Abstract
Self-reported victimization among young women in the Swedish Crime Survey increased during the early- to mid-2010s, resulting in a gender gap, with young women reporting 10% to 15% higher victimization rates than young men. In this study, we use multi-group confirmatory factor analysis to assess whether the increase in victimization rates is due to declines in measurement equivalence across different cohorts of young women during the period 2009 to 2016. The results show cohort differences in how young women interpret the victimization items in the Swedish Crime Survey, especially with regard to the item about sexual victimization. The results show that the increasing victimization rates are likely the result of changes in perceptions about sexual offences, rather than an actual increase.
Keywords
Introduction
In the Swedish Crime Survey, young women (aged 16–24 years) have consistently self-reported higher victimization rates than young men over the last 10 . For example, a total of 39.9% of the women aged 16 to 24 years self-reported having been subjected to an interpersonal crime in the latest Swedish crime survey, compared to 26.8% in the corresponding age group among males (BRÅ, 2022a). This trend challenges conventional notions of gendered victimization patterns, since from a historical perspective these patterns have been characterized by higher victimization rates among young men (Lilley et al., 2023).
One possible explanation that is often highlighted in the Swedish debate on this statistical trend is that the increase in self-reported victimization might not be due to an actual increase. Instead, within the theoretical framework of research on the Nordic paradox, it has been suggested that the high levels of gender equality in the Nordic countries have resulted in a broadening of perceptions about victimization in general. In statistical terminology, this would be described as a decline in measurement equivalence over time. As the large bulk of the increase in self-reported victimization among young women is comprised of sexual offences (BRÅ, 2018) it has been suggested that changes in perceptions of crime has been more pronounced when it comes to sexual offences (BRÅ, 2018; Gracia et al., 2019; Humbert et al., 2021; Wemrell et al., 2021). Consequently, it is believed that a broader set of behaviors have come to be categorized and reported as criminal victimization over time.
However, no studies have to date assessed this explanation empirically. The aim of the present study is therefore to assess whether the formation of the gender gap coincides with a decline in measurement equivalence among young female respondents (aged 16–24 years) in the Swedish Crime Survey. The study period includes the 4 years during which the gap emerged (2013–2016) and, for comparison purposes, the preceding 4 years (2009–2012, during which the victimization prevalence rates of young men and women were relatively similar and stable. We hold that this period is highly relevant to study, since it is the period during which the “new normal,” involving higher victimization rates among young Swedish women, emerged and since it is a period that remains poorly understood and under-researched.
More specifically, we assess whether statistical tests of measurement equivalence (multi-group confirmatory factor analysis) support or reject the notion that the increase in self-reported victimization among young women is the result of changed perceptions about inter-personal crime, and sexual offences in particular. Tests of measurement equivalence allow us to examine whether different populations, in this case different cohorts of young women, interpret and answer survey questions in the same way.
Based on previous research and theory (Gracia et al., 2019; Gracia & Merlo, 2016; Humbert at al., 2021; Wemrell et al., 2019), we hypothesize that reductions in measurement equivalence will be observed when comparing the first 4 years of the selected period with the last 4 years. We also hypothesize that reductions in measurement equivalence and reductions in model fit over time will primarily be due to the sexual offences variable.
The following background will provide an overview of research on the closing gender gap in victimization rates between men and women, describe the crime patterns that characterize the victimization of young women, and provide a description of the trends found in the Swedish Crime Survey.
Previous Research
The Closing Gender Gap in Victimization Rates
Historically, large-scale national crime surveys such as the British Crime Survey (BCS) and the US National Crime Victimization Survey (NCVS) have shown that men are more likely than women to become victims of crime (Office for National Statistics, 2024; van Dijk, 2016). However, the gender gap in victimization rates has narrowed over time in several countries, such as the USA, the UK, and the Netherlands (Lilley at al., 2023; Office for National Statistics, 2024; van Dijk, 2016). Using data from the NCVS, Lilley et al. (2023) showed that the gap in victimization rates has not only decreased over time in the USA, but that “the unadjusted female victimization rate has been higher than that of males for 4 out of the last 5 years of available data” (Lilley et al., 2023, p. 153). The authors found that the closing of the gender gap in victimization rate was not the result of lower rates of violent crime more broadly, but was associated with lower rates of school dropout, improved economic conditions, more women in the workforce, and an increased number of police officers (Lilley et al., 2023).
