Introduction

Chemical property marking involves the labeling of personal possessions with an invisible, traceable chemical designed to indicate ownership. This article describes a randomized controlled trial of the deterrent effects of a chemical property marking program on household burglary in Denmark. More precisely, it examines whether offering households an opportunity to post warning stickers indicating their use of a chemical property marker discourages would-be burglars from targeting treated homes.

The current study is a near-replication of a Danish experiment with chemical marking that was conducted in 2016 (Kyvsgaard & Sorensen, 2020). The new study is based on 12,000 households randomly assigned to one of three groups: treatment, placebo, or control.

Denmark is a small, affluent, relatively homogeneous northern European country of 43,000 km2 and 5.8 million people (Statistics Denmark, 2021). Survey data consistently rank the Danish population as having one of the highest feelings of personal safety in Europe (OECD, 2020: 151). Visitors are therefore often surprised to hear that Denmark suffers from one of the highest rates of household burglary in the European Union—an unfortunate distinction supported by both official police data (Aebi et al., 2014: 49) and victim surveys (European Union Agency for Fundamental Rights, 2021: 65). The reason for this high rate of burglary is unclear, though research indicates it cannot be attributed to simple differences in legal definition or registration practices (Kruize & Sorensen, 2018: 17–18).

Property marking is an example of situational crime prevention (SCP). SCP seeks to reduce crime by making it appear more difficult, more risky, and/or less rewarding in the minds of potential offenders (Clarke, 1997). Property marking is designed to serve several functions: (1) to deter burglars from targeting homes that post warning stickers indicating its use; (2) to ease law enforcement’s ability to determine if an item is stolen; (3) to facilitate prosecution and conviction of suspects in possession of marked property; (4) to make it more difficult for burglars to sell marked items to buyers of stolen property; and (5) to quickly identify and reunite original owners with their property.

Chemical marking (sometimes called “forensic marking”) is the newest technological stage in the evolution of property marking. It marks property via the application of a few drops of an invisible, traceable solution. Traceability is achieved through the use of a chemical composition that contains a unique identification code for every individual marking set sold (The Crime Prevention Website, 2021). Every package of chemical markers sold has its own unique identification code. Purchasers are instructed to register this code in an online database maintained by the product’s manufacturer. This is a vital step, since police cannot identify owners of recovered property unless it is registered in the database. Chemical markers are largely invisible to the naked eye and can be safely applied without damaging most items. Markers generally contain a tracing dye that allows anyone with a UV light to quickly determine if an item is marked. This is useful both to police and professional fences (i.e., buyers of stolen property), the latter of which can use the presence of chemical markers to bargain down the price they will pay for marked goods. Once applied, marks are nearly impossible to remove without damaging the property. Chemical marking packages generally include a small bottle of marking solution, an applicator brush, warning stickers for posting on doors and windows, and the opportunity to register one’s property in an online database for the life of the product.

Prior Research on Property Marking

Prior research on property marking has examined its effects on various outcomes including the deterrence of burglary, the arrest and conviction of offenders, and the successful return of stolen property. Units of analysis have included households, neighborhoods, and cities. Since households are easier to randomize than geographic areas, the comparability of treatment and control groups is generally more convincing when households are used. Research on burglary reduction is in all cases focused on the deterrent effect of warning stickers as opposed to that of the property marks themselves. After all, there would be no reason to expect a deterrent effect if burglars were completely unaware that a household’s property was marked.

Property marking initiatives became increasingly popular in the mid-1960s (Heller et al., 1975: 1). By the mid-1970s, over 80% of US police departments had established “Operation Identification” programs (White et al., 1975: 26) and the technique had spread to Scandinavia and other parts of Europe (Knutsson, 1984: 9). The three earliest studies of property marking used non-chemical markers (UV pens and engraving) and have been reviewed in detail elsewhere (e.g., Kyvsgaard & Sorensen, 2020). They are therefore covered here only briefly.

A meta-evaluation of property marking programs in the USA conducted by Heller et al. (1975) arrived at the following conclusions: (1) subject recruitment is extremely difficult; (2) subjects that did participate had significantly lower rates of burglary than non-participants (though the lack of random assignment means that some or all of these differences may have been due to selection mechanisms); (3) there was no evidence of an effect on citywide rates of burglary; (4) no evidence of an effect on rates of arrest or conviction of burglars; and (5) no evidence of an effect on the rate at which stolen property was returned to its owners (Heller et al., 1975: ix–x). Heller et al., (1975: x) attribute the latter three points to low participation rates.

