I also show results for the intensive margin using both the number of individual cases filed and the total number of charges in a given year using the Inverse Hyperbolic Sine IHS transformation and Poisson models. Criminal behavior is also separated into the four largest crime categories: property, driving, drug and other. F rackingCountyXP ostLeaseht and F rackingCountyXP ostP roductionht are indicator variables equal to 1 for households in fracking counties during the leasing period, to , and during the production period, to , respectively.
I check this assumption in several ways. First, I provide visual evidence that treated and control counties are tracking prior to any treatment. Relatedly, I formally test for pre-divergence using the above regression model with an indicator for the treated group one year before treatment. Additionally, I allow counties to trend differently over time by including county-specific linear time trends. I also include interactions between pre-treatment controls and year effects. In doing this, I allow for counties with different levels of observable characteristics, such as per capita income, to respond differentially to year-to-year shocks.
In all models, robust standard errors are clustered at the county level, allowing errors to be correlated within a county over time. I also report permutation-based inference for the primary specification when considering all crime, similar in spirit to Abadie, Diamond and Hainmueller for inference when using the synthetic control method. To do this, I randomly assign treatment to 17 counties and compare the estimated coefficient to placebo estimates to compute two-sided p-values.
Adjusted FDR Q-values correct for the increased likelihood of rejecting the null hypothesis when making multiple comparisons, and are interpreted similar to p-values. Given that some counties experience larger shocks than others, detected effects could be driven solely by counties with more extreme local shocks. However, it is beneficial to know if smaller economic shocks also affect criminal behavior.
Therefore, I also consider heterogeneous effects by the amount of fracking activity experienced by a county. Specifically, I estimate the treatment effect for the four major oil and gas producing counties as defined by the Labor Market Information Center, namely Dunn, McKenzie, Mountrail, and Williams, separate from the effect in the thirteen minor fracking counties.
Finally, I examine the potentially differential effects of fracking on leaseholders and non-leaseholders. As previously discussed, some households receive large sums of money in the form of royalty payments while others do not. This creates the potential for increased crime due to changes in both income inequality and criminal opportunities. I consider leaseholders and non-leaseholders within fracking counties as separate treated groups, comparing each of them to residents in non-fracking counties.
To the extent that signing or not signing a lease and receiving royalty payments is also a form of treatment, this strategy separates the effect on the two groups living in fracking areas.
Formally, I estimate the following regression model:. They capture both the effect of job opportunities and the additional income received by leaseholders in the form of royalty payments. Alternatively, they capture the effect of higher wages and job opportunities, along with any potential effect of not receiving royalty payments for non-leaseholders. As in the previous models, equation 2 is estimated using two mutually exclusive periods: leasing starting in and production beginning in Notably, leaseholders receive a small signing bonus upfront, with royalty payments closely following production.
I begin by estimating the overall effect of local economic shocks on crimes committed by residents. As noted above, I consider only the population of residents prior to the fracking boom in North Dakota.
In doing so, I am able to exclude all crimes committed in the county by new workers who migrated to the relatively stronger labor markets. In this way, I can distinguish the effect of the economic shock from the impact of the changing demographics on overall crime rates.
First, I graph the estimated divergence over time in crimes committed by residents in fracking and non-fracking counties, relative to the difference between the two sets of counties in and Figure 4 plots the dynamic difference-in-differences estimates for all crimes, controlling for household and year fixed effects.
Importantly, there is no evidence of divergence prior to the start of the fracking boom in This supports the identifying assumption that absent hydraulic fracturing activities, residents in fracking counties would have experienced similar changes in criminal behavior as residents not in fracking counties. Additionally, the figure indicates that the probability of being charged with a crime falls in fracking counties when leasing starts, then rises some during the production process.
This suggests economic opportunity is reducing crime, but the effect seems to be offset at least somewhat by the indirect effects that accompany oil production.
For example, the production period also includes interactions with new workers and increases in disposable income from royalty payments or high-paying drilling jobs. I report the average treatment effects for each period in Table 2. Figure 4. Dynamic difference-in-difference estimates of the effect of fracking on crime Notes: Dynamic difference-in-differences estimates from equation 1.