Victimization Patterns for Young Women
That the gender gap in victimization rates between men and women is decreasing in several countries does not mean that victimization patterns are equivalent for men and women. Both large-scale crime national surveys from different countries (BRÅ, 2022a; Bränström et al., 2022; Hayes & Kopp, 2019; Lilley et al., 2023; Office for National Statistics, 2024; van Dijk, 2016) and qualitative interview studies among risk groups (Kennedy et al., 2021; Øverlien et al., 2020; Wesely, 2006) illustrate that the victimization of young women is characterized by gender-specific crime patterns.
The most pervasive and consistently reported gender difference involves differences in the types of offences directed against women and men respectively. Men are more likely to become victims of physical assault, while women are more likely to become victims of sexual offences (Björklund et al., 2010; Blom et al., 2014; Cater et al., 2014; Ceccato et al., 2021; Lilley et al., 2023; Pezzoli et al., 2020; van Dijk, 2016). Moreover, it would be incorrect to present women as a homogenous group, since there are also considerable within-group differences. With regard to sexual offences, for example, research has consistently shown that young women are targeted more frequently than older women (Ceccato et al., 2021; Lilley et al., 2022; van Dijk, 2016).
In addition, previous research from Sweden has highlighted distinct gender patterns in risk factors for victimization. For example, experiences of early domestic abuse have an impact on the victimization risk among both female and male youth (Ørke et al., 2022; Pezzoli et al., 2020). However, for young Swedish women, the increased risk does not solely relate to physical violence, as it does for young men, but is also accompanied by an increased risk for sexual assault (Blom et al., 2014; Ceccato et al., 2021; Pezzoli et al., 2020).
Some risk factors described in recent studies, such as low socio-economic status (Aaltonen et al., 2016; Office for National Statistics, 2024; van Dijk, 2016), apply to men as well as women, while others appear to be more gender specific. One example is the elevated risk among sexual and ethnic minorities, which appears to be more pronounced for women than for men (Bränström et al., 2023; Ceccato et al., 2021; Fridh et al., 2014; Hayes & Kopp, 2019; Palm et al., 2016).
Available research also points to differences with regard to victimization sites. For young women, the home is one of the most frequently reported sites of victimization, and offenders are often parents, partners, or siblings (Cater et al., 2014; Fridh et al., 2014; Källström et al., 2020). Men instead more frequently report being victimized in public places, such as streets or venues (Fridh et al., 2014), and by peers rather than family members (Cater et al., 2014; Källström et al., 2020).
As regards intimate partner violence, digital platforms have been found to be used to a greater extent as a tool for violence, control, and threats focused on young female victims (Korkmaz, 2021, Øverlien et al., 2020). In addition, the combination of low age and a lack of relationship experience makes it harder for younger female victims to leave their relationships (Kennedy et al., 2021; Korkmaz, 2021; Øverlien et al., 2020; Wesely, 2006). Moreover, women are more likely to be subjected to physical violence by their partner, while men are more likely to be subjected to psychological violence and coercive control (BRÅ, 2022a; Hayes & Kopp, 2019).
In summary, research shows that there are specific characteristics in the patterns of criminal victimization that target young women, and these are distinctly different from those found in relation to men of the same age and older women. However, with the exception of the study by Lilley et al. (2023), little to no research has focused on the observed changes in victimization rates over time. It remains unclear whether increases in self-reported victimization and the closing of the gender gap is due to actual changes in crime and victimization rates, or if it is due to changes in perceptions of crime. Consequently, the present study will contribute to this body of research by assessing whether the increased victimization rates in the Swedish Crime Survey are due to changes in perceptions of victimization among different cohorts of young women.
Gender Differences in Victimization Rates in the Swedish Crime Survey 2009 to 2016: A Reversed Gender Gap
The Swedish Crime Survey (hereafter the SCS) was first introduced in 2005 and was conducted in the form of a structured phone interview from 2005 to 2016, administered to a representative population sample of 20,000 people.
