Socioeconomic differences in violent victimization: Exploring the impact of data source and the inclusivity of the violence concept

Abstract

Research on socioeconomic differences in violent victimization has relied on surveys. Nationally representative register-based data sets, increasingly used in Nordic criminology, have not been used in such research. We analyse socioeconomic differences in violent victimization in Finland using both survey and register-based data, and assess whether these differences vary by severity of violence. The results show that the data source and the inclusivity of the definition of violence affect the observed socioeconomic differences, with differentials being larger for more severe violence in both data sets and in register rather than in survey data. We conclude that the link between socioeconomic status and victimization is unquestionable when the risk of severe violence is studied.

Keywords

Most scholars would probably agree that the risk of violent victimization is not distributed evenly across social strata (Fattah, 2003; Nilsson and Estrada, 2006; Von Hentig, 1948). Moreover, it is possible that the risk of violence is changing so that the relative risk of socially disadvantaged groups is increasing when contrasted with other social groups. Such developments have been reported in the United States and in the Nordic area (Kivivuori, 2003; Levitt, 1999; Nilsson and Estrada, 2003; Thacher, 2004). They may reflect a decline in the general safety of the lowest social strata, an increasing capability of the upper strata to ‘purchase’ safety from the security market, increasing geographical social segregation in metropolitan areas, or any combination of such factors. Clearly, the existence and persistence of social differentials in violent victimization warrants continued attention.

On the other hand, some studies do not find a social differential in violence risk (Van Kesteren et al., 2000; Wikström and Butterworth, 2006). Some of the controversy in this field may have methodological sources, relating to the ways violent victimization is studied. The most basic distinction lies in the traditionally dual nature of criminological data sources. Register-based studies rely on incidents reported to the criminal justice system. Although the research tradition using administrative data has been prominent when analysing offending, official data are seldom used when the determinants of violent victimization are studied. The problem of hidden crime is the most likely cause of this. Consequently, some researchers prefer victim surveys of different kinds because they are believed to be more inclusive of various kinds of violence, irrespective of police reporting. This, in turn, enables the use of omnibus-type measures of violence, where everything from verbal abuse to aggravated physical violence is aggregated into a single outcome measure. However, this conflation of all violence under one label may reduce the level of comparability between surveys and have a significant impact on the associations between predictors and victimization. Inclusivity of the violence outcome should therefore be considered (Dobash and Dobash, 2004; Felson and Ackerman, 2001) before making definite claims about socioeconomic differences in victimization.

The present study seeks to clarify these methodological issues by examining the impact of data source and inclusivity of the violence outcome on socioeconomic differences in violent victimization in two types of data: a nationally representative register-based sample of Finnish citizens, where we can identify victims of police-reported violence, and a national criminal victimization survey, where the outcome is based on self-reports of violent victimization. Our study has three objectives. First, we explore whether violent victimization risk varies by different measures of socioeconomic status (SES). Second, we do this twice using both survey and register data. In doing so, we hope to answer whether these two parallel sets of data give similar results on socioeconomic determinants of violent victimization. Third, because both register and survey data show variation in the gravity of violence, we analyse how the observed socioeconomic differences are affected when the inclusivity of the victimization outcome is varied by seriousness.

Background

Theoretical approaches to differential victimization risks by SES

A combination of lifestyle and routine activity theories has been the most widely applied explanation for why some socioeconomic and demographic groups are more likely to be victimized than others (Cohen et al., 1981; Cohen and Felson, 1979; Hindelang et al., 1978). Although originally developed to answer different questions (Meier and Miethe, 1993), the two theories are commonly presented as one framework, where opportunity structure has a central role (Meier and Miethe, 1993; Wilcox et al., 2003). In this framework, the factors that influence the risk of victimization are usually formulated as exposure to crime, proximity to crime, target attractiveness and capable guardianship (Cohen et al., 1981; Finkelhor and Asdigian, 1996; Meier and Miethe, 1993).

According to the original formulation of lifestyle theory, ‘lifestyle differences result from differences in role expectations, structural constraints, and individual and subcultural adaptations’ (Hindelang et al., 1978: 245). Lifestyles and routine activities modify the exposure to risk, which in turn affects the likelihood of violent victimization in the long run. Additionally, when victims share sociodemographic features with offenders, the resulting lifestyle similarities increase the likelihood of contacts between the two.

Some elements of exposure to crime are more directly influenced by SES than others. Although the term ‘lifestyle’ might imply a freedom of choice, it is evident that economic resources shape individual lifestyles. For instance, violence in the workplace might contribute to the SES–victimization link, because some occupations are more susceptible to violence than others. Choice of area of residence is affected by economic resources, and living in a poor, high-crime neighbourhood increases one’s proximity to crime. Research has often concentrated on neighbourhood-level measures of affluence and levels of victimization (for example, Lauritsen, 2001; Nilsson and Estrada, 2007; Sampson et al., 1997). The economic status of a neighbourhood is also emphasized in social disorganization theory, where poverty is a key factor affecting the breakdown of social control in a neighbourhood (Shaw and McKay, 1942; Smith and Jarjoura, 1988). Finally, victim–offender overlap should also be considered as a contributing factor to the SES–victimization link, because criminal offending has consistently been identified as a major risk factor for victimization (Lauritsen and Laub, 2007). Because offending is associated with low SES, a person’s own criminal behaviour might mediate the link between low SES and victimization (Bjarnason et al., 1999; Savolainen et al., 2009).