Knutsson (1984) conducted a study of property marking with engraving pens and invisible ink in Huddinge, Sweden. Three thousand five hundred single-family homes were offered property marking kits and warning stickers within an atmosphere of heightened media attention. A quarter of these households posted warning stickers as instructed. But an impact evaluation comparing participating and non-participating households found no effect on rates of burglary or clearance.

Laycock (1985, 1991) conducted an experiment with UV markers in three villages in South Wales, UK. Intense media focus later proved detrimental to interpretation of its results. Treatment households were visited by police and offered free UV markers and warning stickers. A process evaluation determined that over 70% of the 2234 households assigned to treatment posted warning stickers as instructed. An effectiveness evaluation indicated a significant decrease in burglary at houses that had posted stickers as compared to those that had not. Laycock interpreted this as a success but cautioned that her results might be dependent on high levels of program compliance, police involvement, and media publicity.

A study by Bowers et al. (2003) in Liverpool, England, was probably the first to include chemical property marking in an anti-burglary program evaluation. Their evaluation looked at the individual and combined effects of four anti-burglary treatments: target hardening, alley-gating, chemical marking (with accompanying warning stickers), and offender rehabilitation. While target hardening (i.e., installation of mortise locks, door chains, and window locks) significantly reduced the risk of burglary and repeat burglary in treated households, chemical property marking was associated with a slight increase in the risk of burglary (ibid: 33–34). Readers should note, however, that the analysis of chemical marking was based on a very small sample (n = 54).

Brooks et al., (2015: 5–11) conducted the first large-scale study of chemical marking in Joondalup, Australia. Perhaps inspired by Laycock’s (1985, 1991) achievement of a 70% participation rate, Brooks’ team hypothesized that saturation treatment was necessary if property marking should impact area-wide burglary rates. Without saturation, they argued, burglary would simply be displaced from one house or neighborhood to another. The study sampled houses in three areas of the city comparable on socio-demographics and urban form: a treatment area (n = 278 houses); an adjacent control area (n = 300); and a non-adjacent control area (n = 300). The three areas were chosen based on their similarity in terms of socioeconomics, physical layout, transportation infrastructure, and major highway access. The experiment was designed to look at two issues: the effect of saturation-level chemical property marking on the incidence of burglary and the extent to which this would displace burglary to the adjacent control area. Houses in the treatment area received visits from neighborhood watch members who offered free property marking kits and, when asked, helped with the marking of property. The kits provided included warning stickers to post on doors and windows. Seventy-nine percent of the houses in the treatment area signed up for the offer, though it is unclear what percentage actually posted stickers as instructed. While sample sizes were small, an effect evaluation indicated that the decline in burglary in the treatment area was (statistically) significantly greater than that in the non-adjacent control area. They interpreted this as support for the hypothesis that saturation-level property marking—with sticker posting—deters burglary. On the other hand, the results also indicated a statistically significant rise in the incidence of burglary in the adjacent control area—which the authors interpreted as evidence for displacement. The authors conclude that saturation-level property marking (> 80%) accompanied by the posting of warning stickers deters burglary but displaces it to adjacent unprotected areas.

Raphael (2015) reports on a series of experiments with saturation-level chemical marking in five high-burglary districts in London. Each treatment area was matched with a control area of comparable socio-economic composition. The treatment and control areas each contained a total of 5000 households. Police visited treatment area households and helped residents mark their property, register themselves in a property marking database, and post warning stickers on doors and windows. They aimed for saturation-level treatment and succeeded in getting property marked and warning stickers posted at up to 85% of all residences in the combined treatment area. Warning signs were also posted in yards and on the streets, and a public awareness campaign was established to advertise the program on Facebook, Twitter, and in the local print media. An effect evaluation conducted 1 year after the start of the experiment showed that burglary had declined in all areas, but that it had had fallen by 21% more in the combined treatment area than it had in the control areas. The experiment was therefore deemed a success. Furthermore, police data indicated the possibility of a diffusion of benefits (Clarke & Weisburd, 1994), as neighborhoods just outside the treatment areas experienced a greater decline in burglary than non-adjacent neighborhoods.