As illustrated in Figure 1, self-reported victimization rates among young women and young men were quite similar during the first 5 years of the SCS. During this period, there is a declining trend for men and women alike. However, the victimization rate among young women increases during the early- to the mid-2010s. The increase is not linear, with the rates dropping somewhat in 2012 and 2014, for example, and spiking in 2013 and 2016. Meanwhile, self-reported victimization rates among young men initially continue to decline, to then increase slightly in 2015 and 2016.
Self-reported victimization among women and men aged 16 to 24 years in the SCS 2005 to 2020.
The 2010s also coincided with an ongoing public debate regarding sexual offences, which resulted in two substantial legislative changes. The first took place in 2013, when the legal definition of rape was broadened to also include cases in which the victim had been passive, for example due to having a frozen fright reaction, having been asleep, drugged, or otherwise in a defenseless state (Justitiedepartementet, 2013). As can be seen from Figure 1, the introduction of this legislation coincides with the formation of the reversed gender gap.
As such, the early trend shown by the SCS does not point to higher victimization rates among young men when compared to young women. Instead, the trend points to similar victimization rates until 2012, whereafter we observe a reversed gender gap with higher rates being reported by young women than by young men. Most of the increase in the self-reported victimization among young women is comprised of sexual offences.
Since 2016, the SCS has undergone major methodological changes. For example, telephone interviews have been replaced by the use of an online survey, and the victimization questions have been revised to include a broader set of victimization items. Consequently, although two parallel data collections were conducted in 2016, using both data collection methods, the two survey periods are not comparable to one another (BRÅ, 2018). In the new SCS, the reversed gender gap is even more pronounced, with young women reporting victimization rates that are 10% to 17% higher than those of men (BRÅ, 2018). The gap was slightly larger during the #metoo year of 2017 and in the following year, during which a consent-based law for sexual offences was introduced in Sweden (BRÅ, 2022b; Pollack 2019). While not comparable to the earlier SCS, results from the new SCS show that the reversed gender gap between men and women has persisted in Sweden.
Since the purpose of this study is to assess whether the formation of the gender gap coincides with a decline in measurement equivalence, the study focuses on the 4-year period during which the gap developed, 2013 to 2016. We also include the 4 years prior to this, 2009 to 2012, when victimization rates among women and men were more similar, for statistical comparison.
Theory
The high self-reported victimization rates among women in Sweden and the other Nordic countries are sometimes theoretically conceptualized as a “Nordic paradox,” which refers to the combination of high levels of both gender equality and interpersonal victimization against women (Gracia & Merlo, 2016; Humbert at al., 2021). In the present study, the Nordic paradox framework’s concept of lowered tolerance will be used to interpret and assess the results.
One possible explanation for the comparatively high victimization rates found in the Nordic countries emphasizes that higher rates of equality might result in a lowered tolerance for various forms of infringements of personal integrity (Gracia et al., 2019; Gracia & Merlo, 2016; Humbert et al., 2021). Consequently, it has been theorized that higher rates of equality lead to a tendency to classify a broader set of behaviors as violative and criminal. This potential explanation was assessed and rejected by Gracia et al. (2019) in a study that assessed the measurement equivalence in the FRA, a European survey on domestic violence, between Sweden and Spain. However, a more recent study using the same dataset found that a large bulk of the differences in victimization rates were due to experiences of violence in previous relationships. In sum, the Swedish women had accumulated experiences from a larger number of previous relationships than the Spanish women, and had thus had a higher risk exposure. The number of violent relationships distributed over the total number of previous relationships was more similar. Further, Humbert et al. (2021) conducted a multilevel analysis on the results of the FRA, adding controls for methodological differences in how the FRA was carried out between countries, socio-cultural factors such as beliefs about violence against women, and individual factors such as relationship status, citizenship, and educational level. Their results showed that when these controls were included, gender equality lost its significance, and the authors highlight the need for more research on the tendency to disclose victimization (Humbert et al., 2021).
With regards to the increase in sexual victimization in the SCS, it might be the case that growing gender differences in perceptions about acceptable ways of showing sexual interest and perceptions of adequate dating and relationship behavior (Wemrell et al., 2019) might be contributing to higher rates of self-reported victimization.