SES and victimization in recent studies

Major victimization surveys include varying measures of SES, and socioeconomic differences in victimization rates are routinely reported in the descriptive reports of the surveys. The National Crime Victimization Surveys (NCVS) in the United States report consistent differences in the prevalence of violent victimization between income groups: those reporting the lowest household income (family income of less than US$7500) have the highest risk of violent victimization, with a yearly prevalence of 18.6 percent, whereas the group with the highest income (family income of US$75,000 or more) has a prevalence of 3.2 percent (Bureau of Justice Statistics, 2010).

The British Crime Survey (BCS), on the other hand, reports much smaller differences between income groups (Flatley et al., 2010). However, being unemployed is associated with the likelihood of victimization in the BCS. The unemployed have a 7.7 percent yearly prevalence of all violence, whereas those in employment have a 3.3 percent yearly prevalence. There are also small differences in victimization risk by the respondent’s occupation and educational qualifications. However, when a variety of variables are controlled in a logistic regression model explaining violent victimization, the increased victimization risk associated with employment status is greatly diminished: sex, age and marital status remain the most robust predictors of victimization in the British data, and SES differences, measured by occupation, education or income, are relatively small (Flatley et al., 2010).

Not all studies find a strong association between measures of SES and violent victimization. Using data from International Crime Victim Survey (ICVS) data, Van Kesteren et al. (2000) found that high income and high education increased, rather than decreased, the risk of violent victimization in a combined sample of 17 countries. In a similar vein, using data from the Peterborough Youth Study, Wikström and Butterworth (2006) found that family social class is a poor predictor of adolescent victimization. As demonstrated by these results and partly by the BCS as well, low SES is not a universal predictor of victimization in empirical studies.

Studies based on Nordic data provide the best comparison for the current study. Using Swedish survey data, Nilsson and Estrada (2003, 2006, 2007) have examined the connection between socioeconomic measures and victimization in a Nordic setting. Nilsson and Estrada (2003) investigated the association between income and violent victimization during an economic recession in Sweden and found evidence of increasing differences between income groups during 1988–99. In another article, Nilsson and Estrada (2006) discovered that the increasing differences (1984–2001) were particularly marked in violence resulting in medical treatment. According to their analysis, the differences between income groups grew both in violence at home and in public places during 1984–2001 (Nilsson and Estrada, 2006).

Methodological issues

Research on risk factors of violent victimization has usually relied on self-report surveys. Despite having the advantage of including unreported crime, there are some issues associated with surveys regarding the relationship between SES and violent victimization. SES-related non-response bias is an obvious one, because response rates are often low in groups with the highest risks of victimization (Brottsförebyggande rådet, 2000; Catalano, 2007. If those in marginalized positions do not answer the surveys, the estimates on socioeconomic differences might be seriously biased. Even if these individuals are reached, only a few surveys employ large enough samples to make analysis of the most severe forms of violence possible.

Victimization surveys have been subject to rather intense methodological research. As one key task of general victimization surveys is to uncover crime rate trends independently of police action, the most studied issue has probably been convergence and divergence in US crime rates in NCVS and Uniform Crime Reports (for an overview, see McDowall and Loftin, 2007). However, studies that would utilize different operationalizations (survey vs. official data) of victimization as outcome variables in individual-level research on determinants of victimization are not easily found, particularly those that would try to compare the two data sources.

One important tradition of research related to our research question is the one focused on determinants of whether a victim reports a violent incident to the police. According to an overview article (Skogan, 1984), socioeconomic differences in police reporting are quite small. A recent analysis using BCS data (Tarling and Morris, 2010) reports a rather similar finding, and sums up prior research by concluding that high SES increases rather than decreases the likelihood of police reporting. Additionally, their results indicate that severity of violence is the most important factor affecting the decision to report a violent incident to police (Tarling and Morris, 2010).

Research question

Recent Nordic studies have reassessed the relationship between SES and offending (Aaltonen et al., 2011; Galloway and Skarðhamar, 2010; Nilsson and Estrada, 2009). Longitudinal register-based data sets that include reliable measurements on an individual’s education, income and employment have been very useful in this type of research, making it possible to use administrative register data to build nationally representative data sets with large samples that have minimal attrition compared with surveys. However, the data sets have not yet been used in victimization research.

The principal research question of this study is whether socioeconomic differences in violent victimization vary by data source and inclusivity of the outcome measure used. We examine the association between several measures of SES and victimization in two nationally representative data sets, the Finnish National Victimization Survey and the register-based Risk Factors of Crime in Finland. However, because the two data sets do not include the exact same measures of SES, we are not able to do a direct comparison for effect sizes. For this reason, we do additional analysis with education as the independent variable, because it is measured similarly in both data. Focusing only on police-reported violence in both data sets enables us to assess the effect of non-response on survey estimates.

