Crime Highways: the Effect of Motorway Expansion on Burglary Rates

A high-quality road network promotes trade between regions, but burglars can exploit this connectivity because it reduces the time needed to travel to distant destinations which decreases the time in transit with stolen goods. To study the effect of motorway connections on burglary rates, spatial variation in connectivity to the motorway network and the timing of new connections are exploited using an annual panel of 562 police sub districts in Ireland during 2004-15. On average, connection to the motorway network causes a 10% rise (5 burglaries) in the burglary rate in sub districts. The rise in crime occurs temporarily in the year of connection, as no persistent effect is found on the level of crime. An estimated 12,150 burglaries were caused by motorway expansion. This paper shows for the first time that major road construction affects the spatial distribution of crime. The results highlight the need to account for the negative crime externalities of an unpoliced motorway network. JEL classification: K42, R41


Introduction
Transport infrastructure is an integral component of the production and consumption process because it helps goods move between locations. The transfer of goods is also required in criminal activity, such as the relocation of stolen goods to the market where they are traded for cash. There is a large literature that documents how transport infrastructure construction reduces trade costs and shapes economic outcomes, such as trade flows, population density and household commuting (Atack et al., 2010;Turner, 2011 andMichaels, 2008). However, little is known about the effect of new transport infrastructure on criminal activity. This paper studies the causal effect of connections to the motorway network on burglary rates in the (Republic of) Ireland. Motorways are the highest standard of road in Ireland and analysis from the Irish police service details how specialized burglary gangs are exploiting the motorways connectivity to burgle distant targets while evading police detection 2 . The empirical analysis in this paper uses a spatially detailed dataset that links motorway connections to burglary rates at a fine spatial dimension, consisting of 562 police sub districts and access to motorway junctions during 2004-15.
The combination of the spatial variation in motorway connectivity, the timing of new connections and the inclusion of geographically detailed fixed effects creates exogenous variation in motorway connections. Sub districts are considered connected to the motorway network if there is a motorway junction within a ≤30km's radius of the centroid (centre). All motorways in Ireland, except one, connect to the captical city of Dublin. Therefore, connection to the motorway network intrinsically measures connection to Dublin City. This is important because Dublin is presumably where the majority of the trade in stolen goods takes place. To supplement the empirical analysis, a simple theorectical framework incorprating insights from the traditional Becker (1968) crime model outlines how motorways affect the criminal location choice, by affecting the probability of being caught. This paper finds that new connections to the motorway network causes 5 burglaries in the average sub district, which equates to a 10% rise in the sub district burglary rate. The effect is small but this is expected because sub districts are geographically small units and presumably burglars seeking to evade police attention will disperse the location of their crimes. The rise in burglaries caused by motorway expansion is temporary, occuring in the year of connection only, as no statistically significant effect is found in the suceeding years. It is estimated that 12,150 burglaries were committed because of motorway expansion. The immediate response of burglars to motorway expansions is a curious result. A possible explanation is that the probability of being caught by the police is lowest in areas recently connected to the motorway network, because motorway-related burglaries have not happened yet and so police vigilance has not been initiated. Even if the police are alerted, it will take time for policing action to be organised and implemented. Criminals may exploit their ability to react quicker to insufficient policing, in comparison to that of the police force. The causal relationship with motorway connections is unique to burglary, and no effect is found among other types of crime such as theft and vandalism. Conceptually, this can be explained by the need to quickly transport bulky stolen goods away from the property, either to storage or to the selling market.
The novelty of this paper is to show that major road construction influences the spatial distribution of crime. This paper argues that the spatial distribution of burglaries is influenced by motorway expansion because motorways reduce a burglars escape time. This paper contributes a new economic outcome to the literarture on the economic effects of infrastructure construction, as this is the first paper to causally identify the effect of road construction on crime. There is a large literature that focuses on understanding the way that transportation infrastructure reduces trade costs and how this affects economic activity. The empirical literature provides causal estimates on the effect of roads, railroads and subways on a range of economic outcomes. Donaldson and Hornbeck (2013) use a sample of U.S counties between 1870-90 and find that counties that receive rail access in a year experience an increase in aggregate agricultural land rent. Baum-Snow et al. (2012) find that railroads and ring road developments caused industrial production to move to the suburbs of large cities. The trade literature has generally focused on the transportation of goods, but other studies find transport costs are important in the movement of people. For example, Duranton and Turner (2011) find that driving increases by about 1% for each 1% increase in the stock of roadways. This paper also contributes to the literarture on criminal location choice by showing that burglars take into account the layout major roads when deciding where to commit offences. There previous economic literarture on the location of criminal activity, and criminology literarture on the location of burglaries (Cracolici and Uberti, 2008;Glaeser and Sacerdote, 1999;Fabricant, 1979;Bernasco and Luykx, 2003;Fabricant, 1979;Polvi, Looman, Humphries, & Pease, 1991).
