|Recorded assault offences 1950-1997||Graphical version|
Recorded assault offences 1950-1997
Changing Structure and Influencing Factors of Assault Criminality from the 1950s to the 1990s
Helsinki 10 May 2000
This report provides an analysis of changes in the structure and trends in assault offences over the last 45 years; the research based on police-recorded assault statistics and other social statistical time series. The study is divided into two parts. The basic features of criminal assaults in light of the police statistics are described in the first section. Also, trends in official assault statistics are compared with data from victim surveys and statistics on other violent crime. The focus of the second part in on the analysis of factors that may influence the level of assaults over time. Among such factors are: the level of alcohol consumption, migration trends, and economic development.
1 Changes in the volume of assault offences
Looking at changes in the volume of police-recorded assault offences in the time span 1950-1997, two main periods may be discerned (figure 1). From 1950 to about the mid-1960s, the volume of assaults fluctuated annually very little, and the volume of assaults either remained stable or fell. Towards the end of the 1960s, assaults began to increase, and since this time, the growth has been stable with minor exceptions. However victim surveys that have been carried out since 1980 give a different picture of the trend in violent crime. The development since the early 1980s look very different depending on whether one looks at police statistics or victim surveys. According to police data, the volume of assault offences has grown steeply, whereas interview surveys indicate a decreasing trend or a stagnated situation. This observation holds even if the analysis is restricted to relatively serious violence only. One explana-tion of this conflict is that, although violent events have not increased, a growing proportion of them is coming to the attention of the police. Indeed, the victim surveys have shown that reporting violence has increased steadily since 1980 (table 1). However, the matter is not quite that simple. Also the most aggravated part of violence, i.e. attempted and completed homicides have increased over the last decades in a manner similar to recorded assault offences. The reportability of these offences has been high during the entire time span and the police statis-tics reflect their volume quite accurately. Considering that the most aggravated part of violence is not readily (or at all) covered in interview surveys; also the growth of recorded assault offences may be in part real. The conflicting trends in police-recorded assaults and those mea-sured by victim surveys may be interpreted as an indication of a different phenomenon: the population has become increasingly polarised with regard to experiencing violence. The violence experienced by the domi-nant part of the population has decreased in a manner revealed by the victim surveys, but among those who have been pushed to the margins of society or who otherwise deviate from the way of life of the dominant part of the population, such as alcoholised people or narcotics, life has become more violent than earlier.
Parallel to the increase in reporting of violence, another trend has been the slowly decreasing clearance rate of assaults since the late 1960s (figure 5). A possible explanation to this is that the volume of assaults may have exceeded the critical threshold with regard to the investigation resources of the police. Therefore, the resources have not been sufficient for the investigation of "surplus" crime. A counter-argument to this point is, however, that the clearance rate of assault offences has been decreasing already for more than twenty years. While the lack of police resources may, in the short term, be manifested by a decrease in the clearance rate, it is unlikely that the situation would not be corrected in the long run. Actually, it is more likely that the decrease in the clearance rate stems from a change in the character of assault offences that has made the investigation/clearing-up of these crimes more difficult. The generally high clearance rate of assaults is often consequence of the fact that the victim either known or is able to identify the assailant and that there are witnesses to the assault. The falling clearance rate may be considered as an indication that assaults are occurring increasingly in circumstances where the identity of the assail-ant is unknown. Typical examples of such circumstances are urban street violence and conflicts in and around bars and pubs that escalate into violence. A number of indicators, such as growth in licensed establish-ments (pubs and bars) in the aftermath of the 1969 alcohol law reform, suggest that a significant change took place in the situational risk associ-ated with assaults in public settings, increasing the criminal opportuni-ties. As indicated by figures 9 and 11 and appendix figure 11, this trend has persisted throughout the period of interest. A characteristic change in assault offences seems to have been a relative increase of assaults between strangers in public places, typical to urban environments. Also the increase in reporting of assaults, revealed by victimisation surveys points in the same direction. The hypothesis is that an increasing propor-tion of violent encounters takes place in circumstances where it is more difficult to avoid the detection and recording of the offences, such as public and semi-public environments. The result is that the police are faced with an increasing proportion of assaults that are difficult to clear. The social context of this change in the structure of assaults is the con-centration trend of the population due to migration to urban centres, which increases the likelihood of spontaneous violent encounters be-tween strangers.