Data and methods

Finnish National Victimization Survey

In the Finnish National Victimization Survey (FNVS), respondents are asked about their violent experiences during the last 12 months. After a series of screening questions, those respondents who report violence during the last year are then asked more detailed questions about a maximum of the three latest violent incidents during that time. To ensure sufficient statistical power for analysis of rare incidents (violence with injury), we combined the two (2003 and 2006) FNVS sweeps based on random samples of the Finnish population into one single data set (N = 8562). The FNVS has been able to maintain a fairly good response rate: 81 percent in 2003 and 77 percent in 2006.

The FNVS includes several measurements for SES. Of the measures used here, household income is the only one based on self-reports. Additionally, register-based measures from Statistics Finland have been incorporated into the survey data. These measures include education, main economic activity and occupational social class. Table 1 shows the distributions of all variables.

Table 1.

Descriptive statistics: Finnish National Victimization Survey (FNVS) and Risk Factors of Crime in Finland (RFCF)

	Survey (FNVS)				Register (RFCF)
		N	Percent			N	Percent
Gender	Male	4300	50.2	Sex	Male	33,088	50.9
	Female	4261	49.8		Female	31,922	49.1
Education	Basic	1522	17.8	Education	Basic	11,766	18.1
	Secondary	4352	50.8		Secondary	33,141	51.0
	College-level	1290	15.1		College-level	8841	13.6
	University	1398	16.3		University	11,262	17.3
Occupational social class	Upper white collar	1631	19.0	Occupational social class	Upper white collar	17,661	27.2
	Lower white collar	2407	28.1		Lower white collar	12,014	18.5
	Manual	2306	26.9		Manual	17,636	27.1
	No occ./missing	2218	25.9		No occ./missing	17,700	27.2
Household income	€5000–7000	1175	13.7	Income	5th quintile	12,994	20.0
	€3000–5000	2497	29.2		4th quintile	13,019	20.0
	€1000–3000	2939	34.3		3rd quintile	13,015	20.0
	< €1000	910	10.6		2nd quintile	13,019	20.0
	Missing	1041	12.2		1st quintile	12,963	19.9
Main economic activity	Employed	6649	77.7	Unemployment 1999–2004	No	47,707	73.4
	Unemployed	589	6.9		< 1 year	9560	14.7
	Student/military service	593	6.9		1–2 years	2762	4.2
	Disability	205	2.4		2 > years	3582	5.5
	Other	525	6.1		Disability retirement	1400	2.2
Victimized during last year	All violence	1236	14.4	Victimized 2005–2007	All violence	1571	2.4
	Physical violence	682	8.0		Petty assault	490	0.8
	Physical violence with injury	254	3.0		Assault	1055	1.6
					Aggravated assault/attempted homicide	164	0.3
Total N		8562	100.0	Total N		65,010	100.0

Risk Factors of Crime in Finland

The register-based Risk Factors of Crime in Finland (RFCF) data set is built around a nationally representative random sample of 150,010 Finnish citizens in 2004. Here, we use a sub-sample of 65,010 individuals. The primary sample was drawn from the Finnish Population Information System, a register that contains information about all Finnish citizens and foreign citizens permanently residing in Finland. Using personal identification numbers to combine data from different administrative registers to the primary sample, the data include several measures on both socioeconomic and demographic characteristics of every individual. Some observations are from only one point in time, but data on victimization experiences, for instance, are longitudinal, from years 2005–2007. Information about violent victimization is derived from the police information system.

The register-based dependent variables that measure SES in the RFCF data are education, occupational social class and personal income. These variables were measured at the end of 2004. To complement these, we have a variable that measures the number of days an individual was registered as unemployed or on disability retirement during the years 1999–2004. This variable was recoded into the categories ‘no unemployment’, ‘under 1 year’, ‘1–2 years’, ‘over 2 years’ and ‘disability retirement’.

Modelling strategy

Because the seriousness of violence is related to the likelihood of police reporting (Tarling and Morris, 2010) and possibly to socioeconomic differences in violent victimization (Nilsson and Estrada, 2006), we use several outcome variables of different severity. In FNVS, we disaggregate violence into three outcome variables: the most inclusive ‘all violence’ (any incident, verbal or physical, defined by the respondent as violence), intermediate ‘physical violence’ and the most restrictive ‘physical violence with injury’. The same categories have been used in descriptive reports of FNVS (Sirén et al., 2010). In RFCF, we use two outcome variables: the first includes all assaults (Finnish penal codes ‘petty assault’, ‘assault’, ‘aggravated assault’, ‘attempted homicides’), and the second includes only aggravated assaults and attempted homicides.

Although the two data sets still include different types of violence, altering the relative inclusivity of the definition of victimization allows us to compare the data sets better. In addition, it allows us to assess how inclusivity of the definition of violence affects socioeconomic differences within each data set. In police-reported violence, it is reasonable to assume that the most serious incidents of violence represent the actual level of such violence the best, because they are reported most often (Tarling and Morris, 2010). Police-reported petty assaults, on the other hand, are likely to capture only a fraction of all such violence that would meet the definition in the Finnish penal code.