Method
Sample and Study Period
To assess whether there has been a decline in measurement equivalence among young female respondents, we use data from the Swedish Crime Survey (SCS) from 2009 to 2016 regarding self-reported victimization among women aged 16 to 24 years. As such, the period selected for assessment includes the final 4 years, 2013 to 2016 years, of the old SCS and captures the period during which the reversed gender gap in the SCS emerges. It also includes 4 years prior to the development of the gender gap, 2009 to 2012, a period during which the victimization rates among young men and women were similar.
During the selected period, 2009 to 2016, the Swedish Crime Survey was conducted in the form of an annual, cross-sectional survey distributed to a representative population sample in Sweden, using a structured phone interview.
As shown by Table 1, the sample size declines over time, primarily due to a reduction in response rates. However, the distribution with regard to age and occupation remains similar over time. The largest proportion of the sample self-report being students, and are engaged in upper secondary school, college, or university education.
Sample Size, Age, Occupation, and Total Victimization Rate.
The SCS does not collect data on sexual orientation, religion, ethnicity, or nationality. According to the Swedish Personal Data Act, which was applicable during the period examined, it is prohibited to collect and process personal data that include race or ethnic origin, political opinions, religion or beliefs, union membership, health, and sexuality (§ 13). Although exceptions from this law can be made for research purposes, the National Council for Crime Prevention, which is responsible for the SCS, is not exempt from the legislation. One of the reasons for the legislation is to protect minorities from potential political persecution, for example by an external repressive regime. The legislation has since been replaced by the General Data Protection Regulation, but the prohibition clause that applies to the National Council for Crime Prevention and the SCS remains.
Ethical Considerations
The project has been approved by the Swedish Ethical Review Authority, case number 2022-01921-01. The data were retrieved from the National Council for Crime Prevention on an encrypted USB, and have been stored on external encrypted hard drives throughout the project. The SCS does not contain sensitive personal data in the form of social security numbers, but contains anonymous, individual-level data that include a few background variables regarding the respondents’ housing situation, occupation, and family, and thereafter responses to a larger set of questions regarding victimization, fear of crime and worry, and trust in the justice system.
Measures
The victimization variables employed in the study are presented below. In the SCS, victimization is measured using an initial victimization question, followed by a question about how many times the respondent had been victimized. In this study, we use the variable that designates how many times the respondent was victimized. Due to the presence of outliers in the dataset, we limited the range of the scale to 6, with the final number representing having been victimized six or more times.
Robbery
Did anyone rob you or try to rob you using threats or violence in the last year?—How many times? (0 = 0 times, 1 = once, 2 = twice, 3 = three times, 4 = 4 four times, 5 = 5 times, and 6 = 6 or more times)
Assault
Did anyone intentionally hit, kick or subject you to any other physical violence, so that you were injured or hurt, in the last year?—How many times? (0 = 0 times, 1 = once, 2 = twice, 3 = three times, 4 = 4 four times, 5 = 5 times, and 6 = 6 or more times).
Sexual Offences
Did anyone sexually assault, coerce or attack you in the last year? This includes both serious and less serious incidents, for example at home, at work, at school or in a public place.—How many times? (0 = 0 times, 1 = once, 2 = twice, 3 = three times, 4 = 4 four times, 5 = 5 times, and 6 = 6 or more times).
Fraud
As a private person, were you criminally cheated out of money or other valuables last year?—How many times? (0 = 0 times, 1 = once, 2 = twice, 3 = three times, 4 = 4 four times, 5 = 5 times, and 6 = 6 or more times)
Unlawful Threats
Were you threatened in the last year in such a way that you were afraid?—How many times? (0 = 0 times, 1 = once, 2 = twice, 3 = three times, 4 = 4 four times, 5 = 5 times, and 6 = 6 or more times)
Analytical Strategy
We use multi-group confirmatory factor analysis (MGCFA) to assess the measurement equivalence for the victimization items in the SCS. MGCFA is used to examine whether a set of variables has a common underlying factor across groups and is an established method that has been used in previous studies to assess measurement equivalence over time and in cross-cultural contexts (Millsap & Everson, 1993; Schmitt, 2011). If respondents have the same perception of a set of variables, in this case victimization variables, the statistical relationship between the variables and the underlying factor should be similar over time.
