«FAMILIES OF NATIONS, VICTIMISATION AND ATTITUDES TOWARDS CRIMINAL JUSTICE PAUL A. NORRIS* School of Social and Political Science, University of ...»
In previous analysis the United Kingdom is typically treated as a single entity (Arts and Gelissen, 2002: pp. 150–151). This reflects how, until recently, England and Wales, Scotland and Northern Ireland were, in most areas of public policy, subject to a unified legislative process. This is in contrast to criminal justice where marked differences can be seen between legal systems, particularly with regards to Scotland (McAra, 2005; Jones and Christie, 2003).
In view of the long-standing differences in legal systems across the United Kingdom, it is by no means certain that it can be considered a single case when attempting to classify nations with regards to criminal justice issues. Indeed, the foundation of the Families of Nations typology, that modern day outcomes are shaped by historical and cultural context, suggests that variations in legal system may well cause differences between nations in terms of modern-day victimisation and justice related attitudes. In view of this, England and Wales, Scotland and Northern Ireland are treated as separate cases in the subsequent analysis.
Although the ICVS is clearly a valuable data source for comparative criminology, it does have limitations (a good review of the strengths and weaknesses of victim surveys, and the ICVS in particular, is provided by Lynch, 2006). Perhaps the most important issue affecting the use of ICVS data in this research is the relatively small sample size which is employed within each jurisdiction. During the 2004/05 survey, the sample size was typically around 2000 respondents per nation. This sample size could be considered small when compared to the samples in most national crime surveys, and this may affect the reliability of national level estimates. This problem may be particularly acute when trying to estimate the prevalence of events which are not particularly common, for instant many types of violent crime.
In order to try and address the above concern, where a jurisdiction has taken part in either the 1996 or 2000 ICVS as well as the 2004/05 survey estimates will be averaged across all the sweeps in which that nation has participated.
Table 2 provides details of which ICVS sweeps each jurisdiction has taken part in. As not every country has appeared in every sweep, it could be argued that averaging estimates across different time-points could introduce bias to the data. However, a key point of the Families of Nations argument is that the clusters identified represent long-term differences between nations. Therefore, averaging estimates across more than one survey would appear a valid approach because the resultant indicators are more likely to represent the general nature of a jurisdiction, and are less likely to be affected by the results of one unusual set of responses.
Table 3 provides details of the different indicators which will be used in this analysis to represent experiences of victimisation and attitudes towards crime and justice. To try and reduce the possibility that any groupings identified are unduly influenced by any single variable, wherever possible two indicators were used to measure each concept under consideration. For instance, the level of property crime within a country is captured by the aggregate victimisation rate for both burglary and attempted burglary. Similarly, both the incidence rates for robbery and assaults and threats are included as indicators of personal crime.
In addition to the level of victimisation within a country, the dataset includes two indicators of the extent to which respondents worry about crime. Although its exact interpretation is open to question, the percentage of the population who 'feel unsafe when out after dark' in their local area is commonly seen as representing the extent to which people worry about personal, or violent crime (Van Dijk et al., 2007: pp. 130–133). The level of worry about burglary should provide an indication as to the level of concern about property crime. Measures concerning whether people install alarm systems or special locks at their home may also be, to some extent, related to the level of concern they have about victimisation. However, it may also be that these indicators provide a measure of whether individuals feel it is necessary to take private precautions to protect themselves from crime. The question concerning the appropriate sentence for a repeat burglar is often used as an indicator of the extent of punitiveness expressed by a population (Mayhew and Van Kesteren, 2002; Van Dijk, 2008).
Finally, the extent to which a population believes the police do a good job in controlling local crime is included as a proxy of current satisfaction with public efforts to address criminality.
The extent to which the variables listed in Table 3 accurately reflect the complex nature of a population's attitude towards crime and justice, or experience of crime is open to question. For instance does a question concerning feelings of safety when out after dark capture a respondent's fear of crime or, TABLE 2 Participation in ICVS Surveys 1996–2004/05
(n = 23) may their response, also reflect wider concerns with insecurity? It is also possible to argue that several issues are missing from the topics covered in Table
3. For instance, the scope of the ICVS means that this analysis does not consider experience of drug crime.
HIERARCHICAL CLUSTER ANALYSIS
Obinger and Wagschal, 2001; Castles and Obinger, 2008). The technique aims to take information about a group of cases (listed in Table 2) measured across a range of dimensions (listed in Table 3) and to group together cases which exhibit similar characteristics while ensuring the maximum amount of heterogeneity between the different clusters identified.
There are numerous different approaches to cluster analysis. These techniques vary in terms of the criteria they use to decide how similar or dissimilar cases are, whether or not they require the researcher to specify the number of clusters they wish to find and the extent to which they allow a case to be a partial member of several clusters (Everitt et al., 2001). The analysis in this paper follows the lead of much of the previous research around welfare regimes by using a hierarchical approach to clustering.
