The specific research questions for this study are as follows:
What is the best way to create a Danish Crime Harm Index based on the principles of the Cambridge CHI?
How does a Danish CHI, when applied to Danish national police data for 2016, change the relative distribution of proportionate impact of different crime types compared to the raw number of each offense type?
If the Danish CHI is applied to annual national police crime data from to 2011 to 2016, is the decrease in crime count frequency paralleled by a decrease in Danish CHI value?
The creation of a Danish CHI has several practical and policy implications: Besides the purpose of identifying harm-spots (supplementing crime count “hot spots”) and finding the “power few” in relation to resource allocation, a Danish CHI can be used as a more informative metric when comparing changes over time in crime. Moreover, a Danish CHI can be used when setting up randomized control trials and comparing groups (treatment vs. control) to investigate if baseline equality between the groups has been met and also for evidence-based targeting of offenders and victims. Therefore, a Danish CHI has the potential to serve as a stepping stone to bring Danish Police one step closer to performing more evidence-based policing.
The data used to create a Danish CHI was a 6-year universe of the counts of all eligible crime types. The crime counts contain all reported crimes classified by the Danish Criminal Code from 2011 to 2016, plus selected types of offenses from “special law.” All reported offenses were registered, categorized and stored in POLSAS by the use of crime codes, and identified by a five-digit number. Offenses recorded prior to 2011 were categorized differently and were therefore excluded from the study.
Crime Counts and Crime Grouping
Crime data used in this research was extracted from Danish Police’s crime management system (POLSAS) via the data extraction tool POLMAP into Microsoft Excel. All reported crimes from 1 January 2011 to 31 December 2016 were extracted and converted into an Excel file. The data extracted on the 1 June 2017,Footnote 1 with all criminal categories in relation to the Criminal Code included. All criminal codes in relation to “special law” (such as all traffic offenses) were excluded, except reported crimes in relation to illegal firearms, illegal possession of knives and other weapons. After the data extraction, police-initiated offenses such as narcotics and handling of stolen goods were excluded.
In total, 504 criminal code numbers are included in the crime count. The total number of offenses each year is depicted in Fig. 1.
On the basis of these 504 criminal code numbers, 46 crime categories were constructed. The crime categories were constructed on the basis of the legal paragraph in the Criminal Code. All categories were reviewed by two prosecutors in order to ensure that all groupings were created according to the Criminal Code and that no categories overlapped each other. An overview of all categories can be found in Andersen (2018: Appendix A).
Fully 93% of all reported crimes from 2011 to 2016 are included in the grouped categories. The percentage of reported crimes that are not included varies very little from 2011 to 2016. This variation ranged little by year, from as high as 6.85% of offenses excluded to as low as 6.05% excluded (Andersen 2018: Table 1). A list of all crimes excluded can be found in Andersen (2018). Some offenses were excluded because they are very rare and are not considered serious crimes, such as “illegal begging” (11 reported offenses in 2016) or “unauthorized access to electronic material” (77 reported crimes in 2016). Others were excluded because the offense is police-generated (e.g., possession of stolen goods, narcotics possession, smuggling), following the principles of Sherman et al. (2016).
Moreover, some offenses were excluded after being grouped. This was the case for the categories blackmailing, counterfeiting, and embezzlement, for technical reasons detailed in Andersen (2018). Further, all terror-related offenses are excluded from the index, since they are low in frequency but can be high in harm. Since all terror-related offenses are handled by the Danish Security and Intelligence Service (DSIS), it is more useful for the purpose of using CHI to leave terror offenses out of the final index.
Data Limitations of Code-Grouping
The crime code number that is given to all reported crimes refers to the section of the Criminal Code the offense relates to, but the criminal code number is not identical to the legal paragraph in the criminal code. In addition, several criminal code numbers can be linked with the same legal paragraph. These complexities pose a great challenge in relation to the groupings of criminal code numbers. To make sure that the grouping of criminal code numbers corresponds with the legal paragraphs in the Danish Criminal Code, a prosecutor was assigned to assist with the grouping. A second prosecutor reviewed the groupings, and a few changes were made to the original grouping. However, some groups had to be combined even though they might differ slightly when it comes to sentencing. This issue is addressed in detail in Andersen (2018).