For interpretational reasons, years are compared two by two, resulting in a total of 28 combinations for each model. Firstly, we present the configural factorial invariance models, which place the least constraints on the data and which are used as a baseline. Secondly, we present metric invariance models in which the factor loadings of each variable are fixed between the years being compared. Finally, we present scalar invariance models in which the intercepts for each variable are fixed between the years being compared.
For all of these models, we present the Comparative Fit Index (hereafter CFI) and the Root Mean Square Error of Approximation (hereafter RMSEA) model fit indices. For interpretation, we use a cut-off value of equal to or above 0.9 for a good model fit using the CFI, and equal to or below 0.05 to indicate a good model fit using RMSEA (Byrne, 1994).
In order to assess the value change of the CFI and RMSEA between the models, we also present ΔCFI and ΔRMSEA in the metric and scalar models. Following recommendations for larger sample sizes, we have chosen to use values equal to or below 0.01 as a cut-off for good model fit for the ΔCFI (Cheung & Rensvold, 2002; Wu et al., 2007), and values equal to or below 0.015 as a cut-off for good model fit for the ΔRMSEA (Chen, 2007).
Lastly, in order to assess whether any observed variation is due to poor measurement in any specific variable, we also present χ² difference tests for the scalar models. During these tests, the restraints placed on the intercepts are released for one variable at the time. Thereafter, a χ² difference test is conducted on the model fit before and after removing the restraint to assess whether the model fit is improved significantly when the constraints on each variable are released. If there is a significant improvement in the model when the constraint for a variable is released, this implies that the variance is primarily due to variation in this specific variable.
Based on previous research and theory (Gracia et al., 2019; Gracia & Merlo, 2016; Humbert et al., 2021; Wemrell et al., 2019), we hypothesize that measurement equivalence will decrease over time. We expect to observe little to no variation from 2009 to 2012, and increasing variance from 2013 to 2016 negatively impacting the model fit as more constraints are introduced. Additionally, since a large proportion of the increase in self-reported victimization among young women is comprised of sexual offences, we hypothesize that variation in this variable will be the main contributor to variance during the earlier and later years of the study period.
Results
The CFI and RMSEA values for the configural factorial invariance models (Table 2) point to a good model fit for each set of years. As such, these baseline models initially point to invariance across years. However, the threshold for achieving a good model fit in the configural factorial invariance models is low. In order to achieve a good model fit in these models, the number of factors needs to be equal between comparisons. In our models, the factor is set to one which means that these models cannot fail this criterion. Secondly, in order to achieve a good model fit, the number of zero and non-zero loadings of the factor needs to be the same. Thus, if one or more of the regression weights between the victimization items and the factor is zero at year 1, they need to be zero at year 2 as well. These models do not provide information as to whether the variables in the models contribute equally to the underlying factor over time. Instead, the contribution of the included variables to the underlying factor is assessed in the next group of models, the metric invariance models.
Multi-group Confirmatory Factor Analysis.
In the metric invariance models the CFI tests show a poor model fit for nearly half of the models (13 out of 28), and the RMSEAs show a poor model fit in four of the models. More concretely, a poor model fit in the metric invariance models shows that the regression weights between one or more variables and the underlying factor vary at year 1 and year 2 in the comparisons. As such, the poor values of the CFI and the RMSEA show that there is variation in the statistical relationship between the underlying factor and the victimization variables in several of the models.
Moreover, the ΔCFI shows that the decrease in the CFI values for all but two sets of years (2009–2012 and 2010–2012) is large (≤0.01), even for several sets of years in which the CFI measure nonetheless exceeds the cut-off value for good model fit. In contrast, the ΔRMSEA only reaches or exceeds the cut-off value for six sets of years, showing that the changes in the RMSEA are small for most sets of years.
While the results from the metric invariance models show variance across several sets of years, the results do not point to any clear linear trend such that the models become poorer as the time-span between the years examined in each model becomes larger. Consequently, the observed variance with regard to how the items load on the factor does not appear to be connected to a gradual change in how the questions are perceived or answered over time, alternative explanatory models are required. As such, the results from these models lead to a rejection of our hypotheses that equivalence would decline over time.