To note, this paper focuses on the importance of the motorway network to burglars, but the results do not show the relative unimportance of the layout of lower level roads, such as local roads. Even when using the motorway, a burglar must interact with lower level roads to travel to destinations (Brantingham and Brantingham, 1981). The key insight is that motorways provide a high-level of connectivity that other road types do not, and this special function is exploited by burglars to allow them to geographically expand their operations to maximise their return. For policymakers, the findings identify that there can be negative crime externalities generated by major road developments. Policymakers should incorporate the externality into any cost-benefit analysis of future road infrastructure developments. The main cost is the need for immediate targeted law enforcement on all new major roads to reduce the externality, such as police checkpoints at motorway entries and exits.
Concerns about potential endogeneity are relaxed using detailed robustness checks. Firstly, a major potential source of endogeneity is the endogenous placement of motorways, as motorways are generally placed in areas of strong economic growth. However, connections to randomly postponed motorway junctions yield no statistically significant effect on burglary rates. Secondly, the uniqueness of the causal relationship between motorway connections and burglary rates over other crime types, suggests that there is no ommited variable bias present. Thirdly, the main result is robust to the inclusion of a control variable measuring local economic activity.
The analysis is extended to show that motorway targeted law enforcement in 2016 was successful in reducing burglaries nationwide. This provides additional weight to the argument that motorways are affecting the location of burglaries. Furthermore, the new policing measures were launched after a period of no motorway growth, suggesting that burglars continue to exploit motorways in the years after connection. It is very likely that burglar's return to previously targeted areas after a period of time, but the revisitation does not appear to be patterned with the year of motorway connection, making it difficult to quantify in this paper. It is for this reason that this paper does not provide an estimation of all motorway-related burglaries and focuses instead on the immediate aftermath of motorway expansion. Due to this, the estimated overall value of stolen items gained from motorwayrelated burglaries cannot be compared to the cost of the motorway. In 2007, the average value of goods stolen from break-ins in Ireland was estimated to be €5,390 ($7,450) 3 . Multiplying this figure with the estimated 12,150 motorway burglaries caused by the initial motorway expansion, €72 million was stolen by burglary gangs. This is a small percentage of the total cost of the motorway (€8billion or $9.8billion), but it is still a significant amount of wealth to be forcibly redistributed from many households and to be held in the hands of a small number of individuals 4 .
An important aspect to the rise in burglaries near new motorways is whether motorway expansion increases the amount of burglary activity in the economy or whether it relocates existing burglaries 5 .
No formal test is conducted, but it is hypothesized that motorways are the artery to attractive crime locations, rather than increasing the stock of burglaries nationwide. The major indication of 3 In 2007, Eircom Phonewatch estimated that the average value of goods stolen from each home during a burglary is €5,930 see "House thefts total €86m in a year" available at https://www.irishtimes.com/news/house-thefts-total-86m-ina-year-1.813645. 4 See "Total cost of motorway links to Dublin €8bn -NRA" available at https://www.irishtimes.com/news/total-cost-ofmotorway-links-to-dublin-8bn-nra-1.667286. 5 Chandra and Thompson (2000) find that highways affect the spatial allocation of economic activity by drawing existing activity towards counties that receive the highway. While Banerjee et al. (2012) find that transportation infrastructure affects the level of output, but has no effect on its growth. reorganisation is that the number of burglaries in Ireland remains relatively steady between years (see figure 2), despite vast motorway growth. Secondly, the rise in burglaries ocurrs temporarily, suggesting that the criminals are commiting the same crimes each year, but in different locations.
Thirdly, it is intuitive that motorways do not induce criminal activity, because by virtue of the criminal tracking motorway expansions, they are signaling premeditation, intent to commit a crime and some level of organisation. Indeed, the Irish police service have outlined that specialized burglary gangs are committing the burglaries, rather than petty criminals. Fourthly, the motorway burglaries coincide with a period of rising burglaries across the European Union, so the inducement into burglary activity is likely driven by international factors. Recent research posits a strong link between recent property crime rates and the change in world prices of metals and other commodites that can be stolen from properties (Draca et al., 2015) 6 .
The remainder of the paper is structured as follows. Section 2 discusses the background of organised burglaries and Section 3 outlines the theoretical framework. In Section 4, the data sources and the variables used in the empirical analysis are described. The empirical strategy is described in detail in Section 5. In Section 6, the empirical results are explained and robustness checks are provided in Section 7. Section 8 extends the analysis to the effect of the introduction of anti burglary measures in Ireland, and Section 9 concludes.