Table 2. Standard errors are in parentheses and clustered at the county level. County controls include per capita income, total jobs, population, total officers, and production in Starting with the leasing period, Column 1 indicates an initial drop of 0.
In Column 2, I formally test for pre-divergence and find no evidence of it, with the coefficient on the lead indicator being close to zero, In Column 3, I allow for county-specific linear trends. This allows for both observable and unobservable county characteristics to change linearly over time.
If results are driven by fracking counties being on a different path than non-fracking areas, then adding a county-specific linear trend should absorb the treatment effect.
However, results indicate the coefficient increases slightly to Finally, counties with different baseline populations, total jobs, police officers, per capita income, and production may respond differentially to year-to-year shocks. For example, if fracking counties also tend to be smaller in population then detected effects could be a result of small counties differentially responding to yearly shocks.
In Column 4, I allow these baseline characteristics, observed in , to differentially affect criminal behavior each year. The magnitude remains stable at Overall, estimates in Table 2 are consistent with Figure 4 in showing that while there is a significant drop in criminality initially, the drop is somewhat diminished in the production period. Column 1 indicates a 0. The permutation-based p-value is marginally significant at Moving across Columns 2 through 4, coefficients remain negative ranging from It appears as though the reduction in criminal behavior from the boost in economic activity may be at least somewhat offset by additional effects on criminal behavior during the production period.
This could be due to the effects of in-migration, such as peer effects and increased social interaction Glaeser, Sacerdote and Scheinkman, ; Ludwig and Kling, ; Bernasco, de Graaff, Rouwendal and Steenbeek, , or to an increase in the number of bars and illegal markets.
Given that estimates are at the household level, a potential concern is that effects are being driven by changes in household composition rather than changes in criminal behavior.
For example, the results could be driven by composition if young men, a more crime-prone demographic, were more likely to move out during the leasing period and then move back during the height of the production period in fracking counties. I examine this concern by restricting the sample to crime committed by household members that are older than 25 years at the start of my period in January The results for the older, more stable sample are shown in Figure A.
They are the same as the results in Figure 4 and Table 2, respectively, suggesting that it is changes in criminal behavior rather than household composition that are driving this pattern of results.
To better understand the type of crime affected by local economic shocks, I present treatment effects separately for financial-related crimes e. DUIs, reckless driving , drug-related crimes e.
The dynamic difference-in-differences estimates, controlling for household and year fixed effects, are plotted for each crime type in Figure 5. Notably, the figures show that residents in fracking and non-fracking counties do not diverge prior to the fracking boom in these types of crime. However, residents exposed to fracking activities change their criminal behavior relative to residents in non-fracking counties in response to the economic shock.
Results show relatively large reductions in driving, drug, and other offenses in the leasing period. However, this reduction is diminished once production starts. Figure 5. Dynamic difference-in-difference estimates of the effect of fracking on crime, by crime type Notes: Dynamic difference-in-differences estimates from equation 1 with household and year fixed effects.
Similar to Table 2, Table 3 reports average treatment effects for each period, with panels for each crime type and adjusted FDR Q-values for statistical inference. Panel A indicates a Similarly, estimates are negative for driving-related cases during the leasing Panel C shows a decrease in drug cases filed of Finally, all other crimes have a similar negative effect during the leasing period ranging from All coefficients are fairly robust to the inclusion of controls and a lead term.
Table 3. Because I consider four types of crime, I also report statistical significance of these estimates using the Adjusted False Discovery Rate Q-values proposed in Anderson These values correct for the increased chance of rejecting the null hypothesis when making multiple comparisons for two treatments across four groups eight categories. The negative effects on driving and property cases are generally not statistically significant once corrected for multiple comparisons.
However, the effect on drug cases filed during the leasing period is sufficiently large across all specifications in Column 1 through 4 as to not have occurred by chance with Q-values of 0. There is no statistical effect on drug cases during the production period. For comparison, I also report the effect of hydraulic fracturing activities on aggregate changes in cases filed per persons. Figure A. Again, counties do not diverge prior to fracking activities.
However, estimates indicate increases in total cases filed, as well as drug, driving, assault, and all other cases during the fracking periods, specifically during production, which is consistent with prior literature. Finally, I test whether the migrants entering the fracking counties were committing crimes at higher rates than the native population.
