Reka Solymosi and Carly Lightowlers
The connection between crime, including violence, and deprivation is firmly established, and many studies seek to deploy measures of deprivation in analyses of crime. One approach to conceptualising and operationalising deprivation commonly used in England is the English indices of deprivation 2019, specifically the Index of Multiple Deprivation (IMD). The IMD operationalises deprivation and provides a useful tool to understand the link between crime and deprivation. However, the IMD is not without problems, as crime is included as one of the indicators of deprivation. Including crime rates as part of the deprivation measure can artificially inflate the correlation with the (crime) outcome and thus overestimate crime levels. In this blog we demonstrate an alternative approach by re-calculating a composite deprivation score while excluding the crime indicator.
Deprivation as a predictor of crime
The link between crime and deprivation may be attributed to either offender motivations or strained social relations within deprived communities (i.e., social disorganisation (Shaw and McKay 2010; Lightowlers, Pina-Sánchez, and McLaughlin 2021).
Settings that are more deprived are believed to provide a conducive environment for violent behaviour, as they promote polarisation and erode the sense of community and trust, ultimately resulting in increased violence (Lightowlers, Pina-Sánchez, and McLaughlin, 2021; Wilkinson, 2004).
As such many studies seek to deploy measures of deprivation in analyses of crime. In England, the English indices of deprivation 2019, specifically the Index of Multiple Deprivation (IMD) is often used to conceptualise and operationalise deprivation. The IMD provide a set of relative measures of deprivation for small geographical areas (Lower-layer Super Output Areas (LSOAs)), based on seven domains (Noble et al. 2019):
- Income (weight: 22.5%)
- Employment (weight: 22.5%)
- Education, Skills and Training (weight: 13.5%)
- Health and Disability (weight: 13.5%)
- Crime (weight: 9.3%)
- Barriers to Housing and Services (weight: 9.3%)
- Living Environment (weight: 9.3%).
Every LSOA in England is assigned a deprivation score based on the above indicators. Higher scores indicate greater deprivation.
Challenges with using the IMD as a predictor of crime
A drawback of using IMD scores to predict crime is that area level measures of crime comprise one of the indicators used to derive the composite deprivation score. This results in several concerns:
Circularity: The inclusion of the crime indicator in the IMD can result in circularity, whereby the predictor variable (deprivation) is also affected by the outcome variable (crime). This can make it difficult to establish causality in the relationship between deprivation and crime.
Overestimation: As a result, using the IMD with the crime indicator may overestimate the relationship between deprivation and crime, as the inclusion of the crime indicator will inflate the deprivation score for areas with high crime rates.
Overfitting: However, if deploying all the non-crime domains as separate indicators we may run into collinearity between these (especially if the sample size is small or model overfitted). Moreover, retaining a single deprivation variable is a more parsimonious modelling strategy, allowing one to intuitively assess deprivation’s overall effect on crime. This reduces the chances of overfitting models, which is crucial in area level analysis (as these are often on relatively small sample sizes).
One approach taken to get around this issue is to select only one indicator (e.g. the income domain), to predict crime. For example, as deployed by Lymperopoulou and Bannister (2022) in their study of the spatial concentration of poverty and crime.
However, if we are concerned with deprivation more generally (as opposed to poverty), adopting this approach raises further conceptual and methodological concerns. Specifically, the following:
Oversimplification: Deprivation is a complex phenomenon that cannot be fully explained by a single factor such as income or employment. By relying solely on income as an indicator of deprivation, we oversimplify this multidimensional concept.
Model misspecification: The relationship between deprivation and crime is not the same in all areas. Using only one indicator, such as income, may not capture the complexity of this relationship across different geographical areas or populations. If income is not a good predictor of crime, then using only this indicator may result in further model misspecification, as this can lead to biased estimates and incorrect inferences about the relationship between deprivation and crime. Endogeneity is a common problem in econometrics, and occurs when a predictor variable is correlated with the error term in a regression model. This means that the predictor variable is not independent of the unobserved factors that affect the outcome variable, which can lead to biased and inconsistent estimates.
Reverse Causality: While it is often assumed that deprivation leads to crime, it is also possible that crime can contribute to deprivation by reducing economic opportunities or decreasing property values.
Recalculating the IMD without the crime indicator
We propose another approach, adopted by Lightowlers, Pina-Sánchez, and McLaughlin (2021), to re-calculate an index of multiple deprivation without the crime indicator.
The IMD score comprises the weighted sum of seven indicators, exponentially transformed to allow combining them (see Ministry of Housing, Communities and Local Government, 2019). To calculate IMD without the crime indicator, we accessed these scores, and re-combined them, excluding the crime indicator. To do this, we simply re-weighted the other six domains using the same ratio to ensure that the total weight is still 1.00 (i.e. divide them all by 1 – 0.093, the weight of the crime indicator). We then create a composite IMD score from the new weighted sum of these scores, resulting in a general deprivation score that contains all indicators except the crime indicator. We are comfortable in adopting this approach because the original weights were also somewhat arbitrary and subject to judgement (see section 3.7 and Appendix G in Ministry of Housing, Communities and Local Government, 2019a), and in any case the revised weights are not that dissimilar to the originals.
