Contents
The NFPA fire experience survey is the annual sample survey of United States fire departments run by the National Fire Protection Association (NFPA). From it NFPA estimates the national numbers of fires, civilian deaths and injuries, and direct property damage reported to fire departments.[1][2] Ahrens and Stewart (2008) call it the source of the "big picture" on the U.S. fire problem.[3] Each January it is sent to all departments protecting 100,000 people or more and to a random sample, stratified by population, of smaller departments.[1] Its results appear each year as Fire Loss in the United States.[4]
The survey asks for summary counts only. Its second main role is to calibrate the more detailed but incomplete National Fire Incident Reporting System (NFIRS). Ratios between survey and NFIRS totals scale NFIRS counts of specific fire problems up to "national estimates".[5] Tables in the 2008 NFPA handbook that combine the two sources carry notes such as "Source: NFIRS and NFPA survey".[6] This article describes the survey and its results as the 2008 NFPA handbook and the 2002 SFPE handbook present them, mostly for the years 1977 to 2004.
Development and purpose
In the early 1970s, before NFIRS existed, NFPA prepared its estimates of U.S. fire incidents, deaths, injuries and losses from its annual survey. It added analyses of causes and other details from its anecdotal Fire Incident Data Organization (FIDO).[7] In 1977 NFPA modified its survey techniques to follow modern statistical design principles, which greatly improved the estimates of the size of the national fire problem.[7] NFPA's published series of civilian fire death estimates starts in 1977, the year the survey was upgraded.[8]
Many firefighters have wondered why they must complete the NFPA survey when they already file NFIRS reports. The reason given by Ahrens and Stewart (2008) is that states have different reporting practices, so NFIRS cannot estimate the total number of reported fires per year.[9] Analysts need the survey results and NFIRS together to develop national estimates. The procedures were set out by Hall and Harwood in a 1989 article, as cited by Ahrens et al. (2008).[4] NFPA and the U.S. Fire Administration have also supported each other's databases over the years.[10]
The survey is timely because it is small. It is completed by summer each year and can produce numbers about a year earlier than NFIRS.[9] NFPA issues a press release on the number of fires and fire deaths each summer. An abbreviated version of the full report Fire Loss in the United States appears in a fall issue of NFPA Journal, and the report is also available on NFPA's website.[1][2]
Sample design and content
The survey is a stratified random sample. In such a sample, a disproportionate share of the sample is placed in the subpopulations that account for a disproportionate share of the total variance. This gives a lower total variance for a given sample size than a simple random sample.[11] Hall and Ahrens (2002) state that the survey captures roughly one-tenth of all U.S. fire departments, including most departments protecting a population of at least 100,000. The stratification preserves randomness, and hence representativeness, while reducing sampling uncertainty.[12]
Ahrens and Stewart (2008) give the numbers for 2004. NFPA sent out almost 13,000 surveys and received 2,830 back, a response rate of 22 percent, with higher response rates from larger communities.[1] The departments that responded protected about 36 percent of the U.S. population.[1] The response rate of 22 percent follows directly from the two counts: 2,830 divided by 13,000 is 0.218.