Our measures of SES are interrelated and overlapping. For this reason, we present the results of logistic regression analysis first as age- and gender-adjusted bivariate (or crude) odds ratios (OR). This gives a better chance of comparing results between the two data sets, because full models would inevitably not be comparable given the different predictors available in each data set. Although our focus is on bivariate associations and comparisons between the two sets of data, we also fitted the full models, and the results from these models are also briefly discussed.

When considering the relationship between measures of SES and victimization, it is important to take age into account. First, because the SES of children and adolescents cannot be measured except by their parents’ SES, we set the lower age limit in all analyses to 19 years. Additionally, given that we want to focus on people who have a relatively high likelihood of being victimized, we set an upper age limit at 50 years of age. In all analyses, we compare 19–30-year-olds with 31–50-year-olds, in order to separate those with relatively stable SES, in terms of education, occupation and income, from those whose SES is only taking form. Although we stratify the analysis into two age groups, we also control for age as a continuous variable in all analyses.

Results

First, we present the survey results from the FNVS, and then we compare the results with those obtained from the register-based RFCF. Finally, we attempt a more specific test of similarity between the data sets and focus on differences by educational level in violent victimization that was reported to police.

Survey-based analysis

We start the survey-based analysis with the most inclusive victimization variable, the one that includes both verbal threats and physical violence. These incidents are relatively common at the population level: 14 percent of 19–50-year-olds in the survey data reported such an incident during the last 12 months (see Table 1). When verbal threats are excluded, the proportion reporting physical violence is 8 percent. Last, when we restrict the outcome to violence that caused a physical injury, the proportion is 3 percent.

Starting with the most inclusive definition of violence, education and income are related to victimization risk in both age groups at the bivariate level, where only age and gender are controlled (Table 2). Differences by educational group are larger for 19–30-year-olds, whereas low household income is a stronger predictor among 31–50-year-olds. Main economic activity and occupational class are not significant predictors in the younger group, but in the older group we find differences between employed individuals and students, or those on disability retirement. Those in lower white-collar positions and without an occupation have a higher risk as well. In the younger group, there are no statistically significant gender differences, but in the older group females have a higher risk (male OR 0.76) when age is controlled.

Table 2.

Logistic regression models: age- and gender-adjusted associations (odds ratios) between independent variables and violent victimization by severity of violence in FNVS data

		19–30-year-olds			31–50-year-olds
		All	Physical	Injury	All	Physical	Injury
Gender (age)	Female	1.00	1.00	1.00	1.00	1.00	1.00
Gender (age)	Male	1.19	1.01	1.17	0.76*	0.76*	0.87
Education (age + gender)	University	1.00	1.00	1.00	1.00	1.00	1.00
	College-level	1.29	1.22	1.82	1.06	0.91	1.43
	Secondary	1.57*	1.78*	2.93*	1.39*	1.35	3.02*
	Basic	1.91*	2.28*	4.97*	1.41*	1.49*	4.83*
Household income (age + gender)	€5000–7000	1.00	1.00	1.00	1.00	1.00	1.00
	€3000–5000	1.15	1.56	1.69	1.04	0.89	0.87
	€1000–3000	1.20	1.36	1.61	1.32*	1.44	2.01*
	< €1000	1.77*	1.99*	1.94	2.36*	2.32*	3.47*
Main economic activity (age + gender)	Employed	1.00	1.00	1.00	1.00	1.00	1.00
	Unemployed	1.01	1.18	1.21	1.20	1.28	2.67*
	Student/military service	0.93	0.98	0.68	2.05*	2.86*	3.51*
	Disability pension	1.12	0.94	2.79	1.70*	2.79*	3.75*
	Other	0.74	0.57*	0.66	0.70	0.77	1.12
Occupational social class (age + gender)	Upper white collar	1.00	1.00	1.00	1.00	1.00	1.00
	Lower white collar	1.21	1.24	1.46	1.96*	1.69*	3.21*
	Manual	1.02	0.99	1.28	1.18	1.09	2.20*
	No occ./missing	1.01	1.08	1.11	1.56*	1.57*	3.33*

Note: * Statistically significant (p < .05) association when controlling for variables in parentheses.

Although the measures of SES showed significant bivariate associations with the most inclusive violence outcome, it could be argued that the effect of the predictor variables was generally quite weak. For instance, the unemployed (ORs 1.01 and 1.20) had almost the same risk as employed persons. Using a more restrictive outcome variable with only physical violence does not modify the associations very much. Education and income show statistically significant associations of similar magnitude. Additionally, in the older group, being a student or being on disability retirement shows stronger effects than in the model with the more inclusive outcome variable.

However, when the outcome measure includes only cases that resulted in physical injury, the bivariate associations become stronger. In the younger group, the changes are smaller, the main change being the growth in educational differences. Those with only basic education have five times the odds of those with university education. Surprisingly, although still statistically significant, the effect of income becomes smaller. In the older group, the changes are more systematic: the differences grow in all variables measuring SES. Educational differences are the largest, and there are significant differences by income and occupation as well. With this outcome variable, the effect of being unemployed becomes significant (OR 2.7). Gender differences are not statistically significant in either group.