A 2015 Change in Rape Recording Practices
In the last 2 years of the data analyzed (2015–2016), a new procedure was used that increased the proportion of rape reports that was counted as reported crimes. In 2015, the Danish Police service was criticized for not recording reported rapes, but instead assigning the report with an investigation code. However, when a case is assigned with an investigation code number, the case will not be a part of the data set extracted via POLMAP. As a consequence of this critique, the Danish Police instructed all officers to use the criminal code number in rape cases from 2015 onward. When rapes were no longer assigned an investigation code number, but instead were assigned a criminal case number, the number of recorded rapes. However, this increase in reported rapes does not necessarily reflect an increase in rapes committed. When comparing numbers—and harm-scores—over time, this is a very important limitation to data. Thus, the analysis below reports the Danish CHI trends nationally over time both including and excluding the data on rapes of adult victims.
Methods: Creating the Danish Index Weightings
The methods used for this analysis followed the five principles described by Sherman et al. (2016). First, any crime harm metric should be democratic, reflecting a procedure for the resolution of conflicting viewpoints adopted by a democratic government accountable to the will of the people. Second, a crime harm metric must be reliable and be consistently applied to all units of analysis. Third, a crime harm metric must not be complex and has to be inexpensive to develop and use. Fourth, the metric has to be valid and measure the harm simply and as objectively as possible. Validity also means being as specific as possible in relation to particular crimes. Lastly, the crime harm metric must be easily operationalized.
In the absence of sentencing guidelines with “pure” starting points, this study planned to consult prosecutors from the 12 Danish police districts was considered. The reason for consulting prosecutors instead of judges (as in Sweden, see Rinaldo 2016) is that in a Danish context, judges are rarely specialized in criminal law. Because judges preside in both civil and criminal cases, years can pass between a judge being assigned a first and second serious crime, such as a rape case. Yet prosecutors are more specialized, working only in court in relation to criminal cases.
Consultations with prosecutors in a pilot test, however, revealed that they all consulted the guidelines from The Danish Director of Public Prosecutions (DPP) specifying what sentence the prosecutor should ask for in court when having a first-time offender charged without mitigation or aggravating factors (Anklagemyndigheden 2017). The prosecutor guidelines are available to the public as well as to the prosecutors and are available on almost all crime areas. The pilot testing revealed that, instead of rating the listed crimes based on estimates or experience, the prosecutors instead turned to the prosecutor guidelines. The prosecutor guidelines were therefore adopted to provide the crime harm metrics for the analysis, as measured by the number of days of recommended imprisonment.
The prosecutor guidelines specify what sentence a prosecutor should ask for in court when a first-time offender is charged with a crime. The DPP guidelines are only to be applied when no mitigating or aggravating factors are present and the offender is over 18 years old (Anklagemyndigheden 2017). These guidelines are not stored in one document, but are to be found in different documents in relation to different sections of the Criminal Code. However, in recent years many of the guidelines are combined in fewer documents, which increases their availability. The prosecutor guidelines are continuously assessed and updated (Anklagemyndigheden 2017).
Testing and Validation Procedure
Based on the prosecutor guidelines, a self-completion questionnaire was created (see Andersen 2018, Appendix C). For each offense, the prosecutor guideline value was assigned and each prosecutor was asked to evaluate if this value was equal to what a first-time offender would receive for the offense with no mitigating or aggravating factors. The questionnaire was set up by using Excel and information about how to fill out the questionnaire was presented in an attached Word document (Andersen 2018: Appendix D). The prosecutors could either “agree” with the value based on the prosecutor guidelines or “disagree.” If the prosecutors disagreed, they were asked to type in their own rating. Applying such an approach opens up for a risk of priming (Bryman 2004). Priming is a theory within psychology that takes into consideration how implicit memory affects responses; in this case, presenting the DPP value before asking the respondent about how he/she would rate the offense. It could be argued that due to the attention to the DPP value, the respondent will be more likely to agree with this value than if the respondent would have had to fill out the questionnaire without any set value. However, the pilot testing indicated that a such approach was not suitable, as the prosecutors would look up the DPP value anyway before answering the questions. It was therefore decided that despite the risk of priming, this was the best method available.
