of phenomena have been investigated using within-family and within-person control models. With Finnish data, Aaltonen & Mikkonen examined the impact of childhood exposures to a parent being convicted, while Finlay, Mueller-Smith & Street5 looked at the effect of moving to a county with a high level of exposure to the criminal justice system in U.S. data. Their sibling-control results are plotted next to one another:
The majority of violent crimes involve recidivism and the majority of those sentenced to state prisons6 are recidivists. With that in mind, it’s natural to wonder whether ex-criminals might be more affected by their socioeconomic circumstances. Sariaslan et al. (2017) assessed whether personal factors like disposable income, welfare benefits receipt, or residential relocation predicted recidivism in within-person models. They also assessed whether local factors like neighborhood income levels, neighborhood welfare usage, and neighborhood crime levels predicted recidivism. In each case, the answer was the same:
There are many other causally informative studies like these. A general conclusion one can draw is that associations between deprivation and crime are either wholly or substantially attributable to selection, rather than to deprivation itself.
Other clever designs have been used to study questions like whether serving a prison sentence has a “criminogenic” effect. The two most common ways to study that question are: judge-instrumental variables and sentencing grid regression discontinuity. The former leverages the purportedly stable decision-making tendencies of judges as instruments; the latter exploits mandatory sentencing guidelines as discontinuities in the likelihood of going to prison.
A recent review of such studies concluded that “the experience of post-conviction imprisonment has little impact on the probability of recidivism” (though there were important qualifiers for the handful of studies that had significant results). In fact, this result is even more fragile than the review suggests. When the results are subjected to a random effects meta-analysis with study-clustered standard errors, the result is just null.7
What, generally, can we say about why people commit violent crime? I believe at least three broad conclusions are justified:
The material status of the individual or family is an insufficient explanation. Neighborhood deprivation, exposure to crime and other such factors are incapable of explaining much of the variation in violent offending.
The factors that do seem to matter are, in many cases, not yet open to intervention. Before the invention of chlorpromazine, this was true for psychosis, and so asylums were needed. After antipsychotics were developed, they were no longer required. Similar technologies may emerge for other “root” causes of crime, such as mental illness.
The impact of expanding law enforcement and criminal justice is highly situational. For example, in places where all but the most marginal offenders have already been incarcerated, jailing more people is unlikely to do much about crime.
If I were to add a fourth conclusion, it would be this: despite the well-documented fact that selection is happening all the time, we’ll continue to be served up explanations that don’t cross into causal territory and therefore don’t teach us anything at all.
Cremieux Recueil writes about genetics, 'metrics, and demographics. You should follow him on Twitter for the best data-driven threads around.
Consider supporting Aporia with a $6.99 monthly subscription and following us on Twitter.
It is theoretically possible to reconstruct these sorts of records for later years too, but that’s beyond the scope of this article.
Which was even observed at universities.
The lack of a robust relationship between inequality and violence makes the case for a relationship due to relative deprivation untenable too.
Sivertsson, Carlsson & Hoherz' non-sibling control results are significant, but their CIs appear smaller than the dots.
This result is more ambiguous than Aaltonen & Mikkonen’s since their within-family estimates were not significantly different from the between-family ones. In addition, the study did not distinguish half-siblings from full-siblings, so while it is a good study based on high-quality data, some caution is advised. A Cox proportional-hazards model would have suited this dataset well and would be well worth fitting.
A group that comprises the majority of those sentenced to prison in the U.S. in general.
The literature cited in the review was consistent with a minuscule effect of 0.0094, with a 95% confidence interval ranging from -0.0196 to 0.0385. But at least there wasn’t any publication bias! The suggested moderators didn’t significantly alter this conclusion.