Lindström and Olsson (2016) conducted an experiment with forensic marking in three burglary-plagued residential neighborhoods in Sweden. Participants were solicited via a neighborhood information meeting where free marking kits were handed out to the first 500 households in each area to register for the experiment. As in the London experiment, the program was highly publicized with signs set up throughout all treatment areas. The program’s effect on burglary was evaluated at both household and neighborhood levels. Household-level evaluations were conducted by means of a survey that measured compliance with the sticker posting protocol and the extent of victimization during the follow-up period. The survey indicated that residents who claimed to have posted stickers as instructed reported fewer burglaries in their homes, sheds, and garages than those who admitted to not having done so. The researchers also looked at official police reports in order to gauge the program’s effect on area-level burglary rates. Data indicated a lower rate of officially reported burglary in the treatment areas as compared to control areas, though this difference diminished with time. The authors acknowledge that the initial effect may have reflected a regression to the mean as rates of burglary had been unusually high in the treatment areas during the year prior to the experiment.

Hodgson et al. (2018) evaluated the effects of a series of forensic property marking programs initiated by the West Mercia Police in England. Treatment at the 14 experimental sites was varied, but generally included giving free forensic property marking kits to victims of burglary and free UV property marking pens to victims’ neighbors. Both types of kits came with accompanying warning stickers to post on doors and windows, and placards advertising the initiatives were put up in all 14 treatment areas. The statistical analysis aggregated data from the 14 areas and treated them as a single treatment area. The rest of West Mercia served as a control area. An outcome evaluation showed that while domestic burglary dropped in all parts of West Mercia, it decreased more in the treatment area than in the control area. Meanwhile, an increase in burglary at unoccupied buildings suggested the possibility of a displacement effect. Hodgson et al. (2018: 47) interpret the results as suggestive, but mention that a more conclusive result would require a bigger sample, randomization of treatment and control groups, and a more homogenized treatment protocol. They also suggest that future experiments include a longer follow-up period, since it is also unclear whether the observed effect was due to the treatments themselves or to the “community buzz” that surrounded the interventions.

The studies cited above have produced mixed conclusions. Saturation-level treatments (e.g., Brooks et al., 2015; Laycock, 1985, 1991; Raphael, 2015) appear most likely to generate statistically significant results. Among non-saturation studies, those conducted at the household- as opposed to area-levels are more likely to result in statistically significant findings. This is not particularly surprising, however, given that it requires extremely large treatment samples if one is to hope to make a dent in area-level burglary rates (Heller et al., 1975: 11). Furthermore, to the extent that displacement from a treated to untreated house occurs—though research suggests it is far from inevitable (Johnson et al., 2014)—target displacement will show up as a success at the household level, but as no net effect at the area level. Almost all of the studies reviewed have taken place in an atmosphere of intense media coverage. While promotion is intended to amplify program deterrence, it is unlikely to be part of the proliferation of chemical property marking in a non-experimental setting and complicates the process of identifying the independent causal effects of the property marking programs themselves. The most significant problem, however, and one that plagues all of the studies discussed above, concerns self selection. Regardless of whether households elect to join a treatment program or are randomly assigned to it, the decision to fully cooperate with experimental protocols—most notably the posting of warning stickers—is never made in a vacuum. Households that agree to register for property marking programs, and that work hard to follow their protocols, are likely to be both more concerned with the risk of burglary and more proactive in its prevention than those that do not.

These methodological concerns were the impetus for a randomized controlled trial on the effects of property marking in Aarhus, Denmark, in 2016 (Kyvsgaard & Sorensen, 2020). The Aarhus Experiment is described in detail below as it serves as the starting point for the North Zealand Experiment, which is the focus of this article.

