The effects of quality control on decreasing error propagation in the LandScan USA population distribution model: A case study of Philadelphia County

LandScan USA is a 90 m population distribution model that is used for a variety of applications, including emergency management. Models should have a measure of accuracy; however, the accuracy of population distribution models is difficult to determine due to the inclusion of multiple input datasets and the lack of quantifiable, observable (validated) data to confirm model output. Validated data enables quantification of: (1) overall model accuracy and (2) changes in model output at different levels of quality control. This article examines the effect of quality control for two national school datasets incorporated as input in LandScan USA for Philadelphia County, Pennsylvania; which had a local, validated school dataset available. The effect of each stage of quality control efforts utilized throughout the LandScan USA process were assessed to determine what level of quality control was required to have a statistically significant change of the model's population distribution. The typical level of quality control for LandScan USA resulted in 36% of schools being moved to the correct location and 20% of missing student enrollments were found, compared to 87% and 98% respectively for the validated dataset. The costs of increasing quality control resulted in a six-fold increase in labor time; however, the additional quality control did not produce statistically significant improvements in the LandScan USA model. Thus, typical quality control efforts for schools in LandScan USA produced a population distribution similar to the validated level of quality control, and can be applied with confidence for policy, planning, and emergency situations.