Module 1.3 Notes on international crime declines (ICVS data, 1988–2004)

Data and Data Source

• Study uses International Crime Victims Survey (ICVS) data from sweeps: 1989, 1992, 1996, 2000, 2005.

• Sample: 26 countries; 5 crimes (theft from cars, theft from person, burglary, assault, car theft).

• Measures:

  • Incidence: crimes per 100 possible victims; formula: $\text{Incidence} = \frac{C}{V} \times 100$

  • Prevalence: victims per 100 population; formula: $\text{Prevalence} = \frac{\text{victims}}{N} \times 100$

  • Concentration: crimes per victim; formula: $\text{Concentration} = \frac{C}{V}$

• Data limitations and choices:

  • Unweighted data used; weighting can bias country-level rates less but is often small.

  • ICVS samples may bias upward/downward due to city sampling, response rates, and overrepresentation of urban residents.

  • 99 country–year observations (3+ sweeps per country) form the analytic panel; overrepresentation of Northern Europe and North America.

• Key sample details:

  • 5 ICVS sweeps; 1989, 1992, 1996, 2000, 2005.

  • Countries meeting criteria: 26; only 5 countries participated in all sweeps (Canada, England & Wales, Finland, USA, Netherlands).

• Centering and controls in modelling:

  • Time trend centered around 1995; non-linear trend allowed for some crimes.

  • City booster variable accounts for overrepresentation of city residents (cities have higher crime rates).

  • Former Communist Country (FCC) dummy controls regional differences post-communism.

• Key findings structure:

  • Main international trends by crime type (incidence, prevalence, concentration).

  • Whether changes reflect changes in victims vs repeat victimization.

  • Cross-country and cross-crime-type relationships and variance components.

Main International Trends (1988–2004)

• Post-1995 international crime declines across multiple countries and crime types.

• Estimated mean declines (1995–2004):

  • Theft from cars: $77.1\%$

  • Theft from person: $60.3\%$

  • Burglary: $26.0\%$

  • Assault: $20.6\%$

  • Car theft: $16.8\%$

• Across countries, slope of over-time changes was broadly similar (no strong country-level differences in the main trend).

• Regional nuance: burglary incidence/prevalence declined faster in Latin America, Africa, and Asia than in Europe, North America, and Australia.

• Trends by measure:

  • Incidence and prevalence generally declined; concentration declined for most crimes except some patterns vary by crime type.

• Visuals referred: Figure 1 (model estimates of incidence), Figure 2 (indexed trend), Figure 3 (unadjusted vs adjusted declines).

Role of Repeat Victimization

• Repeats contributed to declines for most acquisitive crimes; the pattern differs by crime type:

  • Theft from cars: repeat incidents dropped by $28.1\%$ (1995–2004).

  • Theft from person: repeats dropped by $13.2\%$ (1995–2004).

  • Burglary and assault repeats: international concentration did not consistently fall; repeat burglary showed no clear global relationship to burglary levels; repeat assaults did not show a significant drop.

  • For car theft, repeat incidents tended to be more influential earlier in the period (end of the 1990s).

• Overall, reductions in repeat victimization helped explain a substantial portion of the international declines, particularly for acquisitive crimes.

• Policy implication: reducing repeat victimization can be a meaningful component of crime reduction strategies.

Methodology (Multilevel Modelling)

• Analytic framework: multilevel models to separate time trends from sample and regional variation.

• Levels: time (within-country over years) and country (across-country differences).

• Main features:

  • Non-linear trends for some crimes; linear trend for burglary and theft from cars.

  • Additional controls: city booster (percent overrepresentation of urban samples) and FCC (former communist country).

• Outcome measures estimated for 1995–2004 period:

  • Incidence, prevalence, and concentration across five crime types.

• Key advantage: accounts for sampling differences, unobserved country-specific factors, and time correlations.

Cross-Country Variation and City FCC Effects

• City booster effects:

  • Each 1% increase in city overrepresentation raises incidence for some crimes (e.g., theft from cars) by about $0.11$ per 100 car-owning households.

  • The city booster generally elevates repeat victimization statistics; however, the overall decline pattern remains consistent with the general trend.

• Former Communist Countries (FCC):

  • In 1995, estimated thefts from cars per 100 car-owning households were higher in FCCs (approximately 14.4) than the non-FCC international mean (approximately 9.3), reflecting higher baseline rates.

  • FCCs showed similar downward trends as other countries; higher baseline levels did not imply a different trajectory post-1995.

• Summary: despite different sampling and historical contexts, the overall downward trend was robust across country groups, with city sampling and FCC status affecting absolute levels rather than the direction of change.