Finally, the CFI in the scalar invariance models show that there is a poor model fit in more than two-thirds of the models (20 out of 28). In contrast, most RMSEA values still indicate good model fit, with only five models showing a poor model fit.
A poor model fit in the scalar invariance models shows that the intercepts for one or more variables differ at year 1 when compared to year 2. Therefore, these models show that there are differences between the variables at different times.
In contrast to the metric invariance models, the ΔCFI and ΔRMSEA of the scalar invariance models seems to point to a worsening trend over time, with poorer values being concentrated to comparisons between early and late years. For example, the comparison of 2016 and 2015 produces a good model fit, while the model fit is poor in comparisons with all previous years. A similar pattern, albeit not always as clear, can be observed for the comparisons with previous years conducted for 2015, 2013, 2011, and 2010. However, the results from comparisons of 2014 and 2012 with previous years do not point to a linear decline in model fit.
In sum, the model fit indices of the scalar invariance models show that there is variation in the intercepts in nearly all models, and the ΔCFI and ΔRMSEA show that the variation seems to increase over time. These results show that there is some support for a linear pattern of poorer model fit over time with regard to the variable intercepts.
Finally, we conducted χ² difference tests (Table 3) to assess which of the variables vary over time in the scalar models, and whether the variation in these variables has a significant negative impact on the overall model fit.
χ² Difference Tests, Partial Scalar Models.
p-value ≤ .05. **p-value ≤ .01. ***p-value ≤ .001.
Interestingly, the tests show that the model fit would improve significantly over time if the sexual offence variable were to be excluded from the models. As illustrated in Table 3, the differences are non-significant for the period of 2009 to 2012 when the victimization rate remains relatively similar. The significant worsening of the model fit due to the sexual offence variable occurs in the comparisons from 2013 to 2016, which thus coincides with the increase in self-reported victimization. As such, these results support our hypothesis, since they point to a change in the perception of sexual offences among young women, rather than an actual increase in the victimization rate.
Aside from the sexual offence variable, there is also a worsening with regard to the assault variable for the years 2015 and/or 2016 in the models for 2009, 2010, 2012, and 2014. By contrast with the sexual offence variable, self-reported experiences of assault are relatively stable throughout the period.
For the fraud variable, the model fit for the sets 2009 to 2011 and 2009 to 2013 would be slightly improved if this variable were removed. However, the variable does not lead to any significant worsening of the models for any of the other 26 sets of years.
Finally, the tests show that the change in model fit is small and insignificant for all sets of years with regard to the variables robbery and unlawful threat. This means that the models would not be improved by the removal of either of these variables, and that the observed variation is not caused by either of them.
In sum, the results of the analysis show some support for a linear pattern of increasing variance over time. More specifically, the results show that a linear pattern in the level of variance between the items can be seen in different measurement years, but not with regard to the relationship between the items and the underlying factor.
Discussion
The aim of the present study has been to examine whether there has been a change in measurement equivalence across different cohorts of young female respondents (ages 16–24 years) who have answered the Swedish Crime Survey from 2009 to 2016. More specifically, we used MGCFA to assess whether the increase in self-reported victimization among young women is the result of changed perceptions about inter-personal crime, and sexual offences in particular. Based on previous research and theory (Gracia et al., 2019; Gracia & Merlo, 2016; Humbert et al., 2021; Wemrell et al., 2019), we hypothesized that lack of measurement equivalence would be observed when comparing the first 4 years of the selected period with the last 4 years. We also hypothesized that reductions in measurement equivalence and reductions in model fit over time would be primarily due to the sexual offence variable.
Although the results from the metric models did not show a clear pattern with increasing variation over time, this trend was clearer in the scalar models and in the χ² differences tests that assessed the influence of model fit one variable at the time. As such, the results largely support our hypotheses, showing that most of the variation observed is due to changes in how young women perceive sexual offences. Interestingly, the results also point to variation in the assault variable, which in contrast to sexual offences remained at a relatively stable prevalence during the period examined. In sum, based on how MGCFA-tests are conventionally interpreted in relation to self-report victimization data (Gracia et al., 2019; Yount et al., 2022), the results support the notion that the increased victimization rate among young women is the result of changing perceptions about victimization, and sexual victimization in particular, rather than an actual increase.
