Model-based small area estimation techniques in criminology: Theory and applications

Research Institution / Organisation

University of Manchester

Level of Research


Project Start Date

September 2016

Research Context

‚ÄčLarge scale spatially aggregated data hide internal heterogeneity in levels of both crime and perceptions of crime and antisocial behaviour. Thus, low area level approaches are necessary to un-erroneously associate phenomena to social and environmental features.

Victimisation surveys are one of the most important sources of information to analyse emotions and attitudes towards crime and non-reported victimisation at low area level. However, surveys need to record big samples of citizens per small area in order to allow reliable and precise direct estimates. Unfortunately, most available surveys are designed to be representative of large areas. In England, the Crime Survey for England and Wales suffers from small sample sizes at neighbourhood level. Even local surveys, which are used in criminology to measure crime patterns at lower geographical levels, are frequently based on samples that are not large enough to allow precise direct estimates at neighbourhood level.

In this respect, new methodological approaches are used to map emotions and attitudes towards crime at low spatial level, such as mobile phone applications and websites where users report and map their real-time emotions, or collection of big data from administrative records. However, such approaches might be limited by biased social participation, underrepresenting some neighbourhoods and overrepresenting others.
In order to map variables recorded from victimisation surveys at small area level without need to record new data, indirect model-based small area estimation approaches are helpful to produce estimates of adequate precision. Small area estimation techniques make use of already existing survey data and introduce models to borrow strength from related and neighbouring areas. Model-based small area estimation techniques have shown to be a potential tool for low area level mapping in criminology.

Research Methodology

This project will explore different small area estimation techniques and apply them to variables of primary interest for criminologists (e.g. worry about crime, perceived disorder, reporting crime). The research will make use of already existing data recorded from different social and victimisation surveys, such as the European Social Survey, the Crime Survey for England and Wales and the Manchester Resident Telephone Survey, and produce estimates from direct and indirect small area estimation techniques to provide highly precise estimates at small area level.

More specifically, this project aims to explore:

  1. direct estimation techniques, which use data recorded by the original survey for each area and survey weights to produce design-unbiased estimates, which might be unreliable in areas with small sample sizes.
  2. empirical Best Linear Unbiased Predictor (EBLUP) combines direct estimates and synthetic estimates produced from models that link area-level covariates (e.g. census, administrative records), in order to borrow strength from related areas and produce more precise estimates.
  3. spatial and spatial-temporal EBLUP estimators, which add spatially correlated random area effects and temporal random effects, to increase the precision of our estimates.

Interim reports and publications

Buil-Gil, D., Medina, J., & Shlomo, N. (2019). The geographies of perceived neighbourhood disorder. A small area estimation approach. Applied Geography

Buil-Gil, D., Moretti, A., Shlomo, N., & Medina, J. (2019). Worry about crime in Europe: A model-based small area estimation from the European Social Survey. European Journal of Criminology

Buil-Gil, D., Moretti, A., Shlomo, N., & Medina, J. (forthcoming). Applying the Spatial EBLUP to place-based policing. Simulation study and application to confidence in police work. Applied Spatial Analysis and Policy.

Buil-Gil, D., Solymosi, R., & Moretti, A. (forthcoming). Non-parametric bootstrap and small area estimation to mitigate bias in crowdsourced data. Simulation study and application to perceived safety. In C. Hill, P. Biemer, T. Buskirk, L. Japec, A. Kirchner, S. Kolenikov & L. Lyberg (Eds.), Big data meets survey science. Wiley.

Date due for completion

February 2020
Return to Research Map