There are also two features in recorded crime that supports the idea of a changing nature of assault offences. First, the proportion of incidents where the assailant has been under the influence of alcohol has grown steadily (figure 6). This observation fits the interpretation pre-sented above. The second feature concerns the gender balance among the assailants (figure 7). In the 1950s and the 1960s, when the volume of assault offences remained constant for almost two decades, the propor-tion of women among all assailants decreased. In the early 1970s, how-ever, the balance changed as the assault figures began to grow. Simulta-neously with the growth of assault offences, the proportion of women among the assailants has grown. The growth in the proportion of women already by itself means a change in the structure of the assaults. This may also be connected with a deeper structural change if the typical environments of assaults by women are systematically different from the environments typical of male violence.
2 Factors influencing the volume of assault offences
Alcohol plays a significant role in assault offences in a number of dis-tinct ways. In recent years, about three out of four persons suspected of assaults have been intoxicated at the time of the offence. Also, being under the influence of alcohol seems to increase the likelihood of becom-ing a victim of violence. Victim surveys indicate that violent incidents have undergone a systematic and therefore notable change in those violent events where the victim was drunk has become more frequent (table 3). By contrast, in all four waves of the Finnish victim survey (in 1980, 1988, 1993 and 1997), the proportion of intoxicated assailants was nearly constant. The change regarding the victim's intoxication concerns both men and women. Among women, however, the relative change is larger. The significance of this finding is accentuated by the fact that it is based on information given by the victims themselves. Also, the overall frequency of alcohol consumption as well as the frequency of getting drunk is connected with the likelihood of violent victimisation. The findings are unambiguous. The involvement in violent events is systematically more likely the more frequently the person uses alcohol. However, also the stronger connection is valid. Persons who consume alcohol frequently were intoxicated more often than others were also when they became victims of violence (table 4). The finding holds for both for men and for women, and it was confirmed by both the 1980 and the 1988 victimisation surveys 1).
1) Due to a small sample, the analysis was not repeated for 1993 data.
Thus, there is convincing evidence as to the connection of alco-hol and violence at the individual level. However, it is not self-evident that the same connection would exist also on the macro level. The influence of the increase of alcohol consumption on the volume of assault offences depends on several factors: the distribution of the con-sumption increase in the population, the type of environments the in-creased consumption takes place, and the pattern of consumption. If the increase in the total volume of alcohol consumption increases incidents of intoxication, and if intoxication in certain circumstances increases the likelihood of violence, then a statistical relationship between the volume of alcohol consumption and the volume of assaults may be expected. The time series analysis concerning the years 1950-1997 establishes indeed a statistical correlation between changes of private as well as public consumption 2) of alcohol with changes in the volume of assault offences (figure 10). The connection of public consumption to assault offences is somewhat stronger than the one of private consumption. Regression models fitted to the data show, however, that the relative "effects" of both types of alcohol consumption are nearly identical. Crudely, a 10 per cent increase in public consumption corresponds to a growth of about 5-7 per cent in the volume of assault offences in the same year. The corresponding estimated effect of private consumption is 4-9 per cent (Models 5 and 6)3) . Changes in alcohol consumption do not seem to have lagged influences on assault offences: only the synchronic correlation between the time series is statistically significant. The coefficients of the alcohol variables in the estimated models differ significantly from zero. However, considering the standard errors of coefficients, changes in the volume of assault offences cannot be predicted very accurately on the basis of changes in the volume of alcohol consumption alone.
2) Private consumption=consumption of alcoholic beverages through retail outlets.) Public consumption=consumption of alcoholic beverages through licensed serving.
3) Models are estimated by both Ordinary Least Squares (ols) and a robust method of estimation. As a robust high breakdown method, the method of Least Median of Squares (lms) has been used (see Rousseeuw 1984, Rousseeuw & Leroy 1987). Lms estimation using Rousseeuws algorithm has been done by a module programmed by the author, integrated in SURVO 84 statistical system (about SURVO system, see Mustonen 1992). Lms+ols means a two stage procedure. The model is first estimated by lms and standardized lms residuals are calculated. In the second phase the model is re-estimated using reweighted ols. The weight=0 is assigned to five most influential or outlying observations (about 10 % of total data) according to standardized lms residuals.