Previous Research and Implications for Policy
As pointed out in the background section, the crime trends in the SCS manifest a reversed gender gap, not merely a closing of the gender gap such as has been observed in other countries (Office for National Statistics, 2024; Lilley et al., 2023; van Dijk, 2016). As Lilley et al. (2023) illustrated in their study of the NCVS data, the decline in the victimization rate among men is related to fewer school dropouts, an increased number of police officers, improved economic conditions, and larger proportions of women in the work force. The authors note that large-scale preventive measures have been functional in addressing and prohibiting crime involving male offenders and male victims, but that the decline in crime directed against women has been small (Lilley et al., 2023). In general, these large-scale preventive measures have reduced opportunities for men to commit crime against other men but have not been geared toward, nor effective in, countering the kinds of crime that target women.
A similar pattern regarding the scope of crime policy is also found in Sweden. The Swedish crime policy debate has been dominated by a focus on serious gang-related crime between groups of young men (Brodin Danell & Jarl, 2024). When crime prevention policy aims to reduce victimization among women, the focus areas are domestic violence and honor-related violence, a direction illustrated in both current and past prevention programs (BRÅ, 2010; Regeringskansliet, 2024; SOU, 2014). However, according to the SCS, only a small proportion, approximately 10% during the period examined, of young women report being victimized by an offender with whom they are close or to whom they are related. Consequently, crime policies that focus entirely on crimes that take place in such relationships are unlikely to have a broader preventive impact, since the majority of self-reported victims are subjected to crimes that take place in other circumstances.
Moreover, many of these policies are not preventive per se. Rather than reducing the opportunities, incentives, and the driving forces that cause domestic violence, they primarily comprise various forms of victim support. Like previous prevention programs (BRÅ, 2010; SOU, 2014), the recent Swedish National Policy to Counter Violence Against Women for 2024 to 2026 (Regeringskansliet, 2024) places great emphasis on improving exit-programs to facilitate the process of leaving violent relationships and improving measures to detect and support victims. Under the category of crime prevention, the government only presents reactive measures, such as harsher penalties, increased criminalization, and measures to improve the treatment of victims by police investigators (Regeringskansliet, 2024). Such measures may be important in offering support and access to justice for victims, but they do not target the causes of victimization.
Further, when it comes to countering crime that targets young women, there appears to be a theoretical gap within the criminological field. We argue that the study by Lilley et al. (2023) illustrates how so-called gender-neutral theories and so-called gender-neutral crime prevention primarily have an impact on men’s crimes against other men. Consequently, there is a need for further large-scale quantitative studies as well as in-depth qualitative studies into the victimization patterns experienced by young women. For example, these could examine lifestyle and activity patterns among young women to identify gender-specific risk environments. Such studies are needed to form the basis for more strategic crime prevention measures and to develop relevant theoretical concepts and frameworks.
Theory and Future Research
In the present study, we found support for the notion that young women in Sweden have developed a lower tolerance (Gracia et al., 2019; Gracia & Merlo, 2016; Humbert et al., 2021) for various forms of sexual behavior over time. In part, the results show that it is the sexual offence variable that causes most of the variation in the analyses, and that this variation increases over time. The item used in the SCS for sexual offences is broadly formulated when compared to the other victimization items. As such, this item is sensitive to changes in perceptions, and it could be beneficial to assess the measurement equivalence in the new survey used from 2016 onwards.
It should be noted that the increase in self-reported sexual victimization among young women coincided with a broadening of the legal definition of rape to include incidents in which the victim was in a defenseless state (Justitiedepartementet, 2013). Moreover, an extensive public debate regarding explicit consent and sexual autonomy followed in the wake of the 2013 revision of the legal framework for sexual offences, which would later culminate in the introduction of consent-based legislation in 2018 (BRÅ, 2022b; Pollack, 2019). It may be the case that the introduction of the 2013 legislation, along with the extensive public debate, contributed to broader changes in perceptions of sexual offences.
Consequently, we recommend that future research should assess the impact of such significant events on victimization rates in order to further assess self-reported victimization. In the Swedish case, it would be relevant to assess whether the #metoo movement, which had a substantial impact on the public debate in 2017 (Pollack, 2019), and the introduction of the new consent-based rape legislation in 2018 (BRÅ, 2022b) have had an impact on responses to the new SCS.