Background
Approximately one burglary is committed every one and half minutes in the European Union (EU), and some member states register one thousand burglaries per day (Europol, 2017). Organised property crimes carried out by highly specialised and mobile organised crime groups are one of Europol's priority crime areas under the EU policy cycles of 2014-17 and 2018-21 (EU Commission, 2017) 7 . The property crime offences include organised burglaries, thefts, robberies and motor vehicle crimes. Burglary is one of the only crimes in the EU that grew in intensity during 2007 to crime groups in Europe are largely similar and reflect a new strand of organised criminal activity (Savona and Riccardi, 2015). Criminal organisations are changing and adapting their strategies by shifting their interests to new areas of activity that are less risky and less violent (Savona and Riccardi, 2015).
Illegal activity in the Irish economy is estimated to 1.1 per cent of GDP and organised crime in Ireland generates at least €1.7billion a year, which is more than the Irish Governments annual spend on the police service in 2016 (Savona and Riccardi, 2015) 8 . This is smaller than the 7 per cent of Italian GDP, but it is still one of the highest percentages in the EU (Savona and Riccardi, 2015; UNODC, 2011) 9 . The Irish police service has stated that specialized burglary gangs are exploiting the motorways connectivity and speed limits to burgle distant rural targets 10 . Figure 1 illustrates that burglary rates are higher in police sub districts closer to the motorway network, relative to more isolated parts of the country. In 2016, the Irish police service launched nationwide measures to reduce motorway-related burglaries, which decreased the number of burglaries nationally by 30% 11 . Criminals using motorways as an escape route does not appear to be investigated by police services abroad in the same detail as it has been in Ireland, but it has been cited in crime and media reports for several other EU member states, including the United Kingdom and Belgium 12 . The lack of police attention towards motorway burglaries is not necessarily a reflection of a lack of this type of crime in other countries but likely due to the classification of individual burglary incidents as petty criminality (EU Commission, 2017).
The Irish police service estimates that 75% of all burglaries are carried out by 25% of offenders 13 .
The most common items stolen are jewelry, cash and televisions, and the average amount stolen 14 .
There are at least seven criminal gangs involved in the burglaries, with one gang committing up to 25 burglaries per week, and therefore multiple burglaries per day 15 . Unlike traditional Mafia-type gangs, these gangs do not appear to compete territorially with another, but instead they cooperate by Burglary rates in police sub districts and the motorway network in 2015 selling information to rival gangs 16 . The burglars exhibit a preference for targeting rural areas, presumably because police prescence is much lower than in urban areas 17 . Flying drones, night vision goggles and high-powered armored cars are among the tactics used by the professional 15 See "Ireland's burglary gangs: The dirty half dozen who have created a climate of fear", available at: http://www.sundayworld.com/news/crimedesk/irelands-burglary-gangs-the-six-who-have-created-a-climate-of-fear, retrieved and http://www.independent.ie/irish-news/courts/highly-sophisticated-eastern-european-thieves-making-afortune-in-burglaries-targeting-rural-ireland-35605436.html 16 See "Gardai alert over postbox burglars" availible at https://www.irishmirror.ie/news/irish-news/crime/gangstravelling-burglars-posting-stolen-1806747 17 Approximately 30% of Ireland's population lives in County Dublin and so policing resources are concentrated there. Furthermore, a recent media report details how rural burglars receive lighter sentences. See "A new sense of fear is stalking rural Ireland as people's sense of isolation reaches dangerous levels" https://www.independent.ie/irishnews/news/a-new-sense-of-fear-is-stalking-rural-ireland-as-peoples-sense-of-isolation-reaches-dangerous-levels-35497924.html, and "Why do rural burglars receive lighter sentences than Dublin ones?" available at: https://www.irishtimes.com/news/crime-and-law/why-do-rural-burglars-receive-lighter-sentences-than-dublin-ones-1.3352167?mode=sample&auth-failed=1&pw-origin=https%3A%2F%2Fwww.irishtimes.com%2Fnews%2Fcrime-and-law%2Fwhy-do-rural-burglars-receive-lighter-sentences-than-dublin-ones-1.3352167. burglars to select targets and to escape 18 . Although based in Irish cities, the criminal gangs have connections across the country and often hire local informants to carry out reconnaissance missions to scope out targets 19 . The burglars are not all Irish residents and one of the gangs travelled to Ireland from Eastern Europe 20 . The Irish police have detailed how it is difficult to catch these travelling burglars because they often offload their stolen loot as quickly as possible. In some cases, they travel immediately from the crime location to the selling market 21 . In other cases, they store the stolen items; they put stolen items into padded envelopes and post them to a pre-arranged address, or they store the goods in wheelie bins over night 22 .