The development of the volume of assault offences also follows certain other indicators that are connected with the consumption of alcohol, such as the number of restaurants licensed to sell alcoholic beverages, and of pubs licensed to serve beer (figure 11). The availabil-ity of alcohol increases alcohol consumption; the number of licensed restaurants and pubs in particular, public consumption. Independently of this connection, the volume of licensed restaurants and pubs reflects the volume of physical opportunities of assault offences, the number of potential crime settings.
The volume of assault offences is also connected with changes in the economy. As indicators of this dimension, the analysis takes up the gross national product, private consumption expenditure, and the industrial production of consumer goods. Economic cycles and the volume of assault offences fluctuate in a similar manner (figure 12). In times of economic growth, assaults have also often increased; and when growth has stagnated, so has the level of crime. This took place, for instance, in connection with the economic crisis of the mid-1970s. In 1976, the GNP diminished in Finland for the first time after World War II. In the same year, recorded assault offences decreased more than in any other year over the period 1950-1997. Also the recession of the early 1990s - which for Finland was exceptionally deep - was reflected in the volume of assault offences, turning to a decrease (figures 13 and 14).
One explanation for the observed connection between economic development and assault offences could be that when there are more resources to be spent for consumption, more people will spend time in leisure activities and situations that are associated with higher risks of violent interaction. GNP and private consumption expenditure may also be thought as indicators of societal development in general, characterised by urbanisation, changes in the way of life, etc. On the other hand, it may be argued that the connection of economic indicators with the volume of assaults is brought about specifically via alcohol consumption, and would therefore be spurious. This is, however, not the fact. The gross national product as well as private consumption expenditure corre-late significantly with changes in the volume of assault offences also after controlling for the effect of alcohol consumption. The estimation of the unique effects is, however, hampered by the collinearity between the economic indicators and the alcohol indicators. Therefore, it is not possible to make an exact distinction between the relative weights of economic development and alcohol consumption.
Migration and unemployment
In the present setting, migration may be connected with crime in at least two ways. Migration means a movement of the population from low-crime (rural) areas to high-crime (urban) areas, resulting in an overall increase in per capita crime. Immigration transfers people into environ-ments where the risk of violence is higher than in the areas of emigra-tion. Here we are dealing with characteristics of mobility that increase criminal opportunities. Migration may also weaken social cohesion. It may be assumed that this concerns in particular young people without families moving from rural to urban areas. Also the cultural differences between the countryside and the city may contribute to the emergence of conflicts resulting in violence. In this respect, migration is also a factor influencing the motivational element relevant in offending.
Both of these aspects of migration are reflected by variations in the mobility of young men (15-34 years) from rural to urban areas (figure 17). The statistical analysis shows that, in the time period 1967-1996 (for which time series data are available) migration is indeed connected with the volume of assault offences (figure 18). For the entire period under analysis (1950-1996), the volume of migration has been approximated with data on migration between cities. The connection of this indicator with changes in the volume of assault offences is positive but weak (figures 19 and 20).
The effect of unemployment on crime is a controversial topic. Despite the fact that the causal mechanisms (both crime-increasing and crime-decreasing) are typically not defined very clearly, it is generally concluded that unemployment rate has a negative net influence on crime rate. As unemployment raises, so does crime. However, in the time series analysis, there is no sign of that a growth in unemployment would be connected with a growth in assaults. Rather the opposite is true: the unemployment rate and changes in the volume of assault are inversely related (figures 21 and 22). This may be explained by the fact that the unemployment rate is connected with economic cycles and economic growth was, on the other hand, found to be a factor increasing assault offences. If also alcohol consumption and the gross national product (or private consumption expenditure) is included in the regression model explaining changes in the volume of assault offences, migration and the unemployment rate have no additional explanatory power at all.