The questionnaire is short. It asks about fires by type and broad occupancy class, intentional structure and vehicle fires, and civilian and firefighter deaths and injuries.[1] The survey also collects fire department calls, and projections of total fires, casualties, direct property damage and calls are made from it.[2] Because older versions of NFIRS did not forward nonfire responses, the survey was described in 2008 as the only national source of such data until all departments reported in NFIRS 5.0.[9] It collects summary data and lacks most of the detail collected by NFIRS.[2]
The survey covers municipal fire departments only, both career and volunteer. Fires that are never reported to a fire department, fires handled by private brigades or individuals, and fires reported to wildland or military agencies are not captured.[9] National estimates therefore exclude fires reported only to federal or state agencies or to industrial fire brigades.[6] A department can also only report incidents it has documented, so minor fires and nonfire incidents may be missing where a state does not require them to be recorded.[13]
National estimates and the scaling of NFIRS
By comparing the survey totals with the corresponding NFIRS totals, analysts estimate what fraction of U.S. fires, deaths, injuries and property damage NFIRS captured. These fractions are inverted into scaling ratios that convert NFIRS counts of specific problems into national estimates.[5] Hall and Ahrens (2002) describe separate projection formulas for four property classes and four measures of severity. The classes are residential structures, nonresidential structures, vehicles and other fires; the measures are fires, civilian deaths, civilian injuries and direct property damage.[12]
Because many NFIRS reports leave some fields blank, the ratio is usually formed with only the NFIRS fires for which the characteristic of interest is known. This allocates the unknowns and assumes that fires with unreported characteristics would show the same proportions as fires with known ones.[14] The two steps can be written as follows:
| Symbol | Description | Unit |
|---|---|---|
| Scaling ratio for property class k and severity measure m | – | |
| National total estimated from the NFPA survey | fires, deaths, injuries or dollars | |
| NFIRS total for the same class and measure, counting only reports with the characteristic known | same as above | |
| NFIRS count for one value of the characteristic, such as one cause | same as above | |
| National estimate for that value | same as above |
The right-hand form shows that multiplying by the scaling ratio is the same as applying the NFIRS share of a category to the survey total. For fields that allow several entries, NFPA forms the allocation ratio as the total divided by the total minus the entries coded "none", "undetermined" or "not reported".[15] Without allocation, a fall in the share of unknown-cause reports would look like a rise in every specific cause while total fires stayed constant.[14]
Hall and Ahrens (2002) give a worked example for 1996. The survey estimated 4,080 civilian deaths in residential structure fires, and NFIRS contained 1,504, so the scaling ratio was 4,080 / 1,504 = 2.71.[12] In a second example, the survey gave about 480 civilian deaths a year in highway vehicle fires for 1999–2004. NFIRS proportions assigned about 88 percent of them, or 425 a year, to passenger vehicles.[16] As a check, 480 × 0.88 gives 422; the small difference from 425 is consistent with the rounding of both published inputs.
Separate ratios for each measure matter because NFIRS and the survey differ in severity. Except for fire service injuries, the average severity per fire, such as deaths per 100 fires, is generally higher in NFIRS than in the survey.[12] Using a separate ratio for each measure is equivalent to assuming that the difference arises because NFIRS underreports small fires, or because some regions or community sizes are overrepresented in it.[14]
Results
The survey series shows a large fall in the U.S. fire problem. In 1977 it estimated 3,264,000 fires, 7,395 civilian deaths and 31,190 civilian injuries; in 2004 the estimates were 1,550,500 fires, 3,900 deaths and 17,785 injuries.[17] These are falls of 52 percent in fires and 47 percent in deaths. Deaths were stuck at around 6,000 a year from 1982 to 1988, and injuries showed no consistent trend until the mid-1990s.[17] Direct property damage rose from $4.7 billion in 1977 to $9.8 billion in 2004, but most of that rise is due to inflation.[17] The 2001 total of 6,196 deaths includes 2,451 deaths from the events of 11 September 2001.[18]
For 2000–2004, excluding 11 September 2001, the survey attributed about half of all fires to outdoor and other fires, 21.5 percent to homes and 20.1 percent to vehicles.[19] Homes accounted for 80.3 percent of civilian fire deaths, three out of four injuries and over half of direct property damage. Vehicle fires caused nearly three times as many deaths as fires in structures other than homes.[19]
The survey also tracks the workload of fire departments. Calls rose from 10,819,000 in 1980 to 22,616,500 in 2004.[20] In 2004, medical aid made up 62 percent of calls, false alarms 9 percent and fires 7 percent, so fires were only 1 of every 14 calls.[20]
Survey totals are the numerators of the national fire death rate. In 2004 the estimated 3,900 civilian fire deaths, divided by a resident population of 293,655,000, gave 13 deaths per million population.[21] Combined with NFIRS, the survey also yields national estimates by fire cause. In 1999–2002 smoking materials started 7 percent of home structure fires but caused 29 percent of the deaths, 3.4 deaths per 100 fires. Cooking equipment started 28 percent of the fires and caused 9 percent of the deaths, 0.3 per 100 fires.[6]
