Elsevier

Accident Analysis & Prevention

Volume 48, September 2012, Pages 480-494
Accident Analysis & Prevention

Comparing the performance of residential fire sprinklers with other life-safety technologies

https://doi.org/10.1016/j.aap.2012.03.006Get rights and content

Abstract

Residential fire sprinklers have long proven themselves as life-safety technologies to the fire service community. Yet, about 1% of all one- and two-family dwelling fires occur in homes protected by sprinklers. It has been argued that measured sprinkler performance has ignored factors confounding the relationship between sprinkler use and performance. In this analysis, sprinkler performance is measured by comparing ‘like’ structure fires, while conditioning on smoke detection technology and neighborhood housing and socioeconomic conditions, using propensity score matching. Results show that residential fire sprinklers protect occupant and firefighter health and safety, and are comparable to other life-safety technologies.

Highlights

Propensity score matching was used to evaluate the performance of fire sprinklers. ► This matching procedure facilitates comparison between similar fire incidents. ► Results show sprinklers offer life-safety protection over that given by smoke alarms. ► We estimate another 194 deaths a year could be prevented if 10% home had sprinklers.

Introduction

In 2007, 300,500 residential fires in the U.S. resulted in 2350 civilian deaths, 9650 civilian injuries, and $6.2 billion in direct property damages (NFPA, 2008). These numbers have been fairly constant over the last decade. Back in 1975, in an effort to limit fire-related fatalities, injuries, and economic damages, the National Fire Protection Association (NFPA) produced NFPA standard 13D, Standard for the Installation of Sprinkler Systems in One- and Two-Family Dwellings and Manufactured Homes, to ensure fire sprinklers installed in residences would limit flashover and allow for occupant egress and rescue, while maintaining system affordability (Madrzykowski and Fleming, 2008). Since then residential fire sprinklers have proven themselves as life-safety technologies to the fire service community, but homeowner adoption has been remarkably slow. The U.S. Fire Administration (USFA) advocates that all homes be equipped with smoke alarms and automatic fire sprinklers (USFA, 2008a). Still less than 1% of all homes involved in fires are protected by an automatic fire sprinkler system (Butry et al., 2007).

Only recently have residential fire sprinklers been shown to be cost-effective for homeowners, as installation costs have fallen and safety performance has been demonstrated statistically system (Butry et al., 2007, Butry, 2008). Earlier sprinkler research (Harmathy, 1988, Juas and Mattson, 1994, Ruegg and Fuller, 1984) suggested mixed economic performance, at best, which may explain the low historical adoption rate. Thus it was only recently that the International Code Council (ICC) passed a modification (RB64) to the International Residential Code (IRC) requiring all new one- and two-family dwellings to be equipped with a fire sprinkler system beginning in 2011. The IRC is a set of model codes currently in use by 49 states and the District of Columbia.

While recent research has demonstrated that residential fire sprinklers yield reductions in fatalities, injuries, and property damage, it relied on the use of a national estimates approach, and ignored underlying differences between sprinklered and non-sprinklered homes related to structural characteristics, construction materials, demographical factors, and the presence of other fire safety technologies. The national estimates approach, as detailed in Hall and Harwood (1989), scales individual fire incidents reported to the USFA National Fire Incident Reporting System (NFIRS) by participating states with data provided by the NFPA.1

Producing national estimates to compare fatalities and injuries in sprinklered homes with non-sprinklered homes is possible, but scaling incident data parsed into groupings related to sprinkler type, dwelling age, and presence of other fire safety technology (i.e., smoke alarm type), to name only a few differences potentially confounding performance, is extremely challenging, if at all feasible. A deficiency occurs in reporting detailed fire statistics regarding sprinklered and non-sprinklered homes that has less to do with the validity of the national estimates approach, but rather, with the volume (or lack thereof) of observational data used to generate these statistics. Only a small number of reported fires occur in sprinklered homes, and a few incidents (i.e., outliers) may skew the statistics and the findings. Certainly if underlying factors confound the relationship between sprinklers use and the incidence of fatalities and injuries, and these factors are ignored in statistical analyses, the measured effectiveness of sprinklers may be misleading. Could structure age or the presence of other fire safety technology be affecting the number of fire-related deaths, injuries, and property damages, and be highly correlated with the presence of sprinklers?

