Introduction

Does where you live affect how safe you feel?

While many factors contribute to feelings of safety, this study focuses on the connection between socioeconomic deprivation and fear of attack based on skin colour, ethnic origin, or religion. Research in the UK has increasingly examined the relationship between neighborhood characteristics and perceptions of safety. For instance, studies by Webster, Simpson, and Quirk (2014) have explored the impact of area deprivation on social cohesion and fear of crime, highlighting the challenges faced by residents in disadvantaged communities. Other research has focused on the experiences of specific ethnic and religious groups, finding that fear of hate crime is often heightened in areas with histories of racial tension or social segregation (Phillips & Bowling, 2002; Spalek, 2010). Moreover, Bowling and Phillips (2002) have shown how perceptions of discriminatory policing can further erode trust and increase fear among minority communities. This study aims to further explore the specific intersection of socioeconomic disadvantage and the fear of racially/ethnically/religiously motivated attacks in Great Britain. Using data from the British Cohort Study, we investigate whether individuals residing in more deprived areas experience a greater fear of such attacks. Understanding this relationship is crucial for addressing safety concerns in disadvantaged communities.

Data and Methods

Data Source

We use data from the Crime Survey for England and Wales (CSEW) 2017-2018 Teaching Dataset. The CSEW (formerly known as the British Crime Survey) is a large-scale, nationally representative victimisation survey conducted annually by the Office for National Statistics (ONS) (ONS, 2023). It provides valuable insights into crime trends, public perceptions of crime, and experiences of victimisation in England and Wales (Flatley et al., 2010). The CSEW Teaching Dataset is a subset of the full CSEW data, designed for educational purposes, and contains a range of variables related to crime, demographics, and attitudes.

The 2017-2018 survey includes questions about fear of crime, perceived risk of victimisation, and demographic characteristics that allow us to examine the relationship between socioeconomic deprivation and fear of racially/ethnically/religiously motivated attacks. While the Teaching Dataset has limitations in terms of the scope and detail of the available variables compared to the full CSEW, it provides a valuable and accessible resource for exploring these issues. A key strength of the CSEW is its ability to capture experiences of crime that may not be reported to the police, providing a more comprehensive picture of crime and victimisation than official police statistics alone.

Variables

The primary variables of interest are:

  • wraceatt: This variable measures the respondent’s level of worry about being attacked because of their skin colour, ethnic origin, or religion. It is measured on an ordinal scale with the following categories: 1 = Very worried, 2 = Fairly worried, 3 = Not very worried, and 4 = Not at all worried. (Module C).

  • tcemdiqu2: This variable represents the Index of Multiple Deprivation in England. It is measured as a quintile, ranking areas from most to least deprived (1 = most deprived, 5 = least deprived).

  • Sex: This is a demographic variable indicating the respondent’s gender. The categories are: 1 = Male, 2 = Female.

  • Age: This variable represents the respondent’s age in years and is treated as a continuous variable. The age range in the dataset is from 16 to 101.

  • Ethnic: This variable indicates the respondent’s ethnic group. It is a 5-category variable that consists of White, Mixed, Asian or Asian British, Black or Black British and Chinese or Other.

  • Rural: This variable indicates whether the respondent’s area is classified as rural or urban, based on the 2011 Census definition. The categories are: 1 = Rural, 2 = Urban.

Data Preparation

The following steps were taken to prepare the data for analysis:

The following steps were undertaken to prepare the data for analysis, ensuring data quality and facilitating appropriate statistical modeling:

  • Variable Selection: Relevant variables (wraceatt, tcemdiqu2, sex, age, ethgrp2, rural2) were selected using dplyr::select().

  • Handling Missing Data: To ensure the robustness of the analysis, observations with missing values in any of the selected variables (wraceatt, tcemdiqu2, sex, etc.) were excluded. This was achieved using the filter(!is.na()) function in R. While this reduces the sample size, it prevents biased results that could arise from imputing or otherwise including incomplete data.