National estimate tables follow fixed rounding conventions. Fires are rounded to the nearest hundred, civilian deaths and injuries to the nearest ten, and direct property damage to the nearest million dollars; damage is not adjusted for inflation.[6] The notes also warn that casualty and loss projections can be heavily influenced by the inclusion or exclusion of one unusually serious fire.[6]
Accuracy and uncertainty
As a random sample, the survey has negligible sample bias by design, and its sample-size uncertainty is modest. Hall and Ahrens (2002) put it at about 10 percent for deaths and much less for the other measures of fire loss.[14] Hall and Twomey (2008) state that the estimate of total fire deaths is subject to a 95 percent confidence interval of plus or minus just under 400 deaths.[18] As an illustration, applied to the 2004 estimate of 3,900 deaths this margin gives a range of roughly 3,500 to 4,300. The half-width is about 10 percent of the estimate, consistent with the value given by Hall and Ahrens.[17]
Some errors are not due to sampling. The survey can miss deaths that occur outside a reported fire; clothing ignitions are the classic example, totalling fewer than 200 deaths a year on death certificates.[18] One very large fire can dominate a projection, because "with projections, an extreme value is multiplied". Many analysts exclude such incidents from calculations and discuss them separately.[15] Year-to-year fluctuations led NFPA to produce annual averages for most analyses.[22]
Property damage is the least certain measure. The survey's damage values are estimates by fire officers, sometimes informed by insurance estimates.[18] Hall and Twomey (2008) report a modest but growing gap between NFPA's direct damage estimates and those by the Insurance Services Office: for 2002 the two were $10.3 billion and $17.6 billion. The insurance estimate may include indirect losses, whereas the NFPA estimate misses unreported fires even when they lead to insurance claims.[18] The survey value is therefore direct property damage in fires reported to fire departments, only one component of the total fire loss that national cost studies consider.[23]
Quality control is easier for a small survey than for a large incident database. For a database with limited depth such as the NFPA survey, considerable effort can be put into making each report complete, with follow-up calls to fill gaps.[24] For vehicle fires, Long et al. (2008) noted that the survey follows up fatalities to separate deaths caused by fire from deaths caused by crash trauma, whereas NFIRS does not. They concluded that the direct survey estimate of highway vehicle fire deaths may be the more reliable one.[16]
Comparison with other sources
Different databases give different fire death totals because they define fire deaths differently. NFPA's estimates are sample-based projections that include deaths in all types of fires, including arson, self-immolation and vehicle fires.[25] On death certificates these deaths might be classified as homicide, suicide and vehicle accidents. NFPA uses death certificate data for comparisons between states, because of their more complete and consistent coding, and for long-term trends, where no other database is as consistent in method.[25][17] The trends and patterns from the two sources are consistent.[25]
Hall and Twomey (2008) compare the two series for 1993–2002. In 2002 the survey estimated 3,380 civilian fire deaths in total and 2,775 in structure fires, while death certificates with the primary fire codes X00–X09 recorded 3,159.[18] They regard the survey estimates as the best measures of fire loss when fire is considered by itself. For comparisons with other hazards, however, the death certificate data and insurance estimates are more comparable.[26]
Injury counts show a larger gap. The CPSC's National Electronic Injury Surveillance System estimated more than 48,000 nonarson civilian fire injuries treated in emergency rooms from July 2002 to June 2003. More than 21,000 of these, still less than half, occurred at fires attended by the fire service.[25] This is higher than NFPA's estimate of injuries reported to fire departments. In some cases, private individuals or ambulance companies outside the fire service transport patients without notifying it.[25] For passenger vehicle fires, the combined NFPA and NFIRS estimate of 425 deaths a year was about 8 percent higher than the 390 a year from the federal Fatality Analysis Reporting System.[27]
Uses
For engineering analysis in the United States, Hall and Ahrens (2002) recommend starting with national estimates that use NFIRS for detail and the NFPA survey for projection.[12] Valid analysis of incident rates requires a statistically representative database such as NFIRS and the NFPA survey, rather than more detailed databases that are unrepresentative.[28] In U.S. fire statistics, the survey thus supplies the "big picture" totals of fires, home fires, fire deaths and fire injuries, while NFIRS supplies the detail.[4][12]
Hall (2008) describes national estimates based on NFIRS and the NFPA survey as the best national data for designing public fire and life safety education.[29] Several commercial vendors of NFIRS software include the survey in their packages.[1] In the economic analysis of fire, the NFPA survey is one of the sources of the U.S. estimates of the total cost of fire prepared in the manner of the World Fire Statistics Centre.[30]
See also
Notes
- ↑ 1.0 1.1 1.2 1.3 1.4 1.5 1.6 Ahrens and Stewart (2008), p. 3-55.