Relationships Between Crime Types

• Cross-national correlations (incidence) among crimes (Table 2):

  • Burglary with theft from car: r ≈ 0.40

  • Theft from car with car theft: r ≈ 0.64

  • Burglary with assault: r ≈ 0.58

  • Burglary with theft from person: r ≈ 0.92

  • Theft from car with theft from person: r ≈ 0.51

  • Assault with theft from person: r ≈ 0.38

• Cross-national patterns: acquisitive crimes (burglary, car crime, personal theft) tend to co-vary across countries; assault co-varies more modestly with other acquisitive crimes.

• Within-time (residual) correlations (below-trend relationships):

  • Theft from cars and personal theft co-vary over time; car theft appears more independent from other crimes within years.

• Interpretation: high cross-national covariation among acquisitive crimes suggests common underlying drivers; assaults show weaker cross-national linkage to acquisitive crimes.

Summary of Variance and Predictability

• Variance components indicate: overall trend explains a substantial portion of between-country differences for some crimes, but not all.

• For theft from cars and personal theft, trends explain a large portion of both between-country and over-time variance; assaults show substantial residual (unexplained) variation, indicating country- or time-specific factors may matter more for assault.

• Concentration (crimes per victim) tends to vary more over time than between countries.

Key Conclusions and Implications

• Universal downward trend: after 1995, crime fell across many countries and most crime types, particularly theft from cars and theft from person.

• Burglary declined more in non-Western regions (Latin America, Africa, Asia) than in Europe/North America/Australia.

• Repeat victimization contributed meaningfully to declines in most acquisitive crimes; policy should address revictimization to sustain reductions.

• The decline pattern was broadly consistent across countries, suggesting universal or widely shared factors (e.g., improved security, market conditions, policing approaches).

• Assaults did not track the same cross-national pattern as acquisitive crimes, indicating a need for country- or time-specific analysis to understand belated reductions.

• Policy takeaway: focus on reducing repeat victimization and improving security measures that affect repeat offenses (e.g., vehicle security), while recognizing that different crime types may require tailored approaches.

Notes on Data and Limitations (quick reference)

• Data limitations exist due to sample size per country-year and city overrepresentation; however, the trend findings are robust to these concerns given cross-national replication.

• The analysis emphasizes trends over exact country-specific rates, using multilevel modelling to separate common trends from residual variation.

• Appendix data provide country-year participation and descriptive statistics; some outliers were treated and sensitivity analyses suggested similar conclusions when omitted.

Quick Reference Figures/Takes

• Central trend: 1995 as a pivot; post-1995 declines across five crime types.

• Largest international declines (1995–2004): $ext{Theft from cars}: 77.1\%$, $ext{Theft from person}: 60.3\%$, followed by burglary, assault, car theft.

• Repeat victimization contributed significantly to international declines in most acquisitive crimes; less clear for assaults.

• City sampling and FCC status modify levels but not the basic pattern of decline.

Notable Equations (recap)

• Incidence: $I = \frac{C}{V} \times 100$

• Prevalence: $P = \frac{\text{victims}}{N} \times 100$

• Concentration: $K = \frac{C}{V}$

• Car-related trend center: time trend centered at $t=1995$ with possible non-linearity for some crimes.

Methodology (Multilevel Modelling)

• Analytic framework: Multilevel models were employed to meticulously separate time trends from various sources of variation, including sample-specific characteristics and regional differences. This approach allows for a more robust analysis by accounting for the hierarchical structure of the data.

  • Levels of analysis: The models operated on two primary levels:

    • Time (within-country over years): This level captures the longitudinal changes in crime rates within each individual country over the study period.

    • Country (across-country differences): This level accounts for the inherent differences in crime rates and trends observed between different countries.

• Main features of the models:

  • Trend specification: The models allowed for non-linear trends for some crime types, providing flexibility to capture more complex temporal patterns. For crimes such as burglary and theft from cars, a linear trend was deemed sufficient.

  • Additional controls: To refine the analysis and account for known biases or differences, the following variables were included:

    • City booster: This variable addressed the overrepresentation of urban samples in the ICVS data. It accounts for the percentage of city residents in the sample, recognizing that urban areas often exhibit higher crime rates.

    • FCC (former communist country) dummy: This control variable was used to account for potential regional differences that emerged in countries transitioning from communism, which could influence crime rates.

• Outcome measures estimated for the 1995–2004 period:

  • The multilevel models were used to estimate key crime measures across five distinct crime types (theft from cars, theft from person, burglary, assault, car theft):

    • Incidence: crimes per 100 possible victims, calculated as $\text{Incidence} = \frac{C}{V} \times 100$

    • Prevalence: victims per 100 population, calculated as $\text{Prevalence} = \frac{\text{victims}}{N} \times 100$

    • Concentration: crimes per victim, calculated as $\text{Concentration} = \frac{C}{V}$

• Key advantage: The application of multilevel modelling offers significant benefits:

  • It effectively accounts for sampling differences across countries and surveys.

  • It addresses unobserved country-specific factors that might otherwise confound results.

  • It manages time correlations within countries, leading to more accurate estimation of trends and relationships.