This enables me to speak directly to a question of interest in the immigration literature of whether those moving into an area are more criminogenic in general. I measure the propensity to commit crime for a subset of those moving into the county.
Specifically, I calculate the crime rate using the number of cases filed with an out-of-state address over the number of migrant tax exemptions filed in the county. I do the same for all crime committed by those with an address in North Dakota and the number of non-migrant tax exemptions in the county, fixing the total as of First, the crime rate for people moving into the county only considers crime from out-of-state individuals, even though there is some in-migration to fracking counties from other areas in North Dakota.
This also means that any additional crimes committed by those that move into the county from within the state are being considered as crimes committed by non-migrants for this exercise.
Second, migrant and non-migrant tax exemptions are based on whether there is a change in filing county and state. To be conservative, the denominator for the out-of-state crime rate is the total of all inflows from through Similarly, I fix the total number of non-migrants in each county at the total in for each year as migrants that move into the area will be counted as non-migrants in their second year residing in the county.
While I observe reductions in all crime types, results are primarily driven by drug-related crimes. This is in contrast to the county-level results, suggesting that compositional changes play an important role in the criminal response to economic conditions.
Put differently, this suggests that the aggregate increases seen are due largely to additional crimes committed by those who move into the area. In contrast, the effect of the economic opportunity itself seems to have a negative effect on crime. Results thus far have treated all counties on the shale play as receiving the same economic shock.
However, some counties, particularly the four major oil and gas producing counties, experience much larger economic shocks than others. The oil production in each of these four counties was greater than the amount produced in the other 13 counties combined over this time period.
To estimate the differential effect by treatment intensity, I report estimates from equation 1 separately for major and minor fracking counties in Table 4. For the leasing period, to , estimates in Column 1 indicate a 0. This represents a The estimated effect is stable to the inclusion of a lead indicator, county specific trends, and allowing for time-shocks that vary with levels of pre-period observables.
Estimates in Columns 2 to 4 range from a 0. All estimates are significant at conventional levels. Table 4. During the production period, estimates for minor fracking counties are similar in magnitude to the leasing period, ranging from a 0. Estimates for major fracking counties are smaller in magnitude during the production period relative to the leasing period As expected, the effect is larger in magnitude for the major fracking counties than the minor fracking counties initially, although coefficients are not statistically different.
Importantly, this demonstrates that the effect is not driven solely by the four large fracking counties, as counties experiencing more modest economic shocks also see a significant reduction in crime.
Additionally, the effect seems to fade more dramatically in the major fracking counties which also experience larger population and income changes during the production period. This is consistent with the interpretation that it is the other consequences of the in-migration, such as peer effects, and income that lead to a diminished reduction in crime for residents.
In addition to the local economic shock, some residents in fracking counties also receive a large positive income shock in the form of oil royalties during the production period. These payments may affect the decision to commit crime both for the leaseholder and the non-leaseholder as payments increase disposable income for illegal activities by leaseholders while increasing the income inequality and criminal opportunities for non-leaseholders.
In Table 5, I estimate the extent to which the fracking activities may differentially affect residents using equation 2. Table 5. Estimates for lease-holders are mostly negative during leasing Estimates for non-lease-holders range from During the production period, estimates range from This suggests that it is the increase in job opportunities that reduces crime, rather than income per se.
Moreover, the effect of job opportunities seems to be stronger than the effect of increased criminal opportunities. In summary, I find that crime decreases during the leasing period in response to improved job opportunities 0. Effects are largest and most consistent for drug-related crimes with a decrease of 0.
Importantly, the effect is not solely driven by the four largest oil producing counties. During the leasing period, there is a 0. The effects diminish more in the major fracking counties, which suggests that other factors related to production contribute to offsetting the effect of improved labor market conditions. Additionally, I find that effects are strongest for non-leaseholders 0.
This is consistent with those not receiving alternative income streams being most sensitive to the job opportunities. One concern in interpreting the results described above is that the differences over time may be due to changes in the number of police officers. Moreover, empirical evidence has shown that crime decreases in response to increased police presence di Tella, ; Machin and Marie, To test for changes in the police force, I estimate the main model at the county level with total police officers as the outcome of interest.