Hierarchical clustering starts with each case being seen as representing a unique cluster. The 'most similar' clusters are then joined together to create a new aggregate cluster. This process is repeated until all cases are joined in a single cluster. Hierarchical methods are particularly suitable for testing the fit between a typology and a dataset because their results are not dependent on the researcher's pre-conceived ideas about how many clusters should be present within the data.
Cluster analysis does not have the strict assumptions of many statistical methods in terms of linearity, homoscedasicity and multicollinearity. However, if each variable involved in the analysis is to have an equal impact on the decision about which cluster a case should be placed in then it is important that all the variables considered are measured using comparable scales. If different variables are measured using different metrics then those indicators which involve larger absolute values may have a disproportionate impact on the results presented. For instance, in this analysis the indicators concerning perceptions of the police and attitudes to punishment would have a greater influence than those concerning victimisation rates. To address this all variables were subject to a z-score transformation prior to the analysis.
Cluster analysis results can vary depending on which algorithm is used to decide which cases or clusters are 'most similar' and should be joined together in the next stage of aggregation. The analysis in this analysis was conducted using the Ward Method, with Squared Euclidean distances employed as the measure of similarity between cases. These choices reflect those which have often been employed in the social policy literature (for instance Obinger and Wagschal, 2001; Castles and Obinger, 2008).
RESULTSDoes the Families of Nations Typology Fit Criminal Justice Attitudes and Victimisation?
Analysing the variables described in Table 3 using hierarchical cluster analysis produces the dendrogram shown in Figure 1. A dendrogram provides a graphical representation of the extent of similarity and difference between cases and suggests which cases may be grouped together to create homogeneous clusters. The shorter the length of the line linking together two cases, the more similar they are. For instance, the USA would appear to be very similar to Canada, and Northern Ireland very similar to Scotland, while the group of English-Speaking nations (to the left of Figure 1) would appear highly distinct from the rest of the jurisdictions considered.
The 'English-Speaking' cluster in Figure 1 provides the first, and perhaps most clear-cut, evidence that the groupings identified in the Families of Nations Typology could be related to victimisation levels and attitudes towards criminal justice. Figure 2 suggests how those jurisdictions which make-up the 'English-Speaking' cluster compare to the sample as a whole. The dashed line shows the mean score for each variable across the whole sample, while the solid line indicates the mean score for the 'English-Speaking' jurisdictions. Figure 2 suggests that 'English-Speaking' jurisdictions have higher than average victimisation rates, populations who hold punitive attitudes but are generally happy with the police, and are home to individuals who are likely to invest in protection such as burglar alarms and additional household locks. Despite their apparently increased risk of victimisation, respondents within these jurisdictions do not appear to exhibit an increased fear of crime. The finding that these jurisdictions are characterised by 'poor' outcomes with regards to crime and justice fits with a wider pattern concerning social policy outcomes. This adds weight to the idea that the distinctive cultural, social and political context present within these nations may combine to cause less desirable social outcomes, the so called 'awfulness of the English hypothesis' (Castles and Merrill, 1989; Castles, 1993 and forthcoming).
The characteristics of the 'English-Speaking' cluster add weight to the argument that those countries which appear to have the least generous and inclusive welfare regimes experience less favourable outcomes in the criminal justice sphere. The finding that these jurisdictions are associated with punitive attitudes towards punishment would appear to reflect the findings of Cavadino and Dignan (2006) when discussing penal policy more generally. Combined punitive attitudes with the high levels of victimisation associated with this cluster could be seen as providing some evidence to support the argument that there is something criminogenic about Neo-Liberal welfare regimes. However, as argued above, the discovery that there is a strong overlap between Neo-Liberal approaches to welfare and punitive attitudes or high levels of Figure 1. Dendrogram Produced Through Hierarchal Cluster Analysis of ICVS Indicators of Victimisation and Attitudes Towards Criminal Justice Figure 2. Attributes of the English-Speaking Cluster victimisation does not provide conclusive evidence that differences in criminal justice outcomes are directly attributable to differences in welfare provision. It is also possible that the apparent tie-up between crime, justice and welfare can be attributed to how both areas reflect general attitudes or characteristics of society. For instance, the strong focus on individual self-sufficiency could provide one explanation why these jurisdictions appear to be associated with individuals who take greater private precautions against the risk of victimisation. That it is not possible to identify definitively which causal mechanisms explain the unfavourable characteristics of the 'English-Speaking' cluster illustrates how while descriptive typologies may be a starting point for understanding and further investigation, they are unlikely to provide complete explanation.