The volume of assault offences and the clearance rate of the offences display contrasting trends - one is growing, the other is decreasing (figure 5). This does not necessarily ilmply that the clearance rate influ-ences the volume of assault offences or vice versa. Contrasting trends may be explained by seriousness, i.e., the presence of a background factor common to both, such as the "urbanisation" of assaults. But is there a statistical connection between short-term changes in the volume of assaults and the clearance rate that would support the idea that one would influence the other? We already mentioned the possibility that, as a consequence of a sharp increase in assaults reported to the police, the clearance rate may drop with the dilution of investigative resources of the police. Correspondingly, if the volume of crimes decreases, the clearance rate may be expected to grow. Another interpretation has been to understand the clearance rate as in indicator of the likelihood of apprehension, the efficiency of crime control, and of the certainty of sanctions. This interpretation leads to the argument that an increase in the clearance rate decreases (with a short time lag) the volume of crime, and vice versa.
The statistical analysis indicates that there is indeed an inverse relationship between synchronic changes of the clearance rate and of the volume of assault offences (this becomes evident after two most influen-tial and outlying observations are excluded). Additionally, changes in the clearance rate are also connected with the volume of assaults with a one-year lag (table 7 and figure 23). The synchronic correlation supports the resource hypothesis, but does not exclude the interpretation that a falling clearance rate increases the volume of assaults (or that an increased efficiency in clearance work prevents assaults). It is more problematic to explain the remaining connection: there is also a lagged inverse correlation between the clearance rate and the volume of assault offences. One interpretation is that the correlation reflects a deterioration of the mecha-nisms preventing violence. If a decrease in the clearance rate reflects a change in the structure of assaults, as was argued earlier, then the change in the clearance rate may also be interpreted to represent a change in the control of public places, and in this sense, also the risk of apprehension. This hypothesis is valid for all forms of control, whether the formal control carried out by police or by private security guards, or the informal control represented by private citizens. Environments with weak control are lacking one element of crime prevention, which may lead to an increase in opportunities for assaults. A self-perpetuating circle may ensue: people who would be able to prevent assaults by their presence begin to avoid disordely places. As a consequence, informal natural control deteriorates even further. Over the last decades, a devel-opment of this character may have taken place with regard to the control of public places.
The possible connections between changes in the volume of assault offences and short term changes in the sanctions were analysed in light of three indicators: the proportion of imprisonment sentences of all sentences, the proportion of unconditional sentences of all prison sen-tences, and the average length of prison sentences. The relative usage of imprisonment decreased over the 1950s and the 1960s. This did not have a visible influence on the volume of assault offences, which remained practically stable up to the year 1967 (figure 24). It is only in the second half of the 1970s that the usage of imprisonment and the volume of assaults have begun to trend into opposite directions. In fact, there is only a weak connection between changes in the volume of imprisonment and the volume of assault offences, and neither the synchronic nor the lagged correlations differ significantly from zero (figure 25 and table 8).
The usage of unconditional prison sentences in assault offences has decreased in a linear fashion over the entire period of analysis (figure 26). Thus, the trends of this indicator and of the volume of assault offences are opposite to each other. However, no correlation is found between the annual changes in the two time series. Neither are there any lagged connections between the usage of unconditional prison sentences and the volume of assault offences (figure 27 and table 8).
Over the entire period of analysis, the value of the third punish-ment indicator, the average length of prison sentences, has been nearly halved (figure 28). Although the punishment level - as measured by the average length of prison sentences - and the volume of assault offences have changed in opposite directions, there is no statistical connection between the annual changes in the corresponding time series (figure 29 and table 8). Overall, the indicators of the punishment level are problem-atic because the statistical reporting of sentences has undergone several changes in the time period covered by this study.
3 Predicting crime trends - the best fitting statistical determinants of assault volume
This report presents an analysis of the links between the volume of assault offences and alcohol consumption, economic cycles, mobility, the crime clearance rate and the punishment level. As a starting point, all of these factors may be considered to have an effect on the volume of assaults. As the above analysis show, public consumption of alcohol, the gross national product, private consumption expenditure, migration, the unemployment rate, and the clearance rate of assaults are correlated with changes in the volume of assaults over the period 1950-1997 (1996). Taken separately, each of these factors thus "explains" trends in the volume of assaults. The simultaneous observation of several explanatory factors results in a problem in the estimation of models, since some of the factors are strongly intercorrelated, i.e. they are connected with the same underlying dimension. Their explanatory power may change decisively if another variable is added to the model. In a statistical sense, the best combination of explanatory factors, and in this sense, the best predictors of changes in the volume of assault offence may, in any case, be selected mechanically, by stepwise regression analysis. In this case, backward elimination was used, where the best or sufficient explanatory combination is determined by removing, one by one, the variable with the weakest and statistically non-significant (p=<0.10) explanatory power.