Theoretical framework
The economic analysis of criminal behaviour is based on the premise that criminals are concerned with maximizing profit and avoiding detection. The economic incentives of crime are rooted in Becker's (1968) seminal contribution and later extended by Ehrlich (1973). In the model an individual decides to commit a crime if the expected benefits from the crime are greater than the expected costs. The model has mainly been applied to property crimes, and a small literature has applied the model to violent crimes (Grogger, 2000). A simple framework is presented in this Section that adopts insights from the Becker model, to provide the basis to expect the allocation of burglaries to areas near a motorway. The decision to commit a burglary is outlined first, followed by the choice of crime location.

The decision to commit a burglary
A risk neutral agent (i) decides whether or not to commit a burglary by calculating the average expected profit ( per burglary offence: is the probability of being caught and convicted, is the resale value of the stolen goods and is the extent of the punishment. is falling in and , and rising in 23 . The first term on the right hand side of equation 1 is the expected gain from committing a crime that results in getting caught, while the second term represents the gain from crime that does not end in getting caught. The decision-maker chooses to commit the burglary if the expected profits from committing the offence are greater than earnings from legal work (w): (2) = { > >

Location choice
If the individual chooses to commit a burglary, they next decide on a location for the crime. To arrive at this decision, the individual first calculates the expected gains of committing a burglary in j alternative locations 24 . The probability of being caught and convicted for committing a burglary in j is: (3) , , , , , , j= 1, 2, …, k is the travel time from the target location to the base address of the individual or the sellingmarket, is the effectiveness of the local police and is a residual term. The probability of being caught and convicted is rising in and . Motorways connect distant locations through high-speed limits and a lack of traffic barriers, thereby decreasing the time that the individual is in transit with the stolen goods. The less effective the police, the more likely a burglar can evade detection. Policing action affecting the probability of being caught, alongside other factors, will vary by strength 23 The decision to commit a burglary or theft in a certain area is argued to depend on the trade off between the probability of arrest and the possible stolen goods in the area (Deutsch, Hakim, and Weinblatt 1987;Deutsch and Epstein 1998). 24 This is in line with studies in criminology that argue that burglars follow a multiple-stage decision making process when trying to select a crime location (Bernasco et al., 2017) geographically 25 . The expected profits ( ) accumulated from committing a crime in a given neighbourhood are shown in equation 4. The values of and calculated in equation 1 are inputted into the calculation: The individual chooses to commit the burglary in the location where they can collect the greatest expected profit. Burgle is a binary variable that equals one when the offence will be committed in location j, which is conditional on the expected profit in that location being greater than in an alternative location 26 : District diversification can occur if for example travel time from j and police effectiveness in j is lower than in an alternative neighbourhood.

