I’m proud of the method I came up with for this research. The quasi-experiment allows for causal inferences in a field typically bogged down in the swamp of meaningless correlational studies.
Objective: Test for causal effects of monetary possession on crime.
Method: Quasi-experiment comparing days with high cash to days with low cash.
Results: Drug crimes are 47% more likely in the high-cash condition than in in the low-cash condition. Assaults are 21% more likely in the low-cash condition than in the high-cash condition.
Conclusions: The results provide new evidence on the old question of causality in the relationship between economic conditions and crime. Furthermore, the results show that the same financial cause can drive different types of crime in opposite directions.
For nearly as long as criminology has existed as a field of study, theorists have posited relationships between money and crime. In 1893, Carroll D. Wright claimed that crime arises from impoverished economic conditions, especially when the poverty is unequal across society. He pointed out that the majority of prison inmates had no training in a trade, or were not employed at the time of their arrest (Wright 1893: 109). This was the beginning of an enduring tradition of sociologists and criminologists cataloging associations between financial conditions and crime.
In 1911, Cesare Lombroso put forth a different explanation of Wright’s finding on unemployment among prisoners. Lombroso saw the correlation as a result of the criminal’s aversion to work, rather than as a criminogenic effect of poverty (Bonger 1916: 91). Neither Wright’s nor Lombroso’s interpretation could be ruled out by Wright’s data. Since Wright could only observe a simple correlation between two phenomena, the existence and direction of causality could not be validly inferred.
Over the past century, the randomized controlled trial (RCT) has become the standard tool for escaping the problem of spurious correlation. But unfortunately, insurmountable logistical and ethical barriers prevent any substantial random assignment of financial conditions. Therefore our understanding of causality in the relationships between financial condition and criminality continues to be of low epistemological quality. The goal of this study is to analyze one environment in which financial condition naturally varies quasi-randomly, producing a quasi RCT.
This study examines variations in cash over time. Taking somewhat similar approaches, other researchers have found associations between cyclical financial changes and certain types of crime.
Bushway, Cook, and Phillips (2012) analyzed the 13 business cycles (macroeconomic booms and recessions) in the US since 1933 to provide evidence on the question of whether recessions cause crime. They found that recessions are associated with an uptick in burglary and robbery, and a reduction in theft of motor vehicles (Bushway et al 2012: 442).
Though their study is correlational, Bushway et al interpret their results as causal, reasonably rejecting the possibility that small changes in robbery and burglary cause macroeconomic recessions. However, their reason for dismissing the possibility of a third variable is controversial. They claim that since aggregate crime rates do not return to a common mean or equilibrium position, while business cycles do, crime and business cycles are not likely to be driven by the same external force. Bushway et al find it “difficult to imagine” a third process that is causally responsible for both the ‘stationary process’ of business cycles and the ‘random walk’ process of crime (Bushway et al 2012: 442). However, stationary processes can cause random walk processes. Bushway et al make this claim in the case of business cycles causing crime. Stationary processes can also cause other stationary processes. And since one process can be the cause of many other processes simultaneously, it is not clear why a third variable could not cause both business cycles and crime.
Bushway et al see their study of business cycles as a natural experiment. But the third variable problem persists because the experiment lacks any element of randomization in assigning recessions to occur or not occur.
Fritz Foley (2011) got closer to a true natural experiment on financial conditions. He conducted an analysis of daily reported incidents of major crimes in twelve US cities, finding an increase in crime over the course of monthly welfare payment cycles. Temporal patterns in some theft related crimes were observed in jurisdictions in which disbursements are focused at the beginning of the month, and not in jurisdictions in which disbursements are relatively more staggered. The welfare cycle was associated with an increase in crimes which have a clear financial benefit (burglary, larceny/theft, motor vehicle theft, and robbery), but not for other crimes. Foley concluded that these findings indicate that welfare beneficiaries consume welfare-related income quickly and then attempt to supplement it with criminal income (Foley 2011: 111).
The epistemological quality of Foley’s welfare study is relatively high compared to other studies on money and crime. The increased theft can not cause welfare payments, and unlike business cycles, welfare payments have a well understood cause, which partially eliminates the third variable problem. However, it is still possible that the true third variable is the month, which causes welfare payments, and could conceivably cause crime by some intermediary process other than welfare payments. For example, some other monthly event such as rent being due could cause the increase in crime. Foley deals with this by using other cities as a comparison group. But that solution comes with the assumption that other cities are equal in relevant unmeasurable factors, which seems likely, but is unknown. The use of other cities and its required assumption could be avoided if the timing of the payments was effectively randomized.
In order to exploit a naturally occurring, quasi-randomly timed variation in cash, this study uses crime data from Fayetteville, NC, which borders Fort Bragg, the largest military base in the world by population. There is no other urban area of more than 15,000 people within 25 miles of Fort Bragg.
The raw crime data consist of 82,413 arrest incidents from the Fayetteville, NC Police Department (City of Fayetteville 2017). The data cover arrests from the end of 2007 to the beginning of 2017, and is publicly available from the City of Fayetteville website. Each entry includes date of arrest and type of charge. All warrant-based arrest incidents are excluded from analysis because they do not correspond temporally with crimes committed.
The arrest data have been aggregated by date and charge type to generate daily counts of arrests by type of crime. These daily counts have been merged with a vector indicating whether or not each day was a military payday, and a second vector indicating an integer number of days since the last payday (ranging from 0 to 18).
There are four important details about the military that create quasi-random variation in cash, which can be used in a natural experiment.
- All military personnel are paid on the same days.
- Fort Bragg, NC, concentrates approximately 65,000 soldiers in a locality.
- The military disburses pay on the 1st and the 15th of each month, unless those dates fall on Saturdays or Sundays, in which case the disbursement occurs on the Friday preceding the 1st or 15th.