The prosecutors were asked to rate 43 crime categories. In order to minimize the prosecutors’ own interpretation of the crime categories, small written examples of each offense type were given, e.g., “rape, victim over 18 years old” or “possession of child pornography, limited level 2 and level 3 material.” These written examples served as a guidance to the prosecutors in order to rate an average offense with no mitigating or aggravating factors. All examples were written on the basis of a manual examination of cases in all crime categories and reviewed by prosecutors who were not part of the ratings.
Based on pilot testing performed by a former prosecutor, it was estimated that it would take around 1 h to fill out the questionnaire. The questionnaire was electronic and sent out to the participants via email.
The original plan to get one prosecutor from each of the 12 districts to fill out the questionnaire was not possible. Instead, only five prosecutors ended up filling out the questionnaire. All five prosecutors have more than 5 years of experience with the Danish Police. None of the five prosecutors comes from the same district and none of the prosecutors works in the same court.
All the participating prosecutors were contacted by email. Those who agreed to participate were telephoned by the first author for an in-depth explanation about the study and their role in the ratings. Further, it was possible to meet with four out of five prosecutors in order to discuss the study in person. The prosecutors were not given anything in return. The questionnaires were completed by the prosecutors themselves and then emailed back to the first author. On two occasions, the prosecutors were contacted afterward as some categories were not rated. The prosecutors were asked to fill out the remaining categories and once again thanked for their time invested in this project.
The prosecutors reported back that the estimate of time was accurate and that they spent a little under 1.5 h filling it out (all prosecutor ratings are displayed in Andersen 2018, Appendix F).
Testing for Inter-rater Reliability
In order to test how the ratings from the prosecutors differed from the prosecutor guidelines, a Cronbach’s alpha test was performed. The Cronbach’s alpha test is a reliability test for measuring the level of internal consistency (Cronbach 1951). The Cronbach’s alpha will generally increase when the correlations between the items increase. The maximum value for the Cronbach’s alpha is 1 and usually the minimum value is 0 (however, negative values can occur). Two tests were performed in STATA in order to test (a) the correlation between the ratings of the five prosecutors and (b) the correlation between the average of all five prosecutor ratings and the prosecutor guidelines. The correlation between the ratings of the five prosecutors was calculated to α = 0.78. This value indicates a good internal consistency between the prosecutor ratings; however, differences occur. The second test was calculated to test the correlation between the average of the five prosecutor ratings and the value based on the prosecutor guidelines. The Cronbach’s alpha value was calculated to α = 0.93 which indicated a very high level of agreement.
Converting Fines to Crime Harm Weightings
In some cases, e.g., minor theft, the offense is punished with a fine. In order to convert this fine to actual days in prison, the convertor depicted in Fig. 3 was used. This convertor is regulated in Danish Criminal Code, section §§ 50–55, and is reported in detail in Andersen (2018: Table 3).
Research Question A
Based on prosecutor guidelines from the DPP and sentencing ratings from five prosecutors in the Danish Police, a Danish CHI can be constructed, answering research question A. The Danish CHI includes 339,377 reported crimes in 2016 and the index covers 93% of all reported crimes according to the criminal code. Moreover, the index was applied to 6 years of crime data from 2011 to 2016, and more than 2,129,550 reported crimes were included. The reported crime data was grouped into its aggregate crime type and multiplied by the equivalent days’ imprisonment value (crime weight) derived from the DPP prosecutor guidelines and validated by prosecutor sentencing ratings.
Further, the crime types were grouped in broader categories in order to assess changes of crime patterns in Denmark. The composition of groupings and crime weights are discussed in the methods section of Andersen (2018). The grouping resulted in reducing the 504 offense codes in the Danish Criminal Code to 43 crime categories which easily can be applied to any set of Danish crime data and serve as an analytic tool for all 12 Danish police districts. Figure 2 displays the crime weights for all included categories of offense data.
Research Question B
How does a Danish CHI, when applied to Danish national police data for 2016, change the relative distribution of proportionate impact of different crime types compared to the raw number of each offense type? Figure 3 displays the total crime volume for selected and grouped offense data from 1 January 2016–31 December 2016. By comparison, Fig. 4 displays the results of applying the Danish CHI weights to the same crime data.