  • Transforming the Worry Variable: The original wraceatt variable, representing levels of worry about being attacked, was an ordinal categorical variable (e.g., “very worried,” “fairly worried”). To enable quantitative analysis, this variable was transformed into a numeric variable, wraceatt_numeric. The ordinal categories were mapped as follows:

  • “very worried” = 1

  • “fairly worried” = 2

  • “not very worried” = 3

  • “not at all worried” = 4

Fear of Attack Across Deprivation Quintiles: A Visual Analysis

To examine the distribution of fear of attack across different levels of socioeconomic deprivation, a stacked bar chart was generated using the data. The chart visually illustrates how worry about attack differs across deprivation quintiles. While the prominent red segments, representing those “Not at All Worried,” indicate a generally low level of concern across the population, the presence of smaller segments denoting “Very Worried” (dark green) and “Fairly Worried” (light green) highlights the reality of fear of attack for a significant subset.

By examining the distribution of these worry levels across deprivation quintiles, we can explore whether individuals in more deprived areas exhibit heightened concerns about their safety. Comparing the stacked segments across quintiles allows us to assess if those in disadvantaged communities experience a disproportionate level of fear, suggesting a link between socioeconomic circumstances and perceived vulnerability to crime.

Furthermore, the chart facilitates a comparison between genders, revealing potential differences in how fear of attack manifests across deprivation levels for men and women.

Permutation Test

To assess the difference in worry levels between the most and least deprived groups, a permutation test was conducted. This non-parametric test compares individuals residing in the most deprived quintiles (1 and 2) with those in the least deprived quintiles (4 and 5) regarding their levels of worry about being attacked.

Unlike parametric tests (e.g., t-tests) which rely on assumptions about the underlying distribution of the data, the permutation test makes no such assumptions. Instead, it works by repeatedly shuffling the group labels (deprived vs. not deprived) and calculating the difference in means for each shuffled dataset. This process generates a distribution of possible differences in means under the null hypothesis that there is no real difference between the groups.

The observed difference in means from the actual data is then compared to this distribution. The p-value represents the proportion of shuffled datasets that yield a difference in means as large or larger than the one observed in the original data. A small p-value indicates that the observed difference is unlikely to have occurred by chance, providing evidence against the null hypothesis and suggesting a statistically significant difference between the groups.

## 
##  Approximative General Independence Test
## 
## data:  wraceatt_numeric by as.factor(tcemdiqu2) (1, 2, 4, 5)
## maxT = 8.2371, p-value < 1e-04
## alternative hypothesis: two.sided

The extremely small p-value (p < 1e-04) provides strong evidence to reject the null hypothesis of no difference in worry levels between the most and least deprived groups. We can conclude that individuals residing in the most deprived quintiles (1 and 2) experience significantly different (specifically, higher) levels of worry about being attacked compared to those in the least deprived quintiles (4 and 5).

Ordinal Logistic Regression

We investigate the association between socioeconomic deprivation and fear of attack, acknowledging the complex interplay of individual and contextual factors. We fit an ordinal logistic regression models to analyse this relationship, while rigorously controlling for potentially confounding demographic variables. Fear of attack, as an ordinal variable with ordered categories (e.g., “not at all afraid,” “slightly afraid,” “moderately afraid,” “very afraid”), was modeled as a function of socioeconomic deprivation and demographic characteristics.

Two models were estimated:

The Likelihood Ratio Test will helps us to evaluate whether incorporating an interaction between age and sex improves our understanding of the relationship between tcemdiqu2 and the outcome variable.

## Call:
## polr(formula = as.factor(wraceatt_numeric) ~ tcemdiqu2 + sex + 
##     age + ethgrp2 + rural2, data = regression_data, Hess = TRUE)
## 
## Coefficients:
##               Value Std. Error t value
## tcemdiqu2  0.025396   0.014465   1.756
## sex       -0.260620   0.039381  -6.618
## age        0.009249   0.001078   8.583
## ethgrp2   -0.506621   0.027972 -18.112
## rural2     0.482234   0.050028   9.639
## 
## Intercepts:
##     Value    Std. Error t value 
## 1|2  -3.0538   0.1229   -24.8489
## 2|3  -1.9078   0.1163   -16.4073
## 3|4  -0.0746   0.1133    -0.6583
## 
## Residual Deviance: 20306.35 
## AIC: 20322.35