When all the predictors are mutually adjusted, many lose their significance, which is expected given the overlap of the SES variables (analysis not shown). In the younger group, education remains a robust predictor in all three outcome variables. The effect of low income surprisingly decreases with severity, and main activity and occupation are not statistically significant in any of the models. In the older group, education is a significant predictor only in the most severe violence. Low income is a significant predictor only in the most inclusive outcome. Variables measuring main activity and occupational class have significant effects. Even though single variables lose their significance in the full model, if we use the coefficients from the full model to calculate the predicted probabilities¹ for ‘low SES’ and ‘high SES’ individuals, we find that the difference between opposite ends of the social strata grows as the dependent variable becomes less inclusive. Thus, both bivariate and multivariate analysis produced the same main finding: the more serious the violence we measure, the greater is the differential in victimization rates between different socioeconomic groups.

Register-based analysis

Next, we test similar questions with register-based RFCF data. When comparing these results with those from survey data, one should bear in mind that here we are using a three-year prospective follow-up instead of violence during the last 12 months.

Being a victim of police-recorded violent crime is a rather rare incident at the population level. During the three-year follow-up, 0.3 percent of 19–50-year-olds were registered as victims of aggravated assault or attempted homicide, 1.6 percent as victims of assault, and 0.8 percent as victims of petty assaults. Similarly to the survey data, younger subjects had more violent experiences.

Looking at bivariate associations (Table 3) from age- and gender-adjusted logistic regression models, the predictor variables show statistically significant associations in the first outcome variable comprising all assaults. In the younger group, gender differences are larger, as are the differences between educational groups. For other measures (income, occupation and unemployment length) of SES, however, the differences are generally greater in the older group. This was an expected result given the fact that measures for income and occupation of those under 30 are somewhat confounded by the large number of students in the age group. Despite this, the effects are strong at the bivariate level in the younger group as well, unlike in the survey data, where education was the only variable with significant effects.

Table 3.

Logistic regression models: age- and gender-adjusted associations (odds ratios) between independent variables and violent victimization by severity of violence in RFCF data

		19–30-year-olds		31–50-year-olds
		All assaults	Aggravated assault/attempted homicide	All assaults	Aggravated assault/attempted homicide
Gender (age)	Female	1.00	1.00	1.00	1.00
Gender (age)	Male	2.18*	5.73*	1.32*	2.22*
Education (age + gender)	University	1.00	1.00	1.00	1.00
	College-level	1.67	3.07	1.61*	2.75
	Secondary	3.42*	10.28	3.26*	9.11*
	Basic	9.32*	35.63*	5.71*	19.97*
Age-specific income quintile(age + gender)	5th quintile	1.00	1.00	1.00	1.00
	4th quintile	0.94	0.66	1.48*	0.66
	3rd quintile	1.19	1.27	1.48*	2.75
	2nd quintile	1.14	1.07	2.75*	4.45*
	1st quintile	1.70*	2.76*	4.83*	15.50*
Occupational social class(age + gender)	Upper white collar	1.00	1.00	1.00	1.00
	Lower white collar	2.38*	1.71	1.88*	0.99
	Manual	2.41*	2.58	1.69*	2.65*
	No occ./missing	3.17*	5.89*	4.56*	12.05*
Unemployment (age + gender)	No	1.00	1.00	1.00	1.00
	< 1 year	1.45*	1.40	2.44*	2.48*
	1–2 years	1.95*	3.85*	4.36*	11.93*
	2 > years	3.92*	4.62*	5.98*	13.90*
	Disability retirement	0.80	–	2.96*	8.90*

Note: * Statistically significant (p < .05) association when controlling for variables in parentheses.

When we restrict the outcome variable to aggravated assault and attempted homicide, the effects of predictor variables grow substantially. Because practically no subjects with tertiary education show up as victims of these incidents, the coefficient for having only a basic education inflates. We see similar changes in all the variables measuring SES: in both age groups, the victims of the most serious violence are a highly selected (socially disadvantaged) group. For the younger group, the variation is greatest by education, whereas in the older group both income and unemployment length have additionally very strong effects. The greater effect of unemployment in RFCF data compared with FNVS might be explained by the measurement that disaggregates unemployment by its length.

As in the FNVS data, mutually adjusting for all variables modified the bivariate associations (analysis not shown). In both age groups, educational differences remained strong with both outcome variables, as did the effect of long unemployment. The effects of income and occupational class lose their significance. As in the survey data, using the predicted probabilities calculated from the full logistic regression model show increasing differences between opposite ends of the social strata as violence grows more serious.

