Measurement errors – a challenge in police crime statistics
The volume of crimes in the statistics, i.e. the police recorded criminality, may be influenced by the volume of control, among other factors. If in a certain area the volume of a particular type of criminality changes, this is not necessarily a consequence of a corresponding change in the underlying factors of such crime.
The police becomes aware of a small and varying part of crime (with the exception of homicides). There may be many reasons for the police not becoming aware of an offence, such as the efficiency of control as well as the propensity of the victims – or people in general – to report crime to the police. This propensity to report has sometimes been described with the term tolerance limit. The limit varies both by category of crime and in time. Today, the propensity to report – and its increase rather than its decrease – is influenced by factors such as crime news and public attention awarded to individual cases, the social media, increased self-centredness and the more marked emphasis of criminal law in society. People are more susceptible to report various events as crime to the police when they feel that their objects of legal protection have been violated. It may also be the contrary – that the crime passes unnoticed. This may also take place in information network crimes or negligent traffic violations where not even the perpetrator is aware of a crime being committed.
If the authorities do not register, for a reason or other, a case reported to them, it remains hidden criminality. A crime may sometimes be registered but the perpetrator not – if the culpable party is never detected. The reported crime is a registered offence but the perpetrator remains hidden. The share of hidden criminality never entering the statistic may constitute even a considerable volume in so-called mass criminality, such as in usual violent and property crimes. Increased urbanisation, changes in people’s tolerance and resources of the authorities influence the share of registered crime of the entire crime volumes. For example, changes in traffic enforcement have a strong correlation with the volume of traffic offences in the statistics. Also, international studies have shown that the everyday recording practices of the police are heterogenous. There may be geographical differences within and between police departments.
As is well known, the information of criminality registered in police statistics is susceptible to various forms of measurement errors, even consistent. Irrespective of this, the data on the police-registered crimes is widely used in various types of statistical modelling in research that aims at understanding the reasons and consequences of criminality. However, the results may be biased in many different ways.
Indeed, in criminology the nature of research materials makes it different from many other disciplines. For a long time in the 19th century, criminology was linked to official statistics, i.e., detected criminal cases. However, the criminologists were aware of this challenging control barrier in measuring criminality.
Complex mechanisms of measurement errors
The mechanisms affecting the measurements of crime detected by the police are complex. At the very least, they are systematic (too small volumes). They can also be so-called multiplicative (the mistakes increase in proportion to the increase in level of criminality) but the non-reporting of crimes may also be a consequence of some other regional aspect, such as economic inequality. A large number of studies show that several underlying factors of victims affect their propensity to report the events, wrongdoings and crime they have been subjected to, and such links are valid not only in case of individuals but also for geographic areas.
The research information on factual overall volume of criminality are normally based on the methods of so-called total crime rate research. In this research, criminal behaviour is measured directly, for example, in questionnaires, without filters generated by a control system. In Finland, youth delinguency survey and national crime victim survey conducted by the Institute of Criminology and Legal Policy at the University of Helsinki are examples of research on total crime rate.
The control barrier was broken almost a century ago when the criminologists introduced the questionnaire method in crime research. For example, when people are asked anonymously whether they have committed crimes or become victims of crime during a certain period of time, they may tend to give a truthful answer, the young and children, in particular.
Crime victim surveys are normally used as the more accurate indicators of criminality, to complement the measures of criminality detected by the police, but they are often based on relatively small samples in proportion to the whole population, in other words, the results obtained are not necessarily very representative. The observations do not allow for conclusion with more regional precision, but only at the national level.
Accumulating errors of measurement are typical in data describing counts and durations where the observations are easily “right-skewed,” i.e., the observations do not present a normal distributions but the majority of the observations lean to the left and the narrow “tail” of the distribution extends to the right. A general and typical example are the distributions of observations related to income distribution.
The form of measurement errors perhaps most prominent in police statistics is the systematic undervaluation of crime volumes. This is caused, for example, either by the fact that victims and the general public tend not always report crime to the police, or that they do not recognise that crime has taken place. One background factor may also be the fact that the crime has no actual victim.
Crime statistics may cause substantial harm to research
For decades, if not for centuries, it has been known that official crime statistics do not offer a precise depiction of the true volume of crime. Hidden criminality is consistent, and the practices and processes of recording crime are widely non-uniform. Research based on the police crime statistics can provide very biased results.
Measurement error challenges are not related to police statistics only but also to statistical data in epidemiology, psychology or economy. The various adjustment or correction methods of these disciplines to address measurement errors are, however, not directly applicable to police crime statistics because they are affected by diverse, even less usual measurement error mechanisms.
The text is based on the freely accessible international research paper: The developing and progressive statistical software R includes so-called packages or extension which can be used for versatile analyses. Jose Pina-Sánchez with her colleagues in England has developed a package containing a user-friendly tool for sensitivity analyses which can be used to assess the impacts of measurement errors in various statistical models based on data on crime rates from police statistics.
Mika Sutela
Information Analyst
National Police Board
X: @SutelaMika