In the stepwise estimation, it turns out that of all the variables included in the model, the unemployment rate, migration, and the three variables representing the punishment level have no significant explana-tory power. These variables are dropped from the model at different stages of estimation. The outcome is that, out of the variables in the analysis, a sufficient explanation to changes in the volume of assaults is provided by the public consumption of alcohol, the gross national prod-uct, and the clearance rate of assaults. Of these, the clearance rate has the weakest explanatory power but is nevertheless accepted in the final model (p=0.08). The robust estimation (Lms+ols) or OLS after the most influential observations 1969, 1991 and 1996 (largest Cook dis-tances), are excluded, yields models with satisfactory explanatory power (see footnote 3).
4) ASSAULTS=Assaults recorded by the police per 100 000 in population
ALCO=Public consumption of alcohol, litres of absolute alcohol per 100 000 in population over 15
GNP=Gross national product, volume index (1926=100)
CLEAR=Clearance rate of assaults, crimes cleared by the police * 100/crimes recorded by the police
D=Differenced series, LN=log transformation
Expressed in terms of relative change (exponential form, model 12), the model predicts that, on average, a 0,1 litre increase in public consump-tion of alcohol per capita results in an assault increase of 3 per cent. An increase in the volume of gross national product by 10 index points is followed by one per cent increase in the volume of assaults. A one percentage point change in the clearance rate is connected with a one per cent change in the assault volume to opposite direction.
As an application of the fitted model we may consider the year 1998. In 1998, the public consumption of alcohol increased in per capita terms from 1997 by 0,04 litres (or 1,8 per cent), the volume index of gross national product grew by 52,3 units, and the clearance rate of assaults increased by 2,1 percentage points. We may observe that eco-nomic growth has the greatest contribution to the predicted volume of assaults. According to the estimation results, the growth of GNP corre-sponds to a 5,2 percent increase in the volume of assaults in 1998, while the contribution of public alcohol consumption is +1,2, and the contri-bution of clearance rate is !2,1 percent.
The statistical models describe the ways in which changes in the volume of assaults are statistically connected with other time series in the analysis. However, it is clear that the volume of assaults is also deter-mined by qualitative factors connected with opportunities for crimes and criminal motivation that can hardly be compressed into statistical time series or subjected to quantitative measurement. Even many quantitative indicators lack retrospective data that would be necessary for sufficiently long time series analysis. In this sense, the "explanation" provided by a statistical model is always limited. In the post-war era, Finland has experienced large social and demographic changes that can explain long-term trends in crime even if the connections cannot be demonstrated quantitatively with a statistical model. The "Great Migration" of the 1960s, and the rapid urbanisation of Finnish society represent the type of factors that have obviously increased the opportunities for crime, includ-ing crimes of violence. On the other hand, cultural features (that are difficult to quantify) may explain fluctuations in criminal motivation, changes in the dominant attitudes, etc.
Statistical models (or their estimation) are also time-specific. In a strict sense, the results of the estimation only refer to the time period analysed. It is, for instance, quite clearly not reasonable to use the gross national product or private consumption expenditure for the prediction of assault trends very far into the future. It is hard to accept that the assault trend could continue its growth path without limits at the same pace with economic growth. Over the period of analysis there obtains a statistical connection between the volume of assaults and the economic develop-ment, but there is no certainty for this to hold in the future. The ob-served statistical connection belongs to a given historical period of societal development. Also, the connection between alcohol consumption and the volume of assaults may change over time. It is, however, possible that alcohol has a more permanent influence on changes in the vol-ume of assaults as compared with economic factors. The connection of alcohol consumption to fluctuations of assaults is probably historically more stable than the dependence of assaults on the economic development.