Extension
Extending the simple framework in Section 2.2.2 to a population of identical individuals, the number of burglaries in j is equal to the number of individuals who choose j: The theoretical framework is static and per burglary but it can also explain additional offences, as the decision-maker follows the same process for each subsequent burglary. At every period of time, the individual updates the value of the parameters and calculates another set of expected payoffs before deciding where to commit the next burglary. As motorway expansion occurs, the probability of being caught and convicted changes geographically (decreasing in recently connected areas). Police effectiveness may also rise in areas connected in the previous period, if motorway-related burglaries occurred. As the parameter values change, the chosen location changes. If there is no change in the parameter values then the burglar will commit more than one burglary in the same location.

Geographical layout of policing
According to the 2016 Census, the (Republic of) Ireland has a population of 4.76 million people 27 .
Northern Ireland is a political jurisdiction that is part of the United Kingdom, and is therefore not Each SD generally has 1 police station, the strength of which varies from 3 to 100 police officers. In some areas there are stations known as sub-stations, which typically feature one officer conducting administrative work adjacent to the parent station 30 . Dublin City is made up of 9 SD's, and Dublin County contains 39 SD's.

Burglary
An Garda Síochána is the national police service in Ireland and is responsible for policing across the country. As part of their daily duties, police officers are required to record criminal activity on the that are provided on an annual basis for SD's 32 . In 2015, burglary was the third largest crime type in Ireland and the Irish Crime Classification System (ICCS) defines a burglary as the unauthorized entering of a building or part of a building, either with the intent to commit an offence or having committed an offence 33 . To account for heterogeneity in SD size, the ratio of the absolute number of crimes committed in a SD to the population of the SD is used as the dependent variable (expressed per 10,000 persons). Population statistics for SD's are only available in 2011, but Section 6.5 shows that the main results are robust to using population data constructed from population growth figures in the 8 regional authority areas in Ireland. The average number of burglaries in a SD is 46 and the average burglary rate is 48 per 10,000 persons (see appendix A). There are some limitations associated with using Irish crime data that must be considered before analysing criminal activities in Ireland. Firstly, crime counts are often an under representation of the true count, because not all crimes are reported to the police. Furthermore, counting and recording rules typically mean that only the most serious offense is recorded in a criminal transaction. The CSO estimates that 62% of burglaries were reported to the police in 2015, and that burglary was the crime that was most likely to be reported at (73%) (CSO, 2016) 35 . Secondly, recorded crimes may be incorrectly categorised or re-categorised. Thirdly, the length of time between reporting a crime and the recording of the crime on the PULSE system could result in the misspecification or omission of crime data. Of course, the use of yearly data will offset this concern to some extent. Lastly, evidence suggests crime data in Ireland is often incorrectly labeled "detected" or "invalidated". these issues are systematic and patterns of variance are unlikely to exist across SD's, or be correlated with motorway expansion.

Motorway
There is approximately 100,000 km of road network in Ireland. In 2015, there were 12 motorways in Ireland, the highest standard of road, representing 916km of the total road network. In order to construct a measure of motorway access, the location and year of opening of junctions is identified.
There is no official database on junction locations so firstly, the name of motorway junctions in     Table 1 shows the number of annual motorway connections. A common problem in the transport literature is measurement error related to how road access is measured (Picot et al., 2015). This concern does not apply in this paper because using the geographic coordinates for motorway junctions takes into account the exact spatial distribution and local availability of motorways 44 . 43 Studies find a very high correlation between straight line distances and drive time distances (Boscoe et al., 2012;Phibbs and Luft, 1995). However, exceptions exist near physical barriers (Boscoe et al., 2012). If burglary is always low in SD's with physical barriers such as rivers, then this will be accounted for in the fixed effect. 44 Proximity to motorway junctions is used, rather than proximity to the nearest part of the motorway road, because the exact entry and exit point of criminals is necessary to identify the effect of motorway access.