Model 1, examined the relationship between socioeconomic deprivation, gender, age, ethnicity, rural/urban residence, and worry about race-based attacks. The analysis revealed that socioeconomic deprivation, as measured by the Index of Multiple Deprivation, was not significantly associated with worry about race-based attacks (b = 0.025, p = 0.079). However, gender (b = -0.261, p < 0.001), age (b = 0.009, p < 0.001), ethnicity (b = -0.507, p < 0.001), and rural/urban residence (b = 0.482, p < 0.001) were all significantly associated with the outcome. Specifically, females were significantly more likely than males to express worry about race-based attacks. With each increasing year of age, respondents reported greater levels of worry. Residents of urban areas were significantly more likely to report being worried about an attack than rural residents. However, the relationship between ethnicity and worry about race-based attacks requires further clarification, as the negative coefficient and the coding of this variable will be explored and defined for a more precise interpretation.

## Call:
## polr(formula = as.factor(wraceatt_numeric) ~ tcemdiqu2 + sex * 
##     age + ethgrp2 + rural2, data = regression_data, Hess = TRUE)
## 
## Coefficients:
##               Value Std. Error  t value
## tcemdiqu2  0.025752   0.014467   1.7801
## sex       -0.508592   0.112685  -4.5134
## age        0.001311   0.003542   0.3701
## ethgrp2   -0.506411   0.027975 -18.1021
## rural2     0.484481   0.050038   9.6822
## sex:age    0.005049   0.002149   2.3494
## 
## Intercepts:
##     Value    Std. Error t value 
## 1|2  -3.4404   0.2056   -16.7350
## 2|3  -2.2940   0.2015   -11.3830
## 3|4  -0.4598   0.1994    -2.3058
## 
## Residual Deviance: 20300.81 
## AIC: 20318.81

Model 2 included an interaction term between gender and age to explore whether the effect of age on worry about race-based attacks differed for males and females. As in Model 1, socioeconomic deprivation was not significantly associated with the outcome (b = 0.026, p > 0.05), and the effects of ethnicity (b = -0.506, p < 0.001) and rural/urban residence (b = 0.484, p < 0.001) remained significant. The key finding in Model 2 was the significant interaction between sex and age (b = 0.005, p < 0.05), indicating that the positive relationship between age and worry about race-based attacks is stronger for females than for males.

## [1] "\nModel 1: Coefficient Interpretation (Log-Odds):"
##    tcemdiqu2          sex          age      ethgrp2       rural2 
##  0.025396209 -0.260619905  0.009249289 -0.506620689  0.482233710
## [1] "\nModel 2: Coefficient Interpretation (Log-Odds):"
##    tcemdiqu2          sex          age      ethgrp2       rural2      sex:age 
##  0.025751962 -0.508592391  0.001310653 -0.506410652  0.484480787  0.005049307

While the direction and statistical significance of the effects of tcemdiqu2, ethnic, and rural2 on wraceatt_numeric remain largely consistent between Model 1 and Model 2, the magnitudes of their coefficients change slightly due to the inclusion of the sex:age interaction term and potential redistribution of explained variance or correlations between these predictors.

Odds Rations

## [1] "\nModel 1: Odds Ratios (exponentiated coefficients):"
## tcemdiqu2       sex       age   ethgrp2    rural2 
## 1.0257214 0.7705738 1.0092922 0.6025283 1.6196883

Model 1: Odds Ratio Interpretation

  • tcemdiqu2 (Socioeconomic Deprivation): 1.0257

Interpretation: For every one-unit increase in the Index of Multiple Deprivation quintile (i.e., moving from a more deprived area to a slightly less deprived area), the odds of being in a higher worry category increase by a factor of 1.0257, or about 2.6%. However, keep in mind that this effect was not statistically significant in the previous analysis, so this small change is likely due to random chance. Despite the odds ratios, because the p values are higher than the standard significant measures, we will not interpret.

  • sex (Gender): 0.7706

Interpretation: If 1 = Male and 2 = Female: If the gender is male, that means that the odds of being in a higher worry category are 0.77 times lower for males compared to females, holding all other variables constant. Put another way, females have approximately 1/0.77 = 1.30 times greater odds (30% higher odds) of being in a higher worry category than males.

  • age (Age): 1.0093

Interpretation: For every one-year increase in age, the odds of being in a higher worry category increase by a factor of 1.0093, or about 0.9%. So, as people get older, they are slightly more likely to express greater worry about race-based attacks.