- During the workweek, most soldiers are busy, and stay on base. On the weekends, they spend more time and money in the city that borders the base.
Military pay follows a modified semimonthly cycle, while the exposure of the city to the soldiers follows a weekly cycle. Since these cycles have different frequencies, they do not synchronize with each other. And since months are not whole number multiples of weeks, the two cycles do not resonate in a neat pattern. Therefore there are some days when the two cycles happen to align, and some days when they happen to be in anti-alignment. This creates two distinct conditions in the city, which arise merely as artifacts of dissonant timing in the cycles. Figure 1 is a conceptual illustration of these cycles with irregular alignment and anti-alignment.
Figure 1. Conceptual illustration of Dissonant Cycles. When the two cycles align, one distinct condition obtains; when the two cycles are in anti-alignment, another distinct condition obtains.
In concrete terms, alignment of the cycles is when a weekend happens right after a payday. Anti-alignment is when a weekend happens right before a payday. This creates a pool of high-cash Fridays, Saturdays, and Sundays, and a comparison pool of low-cash Fridays, Saturdays, and Sundays. The high-cash days and low-cash days are from the same time of week and same time of month, but some of them happen to be just before payday, and some happen to be just after payday. As an example, Figure 2 shows Fridays clustered by days since pay. The high-cash group and the low-cash group are circled.
Figure 2. Fridays Plotted by Days Since Pay and Number of Arrests. Each blue circle on the scatter plot represents one Friday. The position on the x-axis indicates the number of days since last payday. The position on the y-axis represents the daily total count of arrests. Taking the days on the far left and far right as two distinct conditions allows for a comparison of the arrest counts on many Fridays with high-cash to the arrest counts on many Fridays with low-cash.
The formal rules used for categorical assignment of each day to the high-cash (treatment) group or the low-cash (control) group are as follows:
The high-cash group is defined as days that are either as close to the last payday as possible, or one day later than that. The low-cash group is defined as days that are either as far from last payday as possible, or one day earlier than that. For example, high-cash Fridays are those Fridays that are paydays, or are days directly following paydays. Since there are no paydays on Saturdays or Sundays, high-cash Sundays are those Sundays that are either two or three days after the last payday. Figure 3 visualizes the conditions of high-cash and low-cash for Fridays.
The high-cash and low-cash groups were compared using Quasi-Poisson regression of arrest counts on a binary variable that indicates whether each day is high-cash or low-cash. Separate tests were carried out for various categories of arrests: all arrests, theft arrests, DUI arrests, drug arrests, and assault arrests (Table 1).
Results and discussion
|Charge Type||‘Low-Cash’ Group Mean||‘High-Cash’ Group Mean||t-statistic||% Change|
n = 540 days *p < 0.05
Table 1. Drug and Assault Arrests Differ Significantly Between Weekend Cash Conditions. This table shows results from Quasi-Poisson tests for significant differences between average arrests on low-cash days and average arrests on high-cash days. The ‘%Change’ column shows the change in crime from low-cash to high-cash. The ‘Drug’ and ‘Assault’ categories exhibit significant changes.
The mean daily counts of drug arrests and assault arrests differ significantly between the ‘High-Cash’ weekends (Fri, Sat, Sun) and the ‘Low-Cash’ weekends (Fri, Sat, Sun). These significant differences are visualized in Figure 3. Drug arrests are 47% more likely in the high-cash condition than in in the low-cash condition. Assaults are 21% more likely in the low-cash condition than in the high-cash condition.
Figure 3: Mean Drug and Assault Arrests Per Day in High-Cash and Low-Cash conditions.
There are no significant differences in crime between high-cash weekdays and low-cash weekdays. When tests are performed comparing high-cash workweeks (Mon, Tue, Wed, Thu) to low-cash workweeks (Mon, Tue, Wed, Thu), none of the differences are significant. That the effects only occur on weekends, when the city is exposed to increased soldier presence, validates that the effects are driven by the soldier’s money.
The results show that the same financial cause can drive different types of crime in opposite directions. High cash caused an increase in drug arrests, while causing a decrease in assault arrests.
There is potential for direct application of these results to local policing practices for some departments. For example, if a police department near a military base intends to carry out a drug sting operation, they may benefit from timing their operation on a high-cash weekend.
This study has some limitations. It only examines cash variation in one demographic (military), in one city (Fayetteville), so external validity is limited. Though the demographic is limited, it should be noted that the military is composed of primarily young males, which is a very important demographic in crime.
It is unknown specifically how many soldiers leave base each day, and at what rate they spend their cash, so it is not possible to produce valid elasticity values for cash and crime from the current data. It is also unknown to what extent cash possession influences the rate of soldiers leaving base.
Even though external validity is limited, this study can be taken as corroboration with broader correlational studies. When taken together, the internal weaknesses of large correlational studies can be mitigated by studies like this one, while the external weakness of this study can be mitigated by the large correlational studies. Both are needed to build a strong theory of the relationship between money and crime.
Bonger, W. (1916), Criminality and Economic Conditions, (Boston: Little, Brown, and Company)
Bushway, S., Cook, P. J. and Phillips, M. (2012). The Overall Effect of the Business Cycle on Crime. German Economics Review, 13, 436–446
City of Fayetteville. “Incident Data-Arrest (Cumulative)” 2017. City of Fayetteville Open Data. http://data.fayettevillenc.gov/datasets/b661fd22e9a44a44a2528120eb54e9c9_0
Foley, F. (2011). Welfare Payments and Crime. The Review of Economics and Statistics, 93, 97-112
Wright, C. (1893). The Relation of Economic Conditions to the Causes of Crime. Annals of the American Academy of Political and Social Science, 3, 96-116