The Crime You Can't Chart: Visualizing the Dark Figure

Official crime statistics whether police records, dashboards, or annual reports are often treated as definitive representations of public safety. But a large portion of crime never becomes data at all. Criminologists call this the dark figure of crime: the gap between what happens and what gets documented. In contexts like India, where stigma, institutional barriers, and recording friction influence whether an incident ever becomes a record, the invisible portion can be substantial.

This talk explores why visualizing crime or any public phenomenon is not just about charts and dashboards. It requires careful data engineering, robust statistical reasoning, and ethical interpretation before a visualization can meaningfully reflect reality. Through the lens of crime data, I will demonstrate how gaps in data generation and collection shape what becomes visible, and why ignoring these gaps risks misleading analysis.

The talk will be structured into four parts. First, we will unpack the concept of the dark figure of crime and show how official data often reflects the system that produced it, not the full reality. Second, we will examine where data engineering matters: classification systems, metadata, documentation, and pipelines determine what gets recorded, lost, or misrepresented long before visualization begins. Third, we will discuss statistical approaches for estimating unobserved phenomena, including inference under systematic underreporting, scenario-based projections, and how to communicate uncertainty without false precision. Finally, we will cover visualization strategies that explicitly represent missingness, ranges, and confidence bounds, helping audiences interpret data responsibly.

While this talk uses crime as a case study, the same framework applies to many Indian datasets where missingness is structural from public health and sanitation to education, labour, road safety, and social indicators. In such contexts, engineering and statistics are not “support roles” to visualization; they are essential for building projections with transparent assumptions and meaningful uncertainty.

The intended audience includes data visualization practitioners, analysts, data engineers, civic technologists, and journalists working with public data. Key takeaways include: recognizing structural incompleteness in datasets, designing assumptions and projections responsibly, and creating visualizations that communicate both what is known and what is missing.