Educational differences in police-reported crime – A direct comparison

Most of the predictors used in the analysis are not identical in both data sets. Fortunately, the measure for educational level is based on the same administrative register and provides a better ground for comparing the two sets of data. It is also of substantive interest because of the strong association between education and violent victimization. Additionally, survey data include information about whether the violent incident was reported to the police. Thus, if we form an outcome variable that captures only police-reported violence in the FNVS data, we should be able to compare the results from such models with those from RFCF data. In the survey-based outcome variable, we included only police reports of physical violence, excluding threats. In the FNVS questionnaire, the respondents were asked whether the incident ‘became known to police’, and if one answered yes, they were later asked whether ‘police filled in a report of the offence’. In RFCF data, we altered the outcome variable so that it includes only victimizations from one year (2005). In Table 4, we present results for both of these outcome variables, controlling for gender and age.

Table 4.

Logistic regression models: age- and gender-adjusted associations (odds ratios) between education and victimization in police-recorded crime during one year in FNVS and RFCF data

		FNVS		RFCF
		‘Known to police’	‘Report filled’	All victims	Aggravated assault dropped	Assault dropped
Gender	Female	1.00	1.00	1.00	1.00	1.00
	Male	1.34	1.52	1.50	1.38	0.90
Education	University	1.00	1.00	1.00	1.00	1.00
	College-level	1.87	2.25	1.55*	1.47	2.28
	Secondary	1.70	2.93*	2.48*	2.35*	2.99*
	Basic	3.05*	3.96*	6.37*	6.02*	6.38*
	Percent	1.5	0.9	0.9	0.8	0.2
	N of victims	126	73	555	500	155

Statistically significant (p < .05) association.

As expected, the more restrictive ‘report-filled’ criterion produces larger differences between opposite ends of the education scale (OR 3.96 vs. 3.05). If compared with the results before, we can also see that the gender difference grows (OR 1.52), and men are more likely to be involved in violent incidents that were reported to police. In the RFCF data, however, educational differences still appear larger. First, when we use a simple register-based outcome variable that includes all types of violence irrespective of its severity, the odds ratio for group with basic education only is 6.37. This might yet be caused by inclusion of more serious incidents in the RFCF data. However, even if we exclude all individuals who were victims in attempted homicide and aggravated assault and, finally, in assaults, leaving only those who were victims in petty assaults, we still find similar differences by education in the RFCF data (OR 6.38).

Although this analysis is far from conclusive and the insufficient number of cases hampers the survey-based analysis, the difference between the two data sets is substantial. Despite the fact that both data sets give similar estimates of the yearly prevalence of victimization in the population aged 19–50, there seems to be some level of discrepancy between the FNVS and the RFCF data, even after restricting the analysis to comparable predictors and outcomes. Given that the register-based data certainly cover police-reported violence well, in addition to representing the total population better than the FNVS, it seems that the victimization survey excludes some segment of the population with only basic education and a high risk of victimization. Thus, the survey estimates on the effect of educational level on victimization are biased downwards.

Discussion

This study had three goals: (a) to study the association between SES and victimization; (b) to determine if the detected links are robust when analysed using different data sources; and (c) to assess if the detected links are robust when analysed using different operationalizations of the violence concept. On the basis of our results, we argue that the significant socioeconomic differences in violent victimization in official data are not an artefact caused by exclusion of hidden crime in the higher SES groups. Both survey and register data show that serious violence, in particular, is associated with low SES. The register-based results on aggravated assaults and attempted homicides are especially striking because they show the unequal distribution of these events. Finnish homicide research has shown that both perpetrators and victims of lethal violence are often marginalized persons (Kivivuori and Lehti, 2006), and that the comparatively high level of homicide in Finland is largely caused by mutual violence between marginalized men with serious substance abuse. Our results indicate that the population usually victimized in attempted homicides and aggravated assaults might share these attributes.

With regard to the methodological aspect of the study, we have three main conclusions. First, it is clear that the definition of violence matters. Within both survey- and register-based data, we witnessed increasing socioeconomic differences in violence when less serious violent incidents were excluded from the analysis. On the other hand, if one uses a wide-ranging omnibus measure of violence, socioeconomic factors are generally less important. Thus, to understand socioeconomic differences in violence, it appears that the distinction between more and less serious violence is at least as important as that between reported and hidden violence. For this reason, researchers should be self-reflective concerning how their violence definitions affect their results (Felson and Ackerman, 2001). In addition to measuring violence with a wide scope, it would be important to have more objective measures of serious violence that would be more comparable across subjects and surveys. One solution is to use a physical injury criterion.

Second, the source of the data matters. Although socioeconomic differences can be detected in both sets of data, violent victimization appears much more evenly distributed in survey- than in register-based data. Different indicators of SES give, for the most part, similar results, but the magnitude of the association varies and age changes the relative importance of the predictors. Generally, having only basic education seems to be the most robust predictor, followed by long unemployment in the register data. Expectedly, for older subjects, income and occupation-based social class were better predictors. Overall, when we restrict the FNVS models to violence resulting in injury and compare the results with those from the RFCF data with all assaults as the outcome, the two data sets converge quite well among those over 30. On the other hand, among 19–30-year-olds, the discrepancy remains. Gender differences are larger in the RFCF data (males have higher risk), but are reduced by age as in the FNVS. Disaggregation by types of violence would probably explain some of these findings.