- ↑ 2.0 2.1 2.2 2.3 Ahrens and Stewart (2008), p. 3-58.
- ↑ Ahrens and Stewart (2008), p. 3-43.
- ↑ 4.0 4.1 4.2 Ahrens et al. (2008), p. 3-63.
- ↑ 5.0 5.1 Ahrens and Stewart (2008), pp. 3-55–3-56.
- ↑ 6.0 6.1 6.2 6.3 6.4 Ahrens et al. (2008), p. 3-74.
- ↑ 7.0 7.1 Ahrens et al. (2008), p. 3-61.
- ↑ Ahrens et al. (2008), p. 3-83.
- ↑ 9.0 9.1 9.2 9.3 Ahrens and Stewart (2008), p. 3-56.
- ↑ Ahrens et al. (2008), p. 3-62.
- ↑ Hall (2002c), pp. 1-200–1-201.
- ↑ 12.0 12.1 12.2 12.3 12.4 12.5 Hall and Ahrens (2002), p. 5-67.
- ↑ Ahrens et al. (2008), p. 3-65.
- ↑ 14.0 14.1 14.2 14.3 Hall and Ahrens (2002), p. 5-68.
- ↑ 15.0 15.1 Ahrens et al. (2008), p. 3-77.
- ↑ 16.0 16.1 Long et al. (2008), p. 21-5.
- ↑ 17.0 17.1 17.2 17.3 17.4 Hall and Cote (2008), p. 3-4.
- ↑ 18.0 18.1 18.2 18.3 18.4 18.5 Hall and Twomey (2008), p. 1-7.
- ↑ 19.0 19.1 Hall and Cote (2008), p. 3-21.
- ↑ 20.0 20.1 Hall and Cote (2008), p. 3-28.
- ↑ Ahrens et al. (2008), p. 3-80.
- ↑ Ahrens et al. (2008), p. 3-79.
- ↑ Ramachandran (2002b), p. 5-79.
- ↑ Ahrens et al. (2008), p. 3-91.
- ↑ 25.0 25.1 25.2 25.3 25.4 Ahrens et al. (2008), p. 3-90.
- ↑ Hall and Twomey (2008), p. 1-8.
- ↑ Long et al. (2008), pp. 21-5–21-6.
- ↑ Hall and Ahrens (2002), p. 5-69.
- ↑ Hall (2008), p. 5-117.
- ↑ Ramachandran (2002b), pp. 5-79–5-80.
References
- Ahrens, M., Frazier, P., & Kelch, G. (2008). Use of fire incident data and statistics. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 3-61–3-92). National Fire Protection Association.
- Ahrens, M., & Stewart, S. (2008). Fire data collection and databases. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 3-43–3-59). National Fire Protection Association.
- Hall, J. R., Jr. (2002c). Statistics. In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 1-193–1-202). National Fire Protection Association; Society of Fire Protection Engineers.
- Hall, J. R., Jr. (2008). Using data for public education planning and decision making. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 5-103–5-118). National Fire Protection Association.
- Hall, J. R., Jr., & Ahrens, M. J. (2002). Data for engineering analysis. In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 5-65–5-78). National Fire Protection Association; Society of Fire Protection Engineers.
- Hall, J. R., Jr., & Cote, A. E. (2008). An overview of the fire problem and fire protection. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 3-3–3-30). National Fire Protection Association.
- Hall, J. R., Jr., & Twomey, E. R. (2008). Challenges to safety in the built environment. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 1-3–1-29). National Fire Protection Association.
- Long, R. T., Jr., Colwell, J. D., Ray, R., Grossman, H. L., Thomas, B., & Strassburger, R. (2008). Passenger vehicle fires. In A. E. Cote, C. C. Grant, J. R. Hall, Jr., & R. E. Solomon (Eds.), Fire protection handbook (20th ed., pp. 21-3–21-14). National Fire Protection Association.
- Ramachandran, G. (2002b). Measuring fire consequences in economic terms (J. R. Hall Jr., Rev.). In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 5-79–5-92). National Fire Protection Association; Society of Fire Protection Engineers.