  • ethgrp2 (Ethnicity): 0.6025

Interpretation: This is where you need to be very careful about your coding. Assuming that you are comparing white (0) to non white (1), The odds of being in a higher worry category are 0.60 times lower for non whites compared to whites. Since the odds are 40% lower, non-white people are less likely to worry about a race based attack.

  • rural2 (Rural/Urban Residence): 1.620

Interpretation: the odds of being in a higher worry category are 1.62 times lower for those living in rural areas compared to those living in urban areas, holding all other variables constant. Put another way, urban people are 1/1.62 = .617 times likely to be concerned about a race based attack, so urban areas are 38 percent more likely to be worried about such attacks. Therefore, the people living in urban areas have a greater concern for race based attacks in comparison to those living in rural areas.

## [1] "\nModel 2: Odds Ratios (exponentiated coefficients):"
## tcemdiqu2       sex       age   ethgrp2    rural2   sex:age 
## 1.0260864 0.6013414 1.0013115 0.6026548 1.6233319 1.0050621

The interpretation of the odds ratios in Model 2 requires careful consideration of the interaction term. It is crucial to understand that the effects of sex and age are intertwined, and their individual effects depend on the value of the other variable in the interaction.

Likelihood Ratio Test

The likelihood ratio test compares the fit of two nested models. In this case, Model 1 (main effects only) is “nested” within Model 2 (main effects plus the sex:age interaction term) because Model 2 includes all the variables from Model 1, plus an additional term. The LRT assesses whether adding the interaction term significantly improves the model’s fit to the data.

## [1] "Likelihood Ratio Test"
## Likelihood ratio test
## 
## Model 1: as.factor(wraceatt_numeric) ~ tcemdiqu2 + sex + age + ethgrp2 + 
##     rural2
## Model 2: as.factor(wraceatt_numeric) ~ tcemdiqu2 + sex * age + ethgrp2 + 
##     rural2
##   #Df LogLik Df  Chisq Pr(>Chisq)  
## 1   8 -10153                       
## 2   9 -10150  1 5.5346    0.01864 *
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

The p-value (0.01864) is less than 0.05 (the conventional significance level). This means that we reject the null hypothesis. The null hypothesis in this case is that the interaction term (sex:age) does not significantly improve the model fit.

Therefore, we conclude that the addition of the sex:age interaction term does significantly improve the fit of the model. This suggests that the relationship between age and wraceatt_numeric differs significantly between males and females. Stated another way: Age affects racial attitudes uniquely for men and women and can’t just be looked at individually.

The Curvilinear Effect of Age on the worry about being attacked

To explore the potential non-linear relationship between age and worry about being attacked because of skin colour, ethnic origin, or religion, an ordinal regression model was fitted and age was centered to improve interpretability, and a quadratic term (age2) was included to capture curvilinear effects.

The model revealed a statistically significance effect of age, with the quadratic term suggesting a U-shaped relationship.

To visualize this effect, predicted probabilities for each level of worry about attack were calculated across the age range, holding other variables at their most frequent values. This analysis suggests that the relationship between age and worry about being attacked because of skin colour, ethnic origin, or religion is more complex than a simple linear association, highlighting the importance of considering non-linear effects.

The figure below displays the predicted levels of worry across the age range.

## Call:
## polr(formula = as.factor(wraceatt) ~ tcemdiqu2 + sex + age_centered + 
##     I(age_centered^2) + ethgrp2 + rural2, data = regression_data, 
##     Hess = TRUE)
## 
## Coefficients:
##                        Value Std. Error t value
## tcemdiqu2          0.0304803  1.454e-02   2.097
## sex2              -0.2716360  3.950e-02  -6.877
## age_centered       0.0079597  1.104e-03   7.207
## I(age_centered^2)  0.0003465  5.802e-05   5.972
## ethgrp22          -1.4663274  2.529e-01  -5.798
## ethgrp23          -1.8367197  9.961e-02 -18.438
## ethgrp24          -1.2129321  1.362e-01  -8.906
## ethgrp25          -1.1344194  1.738e-01  -6.528
## rural22            0.4735785  5.022e-02   9.431
## 
## Intercepts:
##     Value    Std. Error t value 
## 1|2  -3.1682   0.0742   -42.6907
## 2|3  -2.0039   0.0634   -31.6047
## 3|4  -0.1439   0.0585    -2.4613
## 
## Residual Deviance: 20152.64 
## AIC: 20176.64

The graph shows an U-shaped relationship between age and predicted worry about being attacked. Worry increases with age until approximately 37, after which it begins to decline. The predicted worry level is lowest for young adults in their early 20s and for older adults over 75.