Third, the final analysis on police-reported crime suggested that non-response is likely to affect the perceived SES–victimization link in surveys. Although it is difficult to quantify the exact difference between the data sets, and the difference could yet be attributed to data quality issues, we can, with some certainty, say that some individuals with low education and high risk of victimization are absent from the survey data. This is in concordance with a Swedish study that discovered low response rates and high victimization rates among marginalized and criminally active people (Brottsförebyggande rådet, 2000).

Conclusion

In many ways, our findings resemble those by Farnworth et al. (1994) on the SES–offending link, where they discovered that the association between low SES and offending is unquestionable once true underclass status and non-trivial crime are measured. Furthermore, their results suggest that using measures that are too inclusive may conceal relevant social differentials. The same seems to apply for victimization although, in our analysis, the associations between SES and victimization resemble continuous gradients rather than thresholds, and the risk of violence also varies outside the most disadvantaged group.

However, in line with Wikström and Butterworth (2006), the explained variance in victimization is relatively low, even after including all the SES measures in the model. Using the RFCF data, we ran similar models with violent offending as the outcome, and found that the overall explained variance (Nagelkerke R²) with the same independent variables was almost twice as large. When compared with offending, victimization remains a harder phenomenon to predict (Lauritsen, 2010). It is thus evident that understanding the causal mechanisms leading to victimization requires looking beyond structural factors. Also, we acknowledge that the current study is very general in its scope, and there are debates on specific types of violence – domestic violence, in particular – that cannot be covered here. The impact of survey design and the inclusivity of violence measure on the estimated prevalence of domestic violence has been a central point of contention (Haggerty, 2001). However, the conflation of physical and verbal violence under one label has been criticized in these debates as well (Dobash and Dobash, 2004).

Finally, the results of this article demonstrate the potential of Nordic register data as a complementary source for studying victims of serious violent crime. Official data, both police reports and hospital discharges (see Estrada, 2006), could probably be used more often as an outcome in research on victims of violence. However, because register-based studies will remain limited to analysing official data, survey methodology should be used and developed further. Although both a difficult and expensive task, survey research would benefit from sustained (for example, Cantor and Lynch, 2007) and even increased attention to non-response, and we should continue to seek ways to include all social strata in our data. Also, because the age–victimization curve peaks early and the discrepancy between our data sets was larger among 19–30-year-olds, the challenge to forestall non-response among youth is acute. Otherwise, we might fail to recognize significant socioeconomic and demographic differences in violence that clearly exist, even in Nordic welfare states.

Footnotes

Notes

References

Aaltonen

Kivivuori

Martikainen

(2011) Social determinants of crime in a welfare state: Do they still matter? Acta Sociologica 54(2): 119–138.

Bjarnason

Sigurdardottir

Thorlindsson

(1999) Human agency, capable guardians, and structural constraints: A lifestyle approach to the study of violent victimization. Journal of Youth and Adolescence 28(1): 105–119.

Brottsförebyggande rådet (2000) Vanor och trygghet – intervjuundersökning av socialt utsatta personer och normalbefolkning. Stockholm: Brottsförebyggande rådet.

Bureau of Justice Statistics (2010) Criminal Victimization in the United States, 2007 Statistical Tables. National Crime Victimization Survey. US Department of Justice. URL (accessed 2 September 2011): bjs.ojp.usdoj.gov/content/pub/pdf/cvus0701.pdf.

Cantor

Lynch

(2007) Addressing the challenge of costs and errors in victimization surveys: The potential of new technologies and methods. In: Hough

Maxfield

(eds) Surveying Crime in the 21st Century. Cullompton: Willan Publishing.

Catalano

(2007) Methodological change in the NCVS. In: Lynch

Addington

(eds) Understanding Crime Statistics. Revisiting the Divergence of the NCVS and UCR. Cambridge: Cambridge University Press.

Cohen

Felson

(1979) Social change and crime rate trends: A routine activity approach. American Sociological Review 44(August): 588–608.

Cohen

Kluegel

Land

(1981) Social inequality and predatory criminal victimization: An exposition and test of a formal theory. American Sociological Review 46(October): 505–524.

Dobash

(2004) Women’s violence to men in intimate relationships. British Journal of Criminology 44(3): 324–349.

10.

Estrada

(2006) Trends in violence in Scandinavia according to different indicators. An exemplification of the value of Swedish hospital data. British Journal of Criminology 46(3): 486–504.

11.

Farnworth

Thornberry

Krohn

Lizotte

(1994) Measurement in the study of class and delinquency: Integrating theory and research. Journal of Research in Crime and Delinquency 31(1): 32–61.

12.

Fattah

(2003) Violence against the socially expendable. In: Heitmeyer

Hagan

(eds) International Handbook of Violence Research. Dordrecht: Kluwer Academic Publishers.

13.

Felson

Ackerman

(2001) Arrests for domestic violence and other assaults. Criminology 39(3): 655–676.

14.

Finkelhor

Asdigian

(1996) Risk factors for youth victimization: Beyond a lifestyles/routine activities theory approach. Violence and Victims 11(1): 3–19.

