Improving Measurement of Community Safety Perceptions with Enhanced Data Inclusivity and Novel Use of Small Area Estimation through Respondent Driven Sampling: A Pilot Study in Detroit, Michigan

Community perception of safety is a fundamental driver of public policies intended to improve safety. Perceptions are not straightforward to measure, often reflecting a complex set of personal, social, environmental, and other components as well as their interactions. These interactions make spatial-contextual thinking an important framework in criminology. While increasing data availability offers an opportunity to examine this human-place interaction at various geographic levels, heterogeneity across microgeographies is reported to be often lost in meso-level analyses. Social surveys are an important methodological component that provide an insight into spatial-contextual processes underlying safety perception. Yet, current practice of surveys is limited in providing such insights into safety perception within microgeographies because of constraints related to sample design, costs, and nonresponse.

To fill this gap, we propose a novel sample design and its inference method implemented in a pilot survey in Detroit, Michigan with a sample size of 1,200. The study attempts to improve measurement of community safety perceptions with enhanced data inclusivity and through practical hybrid sampling that blends address-based sampling and respondent driven sampling on the principles of probability sampling with scientifically supported representativeness. Our data collection method is cost effective, as it will rely on push-to-web surveys.

The sample will be designed to provide reliable estimates at the Census block group level through sophisticated model-based small area estimation techniques that incorporate crime data, contextual data as well as emerging spatial data (e.g., street view images coded for safety relevant features) into our survey data. These estimates will show heterogeneity or homogeneity in block group characteristics related to safety perception and other related measures. The proposed sampling and small area estimation methods are scalable, such that they could be applied to municipalities of varying sizes, and adaptable for various conditions of municipality data collection systems.

Importantly, our project underscores community engagement throughout its planning, execution, dissemination, and closure stages. For example, proposed protocols offer flexibility to introduce the community input to setting analysis outcome variables so that the research products from our study maximize their utility and relevance. Moreover, through the planned community engagement activities, we plan to outline intervention strategies and policy recommendations. Combined with the aggregate estimates at the block group level, these approaches are expected to provide stakeholders with actionable insights which, in turn, help them develop steps specific and tailored to their neighborhood.