The Aarhus Experiment

The sampling frame consisted of 6603 single-family houses within the municipality of Denmark’s second largest city, Aarhus (pop. 336,000 at the time of the experiment in mid-2017). Sampling was limited to houses that had reported one or more burglaries to police during the previous 4 years. These households were randomly assigned to treatment and control. Control households were not contacted. Treatment households received a letter from the mayor of Aarhus describing the upcoming experiment, and offering a chemical property marking kit free of charge. Fifty-six percent of the 3378 households randomly assigned to treatment registered for the experiment. A process evaluation later determined that 32% of the treatment households posted warning stickers as instructed. The total treatment group can therefore be meaningfully divided into three sub-groups: (T1) those who registered for the experiment and posted stickers; (T2) those who registered for the experiment, but failed to post stickers; and (T3) those who dismissed the experimental offer.

An effectiveness evaluation was conducted 15½ months later (Table 1). T1 (registered with visible stickers) had the lowest rate of burglary during the follow-up. This group’s burglary rate (3.8%) was, in fact, 39% lower than that of the control group (6.2%)—a difference that is highly statistically significant (χ2, p = 0.007).Footnote 1 While this finding provided superficial support for the research hypothesis, there was a strong possibility that self-selection might have contributed to the low rate of burglary in T1. After all, the fact that this group both registered for the experiment and carefully complied with its protocols suggests an above average concern for burglary prevention that may differentiate it from other groups. In order to reduce the risk of type 1 error, the effect evaluation was conducted as an intention-to-treat (ITT) analysis utilizing the full treatment group (Gupta, 2011). As shown in Table 1, the full treatment group had a household prevalence rate that is 24% lower than that of the control group (4.7% vs. 6.2%). This difference is both substantively and statistically significant (χ2, p = 0.021).Footnote 2 The article took this as the main result of the evaluation.

Table 1 Number of houses, burgled houses, and total number of burglaries during the 15½-month follow-up, Aarhus, Denmark (n = 6603)*
Full size table

The disaggregated results, however, contained a completely unexpected finding. As shown in Table 1, the part of the treatment group that never registered for the experiment (T3) experienced a lower follow-up rate of burglary than the control group (5.5% versus 6.2%). This was surprising because neither group had received treatment. More importantly, T3’s dismissal of the experimental offer would suggest a below average concern for the risk of burglary—and therefore a below average tendency toward proactive preventive behavior. On the other hand, the randomization process should ensure that the control group has an average concern with the risk of burglary and an average level of proactive preventive behavior. The difference between T3 and the control group was non-significant. So it was clear that the finding might be due to chance.

On the other hand, if meaningful, it suggested the possibility of an availability effect. Known as an availability heuristic or availability bias in psychology, an availability effect refers to a tendency for people’s judgments to be affected by the recency and salience of available information (Tversky & Kahneman, 1973). In the current case, it reflects the psychological impact of the letter describing local burglary risks that was received by all three treatment sub-groups, but not by the control group. This interpretation was supported by a month-to-month analysis of the follow-up period. The analysis revealed a tendency (based on data too sparse for statistical analysis) for the preventive effect in all treatment groups to decline with time, but to do so more quickly in T3 than T1.

The Present Experiment: North Zealand

The Aarhus Experiment provided tentative support for a positive experimental effect. It was, however, impossible to distinguish the effects of treatment from those of availability. A replication study therefore seemed warranted. Doing so would also satisfy Sherman et al.’s (2002) recommendation that experimental results be replicated in at least two study contexts before being accepted.

The North Zealand Police District lies in the suburbs just north of Denmark’s national capital, Copenhagen. North Zealand was chosen as the site of the new experiment on practical grounds. The district had the highest rate of domestic burglary among Denmark’s twelve police districts in the years leading up to the experiment. This meant there was real need for burglary prevention and a will among police to cooperate with the experiment. A second advantage to the choice of the North Zealand Police District was its population density. The district covers just 3.5% of Denmark’s area. This would prove a major cost-saver during the observational study, where student observers visit all of the treatment households that have requested a marking kit to determine whether they have posted warning stickers according to protocol.

The North Zealand Experiment is designed as a methodological replication of the Aarhus Experiment. As in Aarhus, the target population is limited to (stand-alone and row) houses that have reported a burglary to police during the 4 years prior to sampling. These sampling criteria were adopted to increase the likely number of burglaries captured by the experiment and thereby boost experimental power.Footnote 3

The Sample and the Experimental Conditions

The North Zealand Experiment differs from the Aarhus Experiment in two important ways. First, the sample used in North Zealand includes almost twice the number of households (12,000 as opposed to 6603). Second, the North Zealand Experiment introduces a “placebo group” in addition to the treatment and control groups. This placebo group is added to shed light on the potential presence of an availability effect, which may or may not have influenced the results of the Aarhus Experiment.

The 12,000-member household sample is randomly divided into three sub-groups of 4000 households each and exposed to the conditions described below.

The treatment group (n = 4000) received a letter from the District Police Chief in mid-May 2019.Footnote 4 The letter mentioned the high rate of burglary in the area and explained that the household had been randomly selected to be part of a property marking experiment. It then introduced the nature of chemical property marking and offered a free property marking kit (worth circa 100 Euro) if they registered for the experiment. The letter stressed the importance of posting warning stickers on post boxes and/or front entrances, windows and back doors, and of registering the unique identification code in the online database so that recovered property could be linked back to its owner. In mid-June, a reminder letter was sent to 2512 treatment group households that had failed to respond to the original letter. Fifty-six percent (n = 2252) of treatment households ultimately registered for participation in the study and requested a free property marking kit.

The placebo group (n = 4000) received a letter from the District Police Chief in late-May 2019. This letter reminded placebo group households that burglary rates are high in their area and provided some generic suggestions for how their household’s risk might be reduced, e.g., leaving lights on; forming a neighborhood watch group; and cutting down overgrown hedges. No mention was made of property marking or the experiment and they were not offered a chemical property marking kit.

The control group (n = 4000) was never contacted.

The North Zealand trial purposely avoided the promotion of citizen meetings concerning the experiment as well as any mention of it in the media. This was done to help isolate the effects of the experimental treatment, i.e., the posting of warning stickers. As demonstrated in the literature review, the media buzz surrounding almost all of the previous experiments has been an obstacle to the attribution of causality in effect evaluations.

Process Evaluation

An observational study was initiated in June 2019 in order to determine what proportion of the treatment households that had requested marking kits had posted warning stickers as instructed. By that time, 97.4% (2194) of the 2252 treatment group households that would ultimately register for the experiment had already done so. It quickly became clear that relatively few of the treatment households that had requested property marking kits had posted warning stickers. A new letter was therefore dispatched in late November 2019 to remind experimental participants of the importance of doing this. A second phase of observations began in early 2020. This phase was interrupted by the national lockdown instituted by the Danish government on March 13, 2020, in response to the COVID-19 pandemic. The remaining observations were therefore conducted in the beginning of summer 2020.

Twenty-nine percent (1178) of the 4000-household treatment group ultimately registered for the experiment and posted warning stickers as instructed. Meanwhile, another 27% (1074) of the treatment group registered for the experiment but failed to post stickers, while 44% (1748) never registered. Referral back to Table 1 above will show that the distribution in the Aarhus treatment group (32%, 24%, 44%) was remarkably similar to that in the North Zealand treatment group (29%, 27%, 44%). Despite sending two letters with treatment offers and a third letter reminding participants to post their stickers, only 29% of targeted North Zealand treatment group both agreed to register for the experiment and fulfilled its sticker posting requirements. This is significant improvement over the participation rates of “less than 10%” reported by Heller et al., (1975: ix–x) in the USA, but far worse than the “saturation-level” participation rates achieved by Laycock (1985), Brooks et al. (2015), and Raphael (2015).

Outcome Evaluation

Table 2 divides households by group and sub-group and shows the proportion of each that experienced a burglary during the 15½-month follow-up period (May 15, 2019 until the end of August 2020). Treatment group 1 (T1), which registered for the experiment and posted stickers as instructed, had the lowest proportion of burglaries. The risk of burglary is on the verge of being statistically significantly smaller for this part of the treatment group (i.e., T1) as compared to the control group (χ2, p = 0.067).