15.

Flatley

Kershaw

Smith

Chaplin

Moon

(eds) (2010) Crime in England and Wales 2009/10. Findings from the British Crime Survey and Police Recorded Crime (Third Edition). London: Home Office.

16.

Galloway

Skarðhamar

(2010) Does parental income matter for onset of offending? European Journal of Criminology 7(6): 424–441.

17.

Haggerty

(2001) Making Crime Count. Toronto: University of Toronto Press.

18.

Hindelang

Gottfredson

Garofalo

(1978) Victims of Personal Crime: An Empirical Foundation for a Theory of Personal Victimization. Cambridge: Ballinger.

19.

Kivivuori

(2003) Väkivallan uhriksi joutuminen eri tuloryhmissä 1980–2003. In: Sirén

Honkatukia

(eds) Suomalaiset väkivallan uhreina. Tuloksia 1980–2003 kansallisista uhritutkimuksista. Helsinki: Oikeuspoliittinen tutkimuslaitos.

20.

Kivivuori

Lehti

(2006) The social composition of homicide in Finland, 1960–2000. Acta Sociologica 49(1): 67–82.

21.

Lauritsen

(2001) The social ecology of violent victimization: Individual and contextual effects in the NCVS. Journal of Quantitative Criminology 17(1): 3–32.

22.

Lauritsen

(2010) Advances and challenges in empirical studies of victimization. Journal of Quantitative Criminology 26(4): 501–509.

23.

Lauritsen

Laub

(2007) Understanding the link between victimization and offending: New reflections on an old idea. In: Hough

Maxfield

(eds) Surveying Crime in the 21st Century. Crime Prevention Studies, Vol. 22. Cullompton: Willan Publishing.

24.

Levitt

(1999) The changing relationship between income and crime victimization. Economic Policy Review 5 (September): 87–98.

25.

McDowall

Loftin

(2007) What is convergence, and what do we know about it? In: Lynch

Addington

(eds) Understanding Crime Statistics. Revisiting the Divergence of the NCVS and UCR. Cambridge: Cambridge University Press.

26.

Meier

Miethe

(1993) Understanding theories of criminal victimization. In: Tonry

(ed.) Crime and Justice – A Review of Research volume 17. Chicago: University of Chicago Press.

27.

Nilsson

Estrada

(2003) Victimization, inequality and welfare during an economic recession: A study of self-reported victimization in Sweden. British Journal of Criminology 43(4): 655–672.

28.

Nilsson

Estrada

(2006) The inequality of victimization: Trends in exposure to crime among rich and poor. European Journal of Criminology 3(4): 387–412.

29.

Nilsson

Estrada

(2007) Risky neighbourhoods or individuals at risk? The significance of neighbourhood conditions for violent victimization in residential areas. Journal of Scandinavian Studies in Criminology and Crime Prevention 8(1): 2–21.

30.

Nilsson

Estrada

(2009) Criminality and Life Chances. A Longitudinal Study of Crime, Childhood Circumstances and Living Conditions up to Age 48. Stockholm: Stockholm University.

31.

Sampson

Raudenbush

Earls

(1997) Neighborhoods and violent crime: A multilevel study of collective efficacy. Science 277: 918–924.

32.

Savolainen

Sipilä

Martikainen

Anderson

(2009) Family, community, and lifestyle: Adolescent victimization in Helsinki. Sociological Quarterly 50(4): 715–738.

33.

Shaw

McKay

(1942) Juvenile Delinquency and Urban Areas: A Study of Rates of Delinquents in Relation to Differential Characteristics of Local Communities in American Cities. Chicago: University of Chicago Press.

34.

Sirén

Aaltonen

Kääriäinen

(2010) Suomalaisten väkivaltakokemukset 1980–2009. Helsinki: Oikeuspoliittinen tutkimuslaitos.

35.

Skogan

(1984) Reporting crimes to the police: The status of world research. Journal of Research in Crime and Delinquency 21(2): 113–137.

36.

Smith

Jarjoura

(1988) Social structure and criminal victimization. Journal of Research in Crime and Delinquency 25(1): 27–52.

37.

Tarling

Morris

(2010) Reporting crime to the police. British Journal of Criminology 50(3): 474–490.

38.

Thacher

(2004) The rich get richer and the poor get robbed: Inequality in U.S. criminal victimization, 1974–2000. Journal of Quantitative Criminology 20(2): 89–116.

39.

Van Kesteren

Mayhew

Nieuwbeerta

(2000) Criminal Victimization in Seventeen Industrialised Countries. Key Findings from the 2000 International Crime Victims Survey. The Hague: Ministry of Justice, WODC.

40.

Von Hentig

(1948) The Criminal and His Victim. New Haven, CT: Yale University Press.

41.

Wikström

Butterworth

(2006) Adolescent Crime. Individual Differences and Lifestyles. Cullompton: Willan Publishing.

42.

Wilcox

Land

Hunt

(2003) Criminal Circumstance. A Dynamic Multicontextual Criminal Opportunity Theory. Hawthorne: Aldine de Gruyter.