Figure 6a, indicates that the change in the number of police officers was negligible during the leasing period. As a result, changes in police are unlikely to be driving the significant reduction in crime observed during the leasing period.
However, changes in police are potentially part of the treatment during the production period, although this is difficult to disentangle from other factors that changed during that period.
Similarly, reductions in police resources from population increases may lead to fewer reported cases filed Vollaard and Hamed, Figure 6b shows little evidence of changes in the population from to , with large increases during the more labor-intensive production period. Again, population changes are less of a concern during the leasing period, but are likely to be a part of the treatment effects after as previously discussed.
Relatedly, a concern may be that individuals identified as residents may have moved out of the county or, more importantly, the State of North Dakota during the fracking periods. This could be an issue if changes in crime are simply from not observing the criminal behavior of an individual that moved out of the state.
Anecdotally, it seems improbable that residents would disproportionately move out of fracking counties as economic conditions improved. I empirically check for evidence of out-migration using the number of tax exemptions that move out of a county each year. I find no evidence of differential out-migration during the initial leasing period.
I find signs of out-migration only toward the end of the production period when those who had moved into the county begin leaving as shown in Figure 7. Figure 7. Estimates of the effect of fracking on out-migration Notes: Dynamic difference-in-differences estimates from equation 1 with county and year fixed effects.
Outcome is defined as total number of out-migration exemptions. An exemption is classified as a migrant if it is filed in a different county than in the previous year. The exemption would be an out-migrant for the county of filing in the previous year and an in-migrant for the county of filing in the current year. Data on all exemptions is from the Internal Revenue Service. While I am not able to directly test for the mechanism underlying the decrease in crime from improved economic opportunities, I suggest two potential pathways.
First, it is possible that decreases in crime are the result of an incapacitation effect, as individuals become occupied with legal work and thus have less time for criminal activities.
This is similar to the incapacitation effect of school on juvenile crime Jacob and Lefgren, A second explanation is that residents may no longer feel the need to engage in activities related to crime, such as drug use, given their improved economic outlook.
This is consistent with work by Case and Deaton ; and Autor, Dorn and Gordon forthcoming , who document an increasing number of deaths from drugs, alcohol and suicide associated with deteriorating economic conditions. This is also consistent with Becker which predicts individuals are less likely to engage in criminal activity if they have more to lose if apprehended. As a result, a more positive outlook on economic conditions, whether expected or realized, may also lower crime.
Specifically, I exploit the recent boom in hydraulic fracturing activities as a plausibly exogenous shock to local economic conditions. Using detailed administrative data on all criminal cases filed in North Dakota from to , I estimate the effect of increased job opportunities on criminal behavior. An important strength of this study is that by focusing the analysis on all rural residents already living in the area prior to fracking, I can distinguish the effect of improved economic opportunity from the effect of population inflows on aggregate crime.
Results indicate that, consistent with the existing literature, aggregate crime increased in fracking counties relative to non-fracking counties. This was particularly true during the more labor-intensive fracking activities. However, local residents engage in less criminal activity at the start of the boom with a smaller effect in later years. Effects are largest and most robust for drug offenses, and are observed across all counties with fracking activity.
Additionally, I show that effects are most pronounced for residents that do not also receive royalty payments. Taken together, results suggest that residents reduce their criminal activity in response to improved job opportunities, but that other changes from local economic shocks, such as peer composition, seems to offset this effect. This is consistent with economic opportunities reducing crime and highlights the role of compositional changes on the aggregate effects on crime.
Working Paper. CEP Discussion Papers dp This is equivalent to Cyprus, though, is something of an anomaly, in that most of its economic cost is associated with the displacement of its population. In general, violence across the world is getting worse: peace levels deteriorated in 92 countries in , and improved in just 71 countries. However, one bright spot is that overall military spending is falling as a portion of GDP.
According to the Index, 88 countries are spending less and 44 spending more. Average country military expenditure as a percentage of GDP has continued its decade-long decline, with countries spending less. The views expressed in this article are those of the author alone and not the World Economic Forum. Around 48, people were pushed into poverty. This has made ownership unaffordable for many people across the bloc. I accept. Take action on UpLink. Forum in focus. Forum-facilitated guidelines enable more efficient and secure delivery of cash-based humanitarian assistance.
0コメント