Empirical strategy
The causal effect of motorway connections on burglary rates is analysed using standard panel data methods. The fixed effects estimation takes the following form: (7) burglary i,t = + θ i + α t + βΔconnect i,t + i,t 'burglary' is the annual burglary rate (per 10,000 persons), i is the sub district (562) and t is time . 'connect' is a dummy variable that takes on the value of 1 when the first junction within a ≤30km radius of the SD's centroid opens 45 . An alternative variable (denoted 'motorway') indicating the presence of a motorway junction within ≤30km radius is also attempted. Motorway expansion is not the only determinant of burglary location so θ i is included which is a SD fixed effect that removes time invariant unobservable characteristics that vary across SD's e.g. topography and deprivation. α t is a set of year dummies that control for national trends in crime e.g. higher rates of criminal behaviour during a recession (Bell, et al., 2014) 46 . Standard errors are clustered at the SD level.
As detailed in the theoretical framework in Section 3.2, a positive relationship between motorway connection and the burglary rate is expected. The rationale being that connection to the motorway network allows criminals to escape the SD and reach their destination more quickly. Exploiting the combination of: the spatial variation in motorway connectivity, the timing of new connections and the inclusion of geographically detailed fixed effects, creates exogenous variation in motorway connectivity. This is made possible by the geographically detailed nature of the data, which links sub districts to the motorway at a spatially granular dimension and overtime. Therefore, after adjusting for local and national factors that influence the level of crime, β is interpreted as the causal effect of motorway connections on burglary rates.
The main endogeneity issues revolve around the three common issues experienced in transport infrastructure studies: omitted variable bias, reverse causality and the endogenous placement of roads 45 An alternative is the year of connection to the nearest junction i.e. the last junction to be placed nearby in the period of analysis. Using this measure, both fixed effects and differences-in-differences specifications are estimated, they yield statistically insignificant results. This is intuitive conceptually, because it is not about how close an area is to motorway junctions, rather the time at which access by the criminal begins. In this paper, accessibility is defined as a junction within 30km radius of the SD. 46 Hargaden (2016) finds a statistically significant positive relationship between property crimes and the unemployment rate in Ireland during 2003-16. (Picot et al., 2015) 47 . Unobserved effects causing crime may be present that can be time invariant or time varying. In the case of time-invariant effects, using spatially detailed location fixed effects alleviates the concern, but this will not solve the problem of time-variant effects 48 . In Section 7.2, the relationship with motorway connections is shown to be unique to burglary rates, as other types of crimes yield statistically insignificant effects, when used as alternative dependent variables. This provides evidence that the results are reliable because if ommited variable bias were present, other crimes would be effected by the motorway.
The main omitted variable of concern is local changes in economic activity because it varies spatiotemporally. In Section 7.3, the main result is robust to the inclusion of a proxy of local economic activity as a control variable. The uniqueness of the causal relationship of motorways on burglary relaxes reverse causality concerns. For reverse causality to exist, the motorway connections would need to be driven by only burglary rates and no other crime type, which is unlikely. Motorways are generally allocated to specific locations according to unobserved characteristics not orthogonal to their economic potential (Picot et al., 2015). This endogenous placement of motorways might upwardly bias the estimates. This concern is eased because connections to randomly postponed motorway junctions have no statistically significant effect in Section 7.1. If problematic unobserved characteristics at motorway locations are biasing the result, this would be revealed by postponed junctions, because the unobserved variables exist even if the motorway does not yet.

Results
Table 2 presents the main results of the effect of motorway connections on burglary rates. The presence of a motorway nearby is measured using the dummy variable 'motorway' that equals 1 if there is a motorway junction within ≤30km radius of a SD's centroid, and 0 for disconnected sub districts. New connections to the motorway are represented by the dummy variable 'connect', which equals 1 in the year a motorway junction is placed within ≤30km radius of a SD's centroid for the first time i.e. the year of motorway connection, and 0 otherwise 49 . A positive coefficient is expected 47 Many transport studies use an optimal road network instrumental variable to circumvent endogeneity issues. However, this IV is not easily applied to the panel data in this paper because the majority of the motorway is built in the period of analysis, and do the IV is not capable of exploiting the year-to-year variation. 48 For example, Holtz-Eakin and Schwartz (1994) show that the state-level impact of public capital can be entirely attributed to state-level time invariant unobserved effects. 49 A dummy variable equaling one every time a nearby junction is added was attempted, the coefficients are positive and of a similar magnitude to table 3, but not statistically significant. Distance to motorway junctions and changes in this distance where also attempted as an alternative measure of motorway connection. At the Division level (28), dummy on both variables because criminals will commit burglaries in SD's close to the motorway network for ease of escape.

