Identifying Residential FIre Involving Upholstered Furniture within the National Fire Incident Reporting System
Douglas S. Thomas, David T. Butry
This Technical Note develops a method for identifying residential fire incidents, reported to the National Fire Incident Reporting System (NFIRS), with missing information regarding the item first ignited (IFI) or item most contributing to flame spread (IMCFS) that were likely to have involved upholstered furniture. Using NFIRS data, upholstered furniture involvement can only be identified using the IFI or IMCFS fields; however, it is not uncommon for these fields to contain missing values or to be coded as unknown (undetermined). Current fire statistic methods address this issue by allocating the missing and unknown incidents over the distribution of those known. The appropriateness of this depends on the representativeness of the known incidents. This Technical Note explores the ability of a data imputation method to determine the likelihood of upholstered furniture involvement based on statistical correlations with other, non-missing, factors describing the fire incident. Out-of-sample validation is used to assess the ability of the statistical models to accurately impute the missing data. A number of competing statistical models are evaluated, with the best performing models used to estimate the number of U.S. residential upholstered furniture fires, related occupant fatalities and injuries, and property and content losses, by year, from 2006 to 2010.