Police-recorded crimes are used by police forces to document community differences in crime and design spatially targeted strategies. Nevertheless, crimes known to police are affected by selection biases driven by underreporting. This paper presents a simulation study to analyze if crime statistics aggregated at small spatial scales are affected by larger bias than maps produced for larger geographies. Based on parameters obtained from the UK Census, we simulate a synthetic population consistent with the characteristics of Manchester. Then, based on parameters derived from the Crime Survey for England and Wales, we simulate crimes suffered by individuals, and their likelihood to be known to police. This allows comparing the difference between all crimes and police-recorded incidents at different scales. Measures of dispersion of the relative difference between all crimes and police-recorded crimes are larger when incidents are aggregated to small geographies. The percentage of crimes unknown to police varies widely across small areas, underestimating crime in certain places while overestimating it in others. Micro-level crime analysis is affected by a larger risk of bias than crimes aggregated at larger scales. These results raise awareness about an important shortcoming of micro-level mapping, and further efforts are needed to improve crime estimates.