Summary

This study investigates the factors influencing fear of race/ethnic/religious-based attacks in Great Britain, utilizing the Crime Survey for England and Wales (2017-2018). A permutation test revealed that individuals in the most deprived areas exhibit significantly different levels of worry compared to those in the least deprived areas, indicating an overall association between deprivation and fear. While socioeconomic deprivation was not found to be a significant predictor in either ordinal logistic regression model, both gender, age, ethnicity, and rural/urban residence were significantly associated with reported worry. In particular, results suggested that females and residents of urban areas reported higher levels of worry, while further analysis is needed to interpret the effect of ethnicity. Critically, Model 2 revealed a significant interaction between gender and age, suggesting a stronger positive relationship between age and worry for women compared to men. Further analysis with a quadratic term for age also displayed a U-shaped relationship between fear and age, with fear increasing until age 37 and decreasing past that.

Conclusion

This study underscores the complex interplay of factors shaping fear of race, ethnic, or religious-based attacks in Great Britain. While socioeconomic deprivation did not emerge as a direct predictor in our regression analyses, the permutation test highlights its broader association with heightened safety concerns, aligning with research suggesting that area-level disadvantage fosters a climate of anxiety and vulnerability (Sampson & Raudenbush, 1999). The significant demographic effects – particularly the heightened worry among women and urban residents – point to the uneven distribution of this burden of fear. Critically, the interaction between gender and age reveals the unique experiences of women as they age, as they report a stronger positive relationship between age and fear of attack compared to men. Furthermore, The curvilinear effect of age on the relationship with fear emphasises the diversity of experiences faced by all individuals. These patterns highlight the importance of considering intersectional identities in shaping policy and align with previous findings suggesting the influence of area-level disorder on fear of violence (Skogan, 1990). Future research should explore these underlying mechanisms, evaluate targeted interventions to address the specific safety concerns of women and urban populations and investigate how community-based initiatives can foster resilience and challenge the narratives of fear that disproportionately affect marginalised groups (Perry, 2001).

References

  • Bowling, B., & Phillips, C. (2002). Racism, crime and justice. Longman. (A foundational text on race and crime in the UK)

  • Flatley, J., Farrall, S., Graef, R., Phillips, J., & Hough, M. (2010). Measuring public confidence in the criminal justice system: Findings from the 2008/09 Crime Survey for England and Wales. Ministry of Justice Statistics Bulletin. (Or a more recent report if available – look for reports that describe the methodology and purpose of the CSEW).

  • Office for National Statistics (ONS). (2023). Crime in England and Wales: Year ending March 2023. [Link to ONS website]. (Replace with the actual ONS page describing the CSEW and its purpose).

  • Phillips, C., & Bowling, B. (2002). Racist victimisation: Experiences of young people. The British Journal of Criminology, 42(3), 539-559. (Important research on the experiences of young people).

  • Perry, B. (2001). In the name of hate: Understanding hate crimes. Routledge.

  • Sampson, R. J., & Raudenbush, S. W. (1999). Systematic social observation of public spaces: A new look at disorder in urban neighborhoods. American Journal of Sociology, 105(3), 603-651.

  • Skogan, W. G. (1990). Disorder and decline: Crime and the spiral of decay in American neighborhoods. Free Press.

  • Spalek, B. (2010). Religion, spirituality, and social work. Policy Press. (Addresses the intersection of religion, social justice, and related issues).

  • Webster, C., Simpson, D., & Quirk, T. (2014). Social cohesion, fear of crime and area deprivation: testing the ‘social glue’ hypothesis in disadvantaged communities. Journal of Urban Affairs, 36(1), 1-18. (Directly addresses the relationship between deprivation and fear of crime.)