Laserfiche WebLink
or scaling ratio to generate national estimates from NFIRS data But NFIRS is a sample from a <br /> universe whose size cannot be inferred from NFIRS alone. Also,participation rates in NFIRS are <br /> not necessarily uniform across regions and sizes of community, both of which are factors <br /> correlated with frequency and severity of fires. This means NFIRS may be susceptible to <br /> systematic biases. No one at present can quantify the size of these deviations from the ideal, <br /> representative sample,so no one can say with confidence that they are or are not serious <br /> problems. But there is enough reason for concern so that a second data base-the NFPA survey- <br /> is needed to project NFIRS to national estimates and to project different parts of NFIRS <br /> separately. This multiple calibration approach makes use of the annual NFPA survey where its <br /> statistical design advantages are strongest <br /> There are separate projection formulas for four major property classes(residential <br /> structures,non-residential structures,vehicles, and other)and for each measure of fire severity <br /> (fire incidents,civilian deaths,and civilian injuries, and direct property damage). <br /> For example,the scaling ratio for 1998 civilian deaths in residential structures is equal to <br /> the total number of 1998 civilian deaths in residential structure fires reported to fire departments, <br /> according to the NFPA survey (3,250), divided by the total number of 1998 civilian deaths in <br /> residential structure fires reported to NFIRS(1,224). Therefore,the scaling ratio is 3,250/1,224— <br /> 2.66. <br /> The scaling ratios for civilian deaths and injuries and direct property damage are often <br /> significantly different from those for fire incidents. Except for fire service injuries, average <br /> severity per fire is generally higher for NFIRS than for the NFPA survey. Use of different scaling <br /> ratios for each measure of severity is equivalent to assuming that these differences are due either <br /> to NFIRS under-reporting of small fires,resulting in a higher-than-actual loss-per-fire ratio,or <br /> possible biases in the NFIRS sample representation by region or size of community,resulting in <br /> severity-per-fire ratios characteristic only of the oversampled regions or community sizes. <br /> Note that this approach also means that the NFPA survey results for detailed property-use <br /> classes(e.g.,fires in storage structures)may not match the national estimates of the same value. <br /> Calculating National Estimates of Particular Types of Fires <br /> Most analyses of interest involve the calculation of the estimated number of fires not only <br /> within a particular occupancy but also of a particular type. The types that are mostly frequently <br /> of interest are those defined by some ignition-cause characteristic. The six cause-related <br /> characteristics most commonly used to describe fires are: form of the heat that caused the <br /> ignition, equipment involved in ignition, form or type of material first ignited,the ignition factor <br /> that brought heat source and ignited material together,and area of origin. Other characteristics of <br /> interest are victim characteristics, such as ages of persons killed or injured in fire. <br /> For any characteristic of interest in NFIRS,some reported fires have that characteristic <br /> unknown or not reported. If the unknowns are not taken into account,then the propensity to <br /> report or not report a characteristic may influence the results far more than the actual patterns on <br /> that characteristic. For example,suppose the number of fires remained the same for several <br /> consecutive years, but the percentage of fires with cause unreported steoodily declined over those <br /> years. If the unknown-cause fires were ignored, it would appear as if fires due to every specific <br /> cause increased over time while total fires remained unchanged. This, of course,does not make <br /> sense. <br />