In line with Humbert et al. (2023), we see multi-level analysis as a possible tool to further explore changes in victimization rates over time. In the context of such an analysis, the SCS could be supplemented with indicators of equality at the municipal level, which might serve to complement existing qualitative research in this area. On the question of whether high levels of equality lead to a change in levels of tolerance for different offences, we instead recommend qualitative studies among different cohorts of women to investigate these further. In addition, it would be highly relevant to examine the observed measurement variance in physical assault in more detail. Based on studies focusing on values (Oskarsson & Ahlbom, 2021), however, it is possible that there is still a gender gap in the way young men and young women view “milder” forms of sexual offending, which constitutes an important area for future research.
Strengths and Limitations
One of the main limitations of the SCS is that it relies on one victimization item per offence category. From a methodological standpoint it would be more beneficial if the SCS were to include a set of items for each offence category. In the case of sexual offences and assault, which both contributed to variation in the models, this would have allowed us to assess which forms sexual offences and assault vary over time.
On the other hand, the sample size and the proportion of victims in the sample provide good statistical power. Additionally, the SCS data provide the opportunity to assess cohort differences over time. At the same time, our inclusion of an 8-year assessment period limits the level of detail at which the data can feasibly be presented. In this study, we have chosen to focus on model fit indices and chi-square difference tests, which we held to be most relevant based on our aims.
Since the SCS does not collect data on ethnicity, nationality, religion, or sexual orientation, it is not possible to use this survey as a basis for in-depth studies of groups that previous research has shown to comprise risk groups within the group of women (Bränström et al., 2023; Ceccato et al., 2021; Fridh et al., 2014; Hayes & Kopp, 2019; Palm et al., 2016). For this reason, the existing knowledge will need to be supplemented using targeted studies of subgroups of women who may be particularly vulnerable.
Finally, we need to highlight statistical aspects regarding the interpretation of the MGCFA when analyzing measures based on the perceived prevalence of incidents, such as victimization. The method has previously been used primarily in fields other than criminology, such as psychology and education. In these fields, MGCFA is typically used to assess emotional conditions, such as stress, or individual characteristics, such as diligence. This raises questions about whether it is correct to interpret variance as a difference in the way items are interpreted, rather than as an indicator of an actual increase or decrease, when the method is applied to studies focused on the prevalences of victimization. In the present study, the results show that variance over time is primarily found in the scalar models, and more specifically in the item for sexual offenses. Since these models compare the intercepts for the same item from year to year, we interpret this as a good reason to interpret the results in line with the interpretations found in previous studies on victimization that have used MGCFA (Gracia et al., 2019; Yount et al., 2022), that is, that the results indicate a change in perceptions of the item itself rather than a real increase in victimization. Additionally, the increase in self-reported sex crime victimization coincided with a legal reform introduced in 2013 and an extensive public debate on consent-based sexuality.
Conclusions
The results of the present study support the notion that there has been a change in the degree of measurement equivalence across different cohorts of young Swedish women. Moreover, we found that some of the probable roots of these changes may be found in societal shifts in the sex crime legislation and an extensive political debate on sex crime, which coincided with the increase in self-reported victimization. We also found that established Swedish crime policy is ill-suited to countering and preventing the types of victimization reported by those most frequently targeted: young women. As such, this is an area in need of substantial development, ranging from building a proper knowledge base to the development of adequate theory and prevention measures.
MGCFA is a powerful, albeit underused, statistical tool for assessing changes in crime rates. For example, if the results of the present study had not showed differences in measurement equivalence over time, the results would have instead strongly supported the presence of a real increase in victimization rates. Moreover, by using MGCFA, we can assess whether variance is due to changes in the statistical relationship between the items and the underlying factor, and/or between the same item at different times or for different groups. While the interpretation of measurement variance in prevalence items might not be as straightforward as when applied to emotional conditions or individual characteristics, it comprises a valuable statistic tool for assessing both changes in perceptions of crime and changes in crime patterns, especially when combined with additional data on macro-level politics, the media, lifestyle patterns, and societal changes.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The study was funded by The Nordic Research Council for Criminology.