Postponed motorway junctions
Motorways are not assigned to locations at random, but on the basis of unobserved characteristics that affect economic activity, which in turn affect the level of criminal activity. Motorways will generally be placed in areas of above average economic prosperity, or areas with economic potential.
The variable measuring connections to the motorway may be inflated as it fails to adequetely disentangle this relationship. Sections of the Irish motorway planned to be built during 2004-15 were postponed and connections to these postponed junctions are included as a useful placebo test (table   3). There are twenty-two postponed junctions, belonging to two new motorways and extensions to 51 The results are also robust to the exclusion of the Greater Dublin Area (consisting of the counties of Meath, Kildare, Dublin and Wicklow), which is defined as a core socio-economic region in Ireland.
three existing motorways illustrated in Appendix C 52 . As the year of opening of postponed junctions does not exist, the earliest year of reference that can be found for each motorway expansion is used.
It is expected that connections to the motorway are exogenous, and that there will be no causal effect associated with the non existent junctions. To note, theft refers to the action of stealing, while burglary refers to breaking an entry with the potential intent to commite a crime, such as theft. A 1 unit increase in the theft rate increases the burglary rate by almost the same amount (0.91). This is an intutive result as burglary and theft are related crimes.
One might expect that motorways will increase other crimes that require a physical escape, such as theft and vandalism. To test this, other crime types are used as alternative dependent variables and the results are presented in table 5. SD's connected to the motorway network experience more property damage offences (i.e. vandalism), and the effect is statistically significant at 1%, but new connections have no effect. Drug trade is negatively affected by motorway connections, but the effect is Dublin-specific. A crucial result that reduces endogeneity concerns is that motorway connections do not cause the theft rate. This is important because the theft rate is closely related to the burglary rate in table 5. If there were an omitted variable being captured by connections, this available at www.breakingnews.ie/ireland/cork-to-limerick-motorway-and-dublin-airport-included-in-115bn-nationaldevelopment-plan-826540.htm. 54 This result is robust to different years of opening of postponed junctions. 55 Kidnap, fraud and weapon related offences are excluded because of low counts. 56 The presence of a motorway also remains statistically insignificant when other crime types are controlled for in the regression.
would be revealed through connections having a statistically significant effect on other crimes, particularly theft. Conceptually, the uniqueness of the relationship is perhaps because burglary requires the expedient transit of bulky stolen items such as televisions.

Local economic activity
The SD fixed effect remove unobserved effects that vary over space, and the year dummies remove unobserved effects that vary over time, but neither account for the variables causing the local crime rate that vary over both space and time. Left unaddressed these effects can cause an omitted variable bias because identification rests on spatio-temporal variation. The main culprit is changes in local economic activity e.g. connections to the motorway network might simultaneously raise economic activity and the burglary rate, and in turn the burglary rate might rise with economic activity. The geographic coordinates of welfare offices are found in Google maps and the distance from the SD's centroid to welfare offices is used to identify the local number of registrations. Information on the live register is published by the Central Statistics Office in the form of a monthly release availible at http://www.cso.ie/px/pxeirestat/Database/eirestat/Live%20Register/Live%20Register_statbank.asp?SP=Live%20Registe r&Planguage=0. 59 The live register is also scaled by population e.g. the population size in the SD and the County the office is located in. Although the results on the live register are sensitive to the scaling, the coefficient on motorway connections is not. As the scaling does not affect the main result, and as there is no obvious population to scale by, the live register is left unscaled for ease of interpretation.   showing that motorway connections are not capturing changes in local economic activity. previously targeted sub districts, as it is likely that they do return (Bernasco and Luykx, 2003). It indicates that the pattern of revisiting is not based on the year of motorway connection, for example, returning 3 years after the year of connection, and thus the pattern of return is not testable in this paper, this discussion is continued in Section 8.

The border with Northern Ireland
The border between the Republic of Ireland and Northern Ireland is a hard policing border but with free unchecked movement because both countries are in the European Union. Each jurisdictions police force is not legally permitted to continue the chase or make an arrest in the other jurisdiction.
From a burglar's perspective, the probability of apprehension may be smaller closer to the border, because the criminal can flee across the border after committing a crime. If the effect of the border is time invariant, it will be removed in the SD fixed effect, but road changes in Northern Ireland may cause the border effect to vary overtime. It is not possible to include a border dummy variable alongside the SD fixed effect, so SD's (75) whose centroid is within a ≤30km radius of the border are excluded, and the results are shown in Appendix D. Connections to the motorway network remain statistically significant at 1% and the coefficient is a similar magnitude to the main result.
This provides reassurance that the effect of the border on burglary rates is not disrupting the main result.