In this analysis, a nationwide, but community level, examination of sprinkler performance is conducted by comparing ‘like’ structure fires in one- and two-family dwellings, while conditioning on smoke detection technology and neighborhood housing and socioeconomic conditions, using propensity score matching. Performance is measured as the reduction in fatalities, injuries, and property losses. Propensity score matching (PSM), commonly employed in labor economics and epidemiology to measure the effects of programs and treatments, facilitates comparisons between ‘treated’ units (e.g., sprinklered homes) with similar ‘untreated’ units (e.g., non-sprinklered homes). Matching is conducted over a set of covariates known to affect both treatment status and treatment performance, thereby eliminating the effects of confounding factors. An advantage of this method over a national estimates approach is that it is perfectly designed to handle the high dimensionality of confounders.

This research makes the following contributions to the literature: (1) compares sprinkler performance, measured using PSM, with estimates produced by the national estimates approach; (2) investigates the potential impact that residential sprinklers would have had in 2006 if they had accounted for 10% of the housing stock (i.e., simulates the effect of greater market penetration); (3) benchmarks the life-safety performance of residential fire sprinklers against automotive seat belts and air bags; and (4) compares the cost of life saved of sprinklers versus other common life-safety technologies published in the literature.

The rest of the paper is as follows: Section 2 provides the methods used in this analysis and presents the propensity score matching model; Section 3 describes the data used; Section 4 presents the results; Section 5 discusses the results and places them in context to other common life-safety technologies—automotive seat belts and air bags; Section 6 concludes with some final thoughts.

Section snippets

Methods

In randomized studies, the probability of treatment is set by the researcher to be the same across subjects. Thus, any difference in the outcome from the experiment, between the treatment and control groups, is causally due to the treatment. In non-randomized observational studies, the probability of treatment varies across subjects (observations) and is neither defined nor observed by the analyst. When treatment status occurs through a self-selection process, meaning subject-specific factors

Data

Data on fire incidents were taken from the USFA's NFIRS 5.0 over 2002–2006. In this analysis, fire incidents were identified for all reported structure fires occurring in one- and two-family dwellings,7 which were not under construction and contained reported information regarding the smoke alarm and

Results

Of the 11,645 observations available for analysis, only 11,183 were used in the analysis. All observations (458) from Birmingham, Alabama were dropped due to the apparent misreporting of sprinklered dwellings,9 while another four were dropped due to no reported population found in the census block group. The propensity score estimation model was highly significant and explained 13% of the

Discussion

The full sample analysis found significant treatment effects on civilian fatalities and injuries, and on firefighter injuries, while a sub-analysis restricted to homes with battery-powered smoke alarms found significant effects on civilian fatalities, firefighter injuries, and property damage. The sub-analysis restricted to homes with hardwired smoke alarms indicated that treatment effects existed on civilian fatalities and firefighter injuries. These analyses provide strong evidence that

Conclusion

This analysis has taken care to control for those factors confounding the measured effect sprinklers have on fatalities, injuries, and property damage. Potential confounders included factors related to smoke alarm technology, community housing characteristics and socioeconomic characteristics. Notably, factors related to structure age, size, and occupants of the community were controlled. Propensity score matching was used to direct comparisons between sprinklered and non-sprinklered homes.

References (52)

  • R.E. Bellman

    Adaptive Control Processes

    (1961)
  • M.H. Brown

    Economic Analysis of Residential Fire Sprinkler Systems

    (2005)
  • D.T. Butry

    Economic performance of residential fire sprinkler systems

    Fire Technology

    (2008)
  • D.T. Butry et al.

    Benefit-Cost Analysis of Residential Fire Sprinkler Systems

    (2007)
  • Census, 2006. American housing survey. U.S. Census Bureau (Census). http://www.census.gov/hhes/www/housing/ahs/ahs.html...
  • Census, 2007. American housing survey for the United States: 2007. Current Housing Reports, Series H150/07, U.S....
  • G. Chow

    Demand for Automobiles in the United States

    (1957)
  • R.H. Dehejia et al.

    Causal effects in nonexperimental studies: reevaluating the evaluation of training programs

    Journal of American Statistical Association

    (1999)
  • R.H. Dehejia et al.

    Propensity score-matching methods for nonexperimental causal studies

    The Review of Economics and Statistics

    (2002)
  • ESRI, 2006. ESRI Data & Maps: Block Group Layer. ESRI,...
  • ESRI, 2008. ESRI ArcMap 9.3.: ArcGIS Desktop Help. ESRI,...
  • FHA, 2007. Annual vehicle distance traveled in miles and related data by highway category and vehicle type—2006....
  • D. Glassbrenner et al.

    Seat Belt Use in 2007—Overall Results. Traffic Safety Facts: Research Note. DOT HS 