Trends, patterns and projection of crime in metropolitan Bangladesh, 2010-2024

Data underlying a study of recorded crime in the eight metropolitan police areas of Bangladesh. These are a subset of the administrative urban; the roughly 276 smaller pourashavas are outside the analysis and this deposit. The deposit holds the analysis-ready dataset at the three scales the evidence exists at, together with the posterior summaries of every model reported in the accompanying article, including the trend model projected to 2030 and the backtest that says what that projection is worth. The metropolitan tier is used because it is the only part of urban Bangladesh that is unambiguously a city: the administrative urban runs from Dhaka to towns of a few thousand people, so a single analysis of it pools settlements two orders of magnitude apart. The eight metropolitan areas each have a police force constituted for them, and together they are the only units in the country with a consistent offence breakdown across the whole window. Three scales are deposited and are not interchangeable: 1. **The city** (8 units). 1,278 city-year-offence observations covering 14 gated years and 17 offence categories, with a structural profile of each city: census demography 1991-2022, sex ratio, slum households, transient accommodation, schooling by sex, medical provision and police strength. 2. **The metropolitan thana** (82 units). Police establishment and arrests broken down by sex, for the six forces whose thana sheets exist. This is the only unit inside a city for which anything is published about the people arrested. 3. **The areal lattice** (507 units). The national adm3 geography with queen contiguity, arrests and population, in which the city corporations sit as ordinary members with real neighbours. This is the only geography in the study with genuine adjacency, and it is what makes a spatial model estimable.