Extension
Before 2016, there was virtually no targeted law enforcement against criminals on motorways. In 2016, a €5 million national anti-crime strategy called 'Operation Thor' launched across the country, 62 The Regional Authority is the smallest geographical level available that provides yearly growth rates. The Eight Regional both motorway and non-motorway connected regions experience a decrease in burglaries, but the fall is larger in places connected to the motorway network.
A differences-in-differences model is estimated to determine causality between the introduction of the targeted law enforcement at motorways and the fall in burglaries in divisions with a motorway. Since all divisions receive a treatment and a subsequent decrease in burglaries, it is difficult to arrive at a precise estimate of the number of motorway-related crimes the policing measures prevented.
Furthermore, media reports outline that burglars reacted to the policing measures at motorways by reallocating some of their activity towards the old primary road network where police presence is less i.e major roads that were downgraded as higher quality roads were built. This indicates that the rise in policing at motorways dispersed some of the exisiting activity, which provides an explanation as to why the estimated treatment effect is small. Since divisions are large areas, crime can move from the part of the division where there is a motorway towards the part of the division further away from the motorway, theoretically leaving the number of burglaries in the division unchanged 66 . It is not possible to investigate this further because quaterly data is not available at a lower geographical level. Figure 2 also shows that the number of burglaries increased in Dublin in 2014, while simultaneously decreasing in other motorway counties. This suggests that burglars substituted some part of their previous motorway activity outside of Dublin for Dublin burglaries, in response to increased policing near motorways. The effectiveness of the targeted action in 2016 occurred after a period of limited motorway growth, there were no expansions during 2011-14, and only three junctions were built in 2015 (table 1). The continuance of motorway-related burglaries during this period shows that criminal's return to previously targeted areas, and that motorway-related burglaries are not restricted to the year of connection. Most likely, burglars return when police vigilance has lapsed. In the theoretical framework presented in Section 3, this will be represented by a decrease in police effectiveness, and a subsequent rise in expected profits, however there is no discernable pattern to explain when police vigilance falls back to pre-motorway-burglary levels.

Conclusion
This paper studies the causal effect of motorway connections on burglary rates using data from Ireland. The spatio-temporal variation in motorway connectivity is exploited in a fixed effects analysis, consisting of crime rates in 562 police sub districts and connections to motorway junctions during 2004-15. Connection to the motorway network causes a 10% rise (5 burglaries) in the number of burglaries in the average sub district. The rise in crime occurs in the year of connection only, and there is no persistent effect on the burglary rate in the years after connection. It is likely that burglars return to previously targeted areas, but the year of return is not patterned with the year of motorway expansion. For this reason, this paper focuses on the immediate aftermath of motorway expansion and it is estimated that motorway expansion caused 12,150 burglaries in recently connected sub districts during 2004-10. The immediate response of burglars to motorway expansions is a novel result and a possible explanation is that the probability of being caught by the police is lowest in areas recently connected to the motorway network, and criminals exploit the fact that the police may not have had sufficient time to increase policing in these areas. The novelty of this paper is to show that major road construction influences the spatial distribution of crime. For policymakers, the findings indicate the need to incorporate crime-externalities into any cost-benefit analysis of future road infrastructure developments. The measure of motorway connections can confidently be thought of as exogenous because the main results are robust to various detailed checks: postponed motorway junctions yield no statistically significant effect on crime, controlling for local economic activity is not disrupting the main result, and no other crime shares this causal relationship with motorways. A possible explanation for the uniqueness of the effect of motorways on burglary, is that the bulky stolen goods need to be transported from the property as quickly as possible to avoid detection.The national policing action against motorway burglaries launched in 2016 was effective in reducing burglaries, providing further weight to the argument that motorways are affecting the spatial pattern of burglaries.

Appendices
Appendix A     Table 14.
The effect of motorway connections on sub district burglary rates (per 10,000 persons), using yearly population estimates 2004-15 Notes: ***p<0.01 **p<0.05 *p<0.1. Standard errors are clustered by sub district and are in parentheses.
Numbers are rounded to two decimal places where possible.