Steps taken to recalculate the IMD in this way can be found here https://github.com/maczokni/recalculating_imd.
The impact of excluding/including the crime indicator
To test the impact of this approach we furnish an illustrative example and hypothetical research question: Is there an association between area-level deprivation and the number of violent crimes in Cleveland? Data about the outcome – in this case the number of violent crimes at LSOA level, was available from the data.police.uk website.
We obtained one year’s worth of violent crime data from March 2022 to March 2023 for Cleveland and pursued an analysis of the spatial correlation of violence with deprivation. To do so we deploy the original IMD score LSOAs in Cleveland, using a simple spatial error model.
We then compared this to the same analysis using the recalculated IMD (minus the crime domain). Details of the analytical steps taken can be found here https://github.com/maczokni/recalculating_imd.
We found the model deploying the original IMD score suggested there were, on average, 2.89 more crimes per LSOA that year for each increase in the (continuous) deprivation score (model 2). This compared to 2.79 more violent crimes for each increase in the deprivation score of an LSOA when deploying the recalculated IMD (without the crime domain) (model 1). This finding corroborates our suggestion that using the original IMD inflates its effect.
In this blog we have demonstrated an alternative approach to the IMD by re-calculating a composite deprivation score while excluding the crime indicator.
Using the example of violent crime rates in Cleveland, we found a positive association between multiple deprivation score and violent crime. Our results also demonstrated that including crime rates as part of the deprivation measure will artificially inflate the correlation with the (crime) outcome and thus overestimate crime levels. However, as the change in effect size was not dramatic, this did not alter the conclusions drawn from these data.
Nevertheless, this solution solves the conceptual and statistical issues referred to earlier and is a useful approach for those wishing to study the association between crime and deprivation more precisely in their work.
To access the code for the above recalculation and modelling outlined here please visit https://github.com/maczokni/recalculating_imd
A special thanks to Stephen Clark (University of Leeds) who introduced and advised Lightowlers on this as a possible approach to pursuing the analysis in Lightowlers et al. (2021) and Dr Jose Pina-Sánchez (University of Leeds) for comments on an earlier draft of this blog.
Lymperopoulou, K. and Bannister, J (2022). The spatial reordering of poverty and crime: A study of Glasgow and Birmingham (United Kingdom), 2001/2 to 2015/16. Cities 130 (103874). https://doi.org/10.1016/j.cities.2022.103874.
Lightowlers, Carly, Jose Pina-Sánchez, and Fiona McLaughlin. 2021. “The Role of Deprivation and Alcohol Availability in Shaping Trends in Violent Crime.” European Journal of Criminology, 14773708211036081.
Ministry of Housing, Communities and Local Government (2019) English indices of deprivation 2019: technical report. Available: https://www.gov.uk/government/publications/english-indices-of-deprivation-2019-technical-report Accessed: 09/03/2022.
Ministry of Housing, Communities and Local Government (2019a) ‘File 9: transformed domain scores’. Available: https://www.gov.uk/government/statistics/english-indices-of-deprivation-2019 Accessed 28/03/2023
Noble, Stefan, David McLennan, Michael Noble, Emma Plunkett, Nils Gutacker, Mary Silk, and Gemma Wright. 2019. “The English Indices of Deprivation 2019.” London: Ministry of Housing, Communities and Local Government.
Shaw, Clifford Robe, and Henry Donald McKay. 2010. “Juvenile Delinquency and Urban Areas: A Study of Rates of Delinquency in Relation to Differential Characteristics of Local Communities in American Cities (1969).” In Classics in Environmental Criminology, 103–40. Routledge.
Wilkinson, Richard. 2004. “Why Is Violence More Common Where Inequality Is Greater?” Annals of the New York Academy of Sciences 1036 (1): 1–12.
About the authors
Dr Reka Solymosi is a Senior Lecturer in Quantitative Methods at the Department of Criminology at the University of Manchester. Her research focuses on using new forms of data to gain insight into people’s behaviour and subjective experiences, particularly focusing on crime, victimisation, transport, and spatial research.
Dr Carly Lightowlers is a Senior Lecturer in Criminology at the University of Liverpool, with research experience as an academic and in local and central government. Alcohol consumption, associated offending and sentencing are key research interests, although she has researched violence more broadly (e.g. 2011 English riots and modern slavery).
Contact
Dr Reka Solymosi
Department of Criminology,
The University of Manchester
reka.solymosi@manchester.ac.uk
@r_solymosi
https://rekadata.net/
Dr Carly Lightowlers
Department of Sociology,
Social Policy and Criminology
School of Law and Social Justice,
University of Liverpool
c.lightowlers@liverpool.ac.uk
@Carly_LL
This article gives the views of the authors, not the position of the British Society of Criminology or the institution they work for.
The BSC Blog 
You must be logged in to post a comment.