Theft and related offenses immediately stand out as presenting a much smaller percentage of harm in Fig. 4 (16%) than indicated by volume in Fig. 3 (51%). Moreover, the harm approach de-emphasizes the relative proportion of vandalism, comprising 1% of harm versus 7% of volume. Robbery presents higher harm (10%) than indicated by volume (1%), and also sexual assault increases from 1% when measured by volume, compared to 10% measured by harm. In total, crimes against persons stand out when harm is compared to volume, which is illustrated in Figs. 5 and 6.
The application of the Danish CHI to national crime data from 2016 changes the relative distribution of crime types. Looked at by volume, property crime constitutes 81% of Denmark’s total. However, in terms of the harm caused, that percentage is 53%. When applying the Danish CHI, crimes against persons becomes almost five times the proportion of total harm as it is of total volume. Police Chief Constables already know that these crimes are important, but the statistics make a precise and powerful case to do more for the higher harm crimes or areas, offenders or victims with such crimes.
Research Question C
If the Danish CHI is applied to annual national police crime data from to 2011 to 2016, is the decrease in crime count frequency paralleled by a decrease in Danish CHI value?
By focusing on crime frequency alone, the trend in reported crime volume is for all reported crimes, including those ineligible for the CHI, and is displayed in Fig. 7.
Figure 8 shows that excluding certain crimes to calculate the Danish CHI based on 93% of crimes does not challenge the downward trend. Almost the same linear trend can be found when looking only at the reported data included in the Danish CHI.
However, when applying the Danish CHI to the eligible national crime data from 2011 to 2016, the crime trend changes and reveals an increased crime harm as shown in Fig. 9. Note that the metric on the left axis is in the total number of days of imprisonment (or the equivalent in monetary fines) recommended by prosecutorial guidelines for all reported crimes eligible for calculating the Danish CHI.
Unlike the volume trend, the Danish crime harm trend is not linear and not steadily decreasing. While it initially decreased, it reverses in 2015 to rise to a harm level in 2016 that exceeds the harm level for all other years, despite the fact that there were 54,000 more reported crimes in 2011 than in 2016. The crime level has gone down from 2011 to 2016, but the data indicates that the crimes that are being committed are more harmful, especially from 2014 to 2016.
Considering the data, no single offense type or category can explain the increased harm level from 2014 to 2016. However, as outlined in the “A 2015 Change in Rape Recording Practices” section, Danish police changed the standard procedure for how to register reported rapes in mid-2015. As a consequence of this change, the number of rapes reported against persons over 15 years old has increased in 2015 and 2016, but the increase in reported rapes does not necessarily reflect an increase in rapes being committed. To increase the validity of this study, it is therefore important to know if the increased harm level can still be detected when the category rape against person over 15 years old is excluded from the data set. The robust evidence is illustrated in Fig. 10, showing again that the total recommended imprisonment across eligible offenses excluding rape rose from 9,714,057 days in 2011 (after falling to 8,535,745 days in 2014) to 9,760,697 days in 2016.
Splitting the data into the categories (a) crimes against person and (b) property crime also shows a trend of increased harm in both crimes against persons (including rape) and property crime becomes obvious (illustrated in Figs. 11 and 12). In this analysis, however, we see that for crimes against persons, crime volume has risen as well as crime harm (Fig. 12). While that fact alone could have signaled increasing harm overall, that fact was clearly swamped by the much greater volume of property crime that dropped substantially in 2011–2016.
Considering the data presented in Fig. 11, the harm associated with property crime has risen from 2011 to 2016, despite the fact that the frequency has decreased during this time period. Looking closely at the data, the decrease in frequency can be explained by a significant drop in reported burglary.Footnote 2 However, in terms of frequency, the category “fraud” has risen from 10,748 in 2011 to 32,802 in 2016. Even though fraud does not have a harm weight more than 60 days of imprisonment, the increase in reported crimes causes large effects on the harm score.
According to the data, “homicide,” “sexual offenses,” and “rape” are all categories where the reported crime has increased in both frequency and harm from 2011 to 2016. Given the changes in how to record rape occurred in 2015, that category was excluded in a separate analysis. That procedure showed that even after excluding the crime category rape, crimes against persons still generated an increase in relation to harm in the period observed.
The fact that the increased harm score from 2011 to 2016 cannot be explained by a single crime category emphasizes the value of a Danish CHI as a useful tool for Danish Police. These trends would not have been detected if only traditional crime counts were available.