Table 2 Treatment vs. control group: number of houses, burgled houses, and total burglaries during the 15½-month follow-up, North Zealand, Denmark (n = 8000)*
Full size table

However, comparison of T1 on its own to the control group would violate norms of causal inference. This is because subjects who enthusiastically embrace treatment are more likely to achieve positive results. T1 households that have signed up for the trial and posted warning stickers as instructed are neither a random nor a representative sub-sample of the entire treatment group. As a group, they can be expected to be more likely than the other (non-compliant) household groups (T2 and T3) to engage in additional activities that reduce their risk of burglary. According to the intention-to-treat (ITT) principle (Gupta, 2011), all houses in the overall treatment group—including those that refused treatment or failed to follow its protocols—must be included in the comparison to the control group when assessing the experiment’s overall effect. This same approach was taken in the Aarhus Experiment. The difference in burglary rates observed for the overall treatment group (4.6%) and the control group (4.9%) in the North Zealand Experiment is substantively small and not statistically significant (χ2, p = 0.495).

The North Zealand Experiment introduced a placebo group (letter, but no marking kit) in order to test for the presence of an availability effect (Tversky & Kahneman, 1973). Table 2 shows that there is no difference in the proportion of placebo and control group households that experienced a burglary during the follow-up (χ2, p = 0.682). This has two implications: first, that it takes more than subtle messaging for people to increase their proactive preventive behaviors; and second, that if a difference between the overall treatment group and the control group had been found, it could be interpreted as solely due to the effects of treatment, i.e., chemical property marking advertised by the posting of warning stickers.

Conclusions

The North Zealand Experiment produces the following conclusions:

  1. (1)

    Due to low rates of compliance, simply offering households a forensic property marking system is unlikely to affect their overall risk of burglary.

  2. (2)

    Nonetheless, residents who actually post warning stickers may reduce their risk of burglary, though the reductions obtained are likely to be small and at least partly attributable to selection effects.

  3. (3)

    Receipt of a letter that simply calls residents’ attention to burglary risks and strategies of prevention (i.e., like the one received by the placebo group) has no effect on a household’s risk of future burglary.

The replication study in North Zealand was not only designed to test whether warning stickers are a deterrent to burglary, but also to shed light on the interpretation of the results from the previous (Aarhus) study, i.e., whether simply having received a letter concerning the experiment and the risk of burglary could, on its own, have minimized victimization by increasing the incentive of households to take preventive actions. It was this concern with interpretation that motivated the inclusion of a placebo group in the current experiment. Seen in retrospect and with the benefit of other studies, it was probably naive to assume that a single letter could have led to a measurable effect. True, there are numerous examples of campaigns that have prompted behavioral change—most notably in the anti-smoking arena (e.g., Vallone et al., 2018: 543–551; Duke et al., 2020: 645–651). However, their circumstances tend to differ from the current context in at least two important ways. First, almost all of these campaigns have been far more extensive than a single letter. Second, tobacco campaigns typically target young people who may be easier to influence than older smokers.

Relatively gentle nudges, like those resulting from a single letter, are rarely found in criminological research. Initiatives that have been evaluated more often concern comprehensive, multi-prong public campaigns that target offenders as opposed to victims.Footnote 5

A literature review conducted on behalf of the Danish Crime Prevention Council concludes that campaigns are ineffective on their own (Det Kriminalpræventive Råd 2015: 2). It therefore seems far from certain that small-scale preventive campaigns could influence behavior significantly, which contradicts the expectation that a letter sent to the placebo group was likely to affect burglary outcomes.

The experiment in Aarhus produced a short-term effect that was described as an uncertain finding (Kyvsgaard & Sorensen, 2020). In retrospect, this uncertainty cannot—as previously assumed—be explained by an availability effect. The result of the Aarhus Experiment is therefore probably best seen as a result of chance based upon among other things the dubious results, i.e., lower risk of burglary than the control group, obtained for households that received the treatment offer letter, but never registered to participate in the experiment. The lower rate of burglary observed for this group (T3) and/or the higher rate in the control group may have simply been the result of random variation.

In sum, the current experiment suggests that actually posting stickers may reduce burglary risk, though the effect is small and at least partially due to selection effects. An intention-to-treat (ITT) analysis comparing the full treatment group with the control group finds no clear evidence to support the hypothesis that the posting of warning stickers indicating a household’s use of chemical property markers deters burglary. The failure of the experiment to identify a statistically significant effect for property marking may be partially attributable to the fact that only a small proportion (29%) of the targeted treatment group actually followed the sticker posting protocol. Whether a deterrent effect would have been demonstrated if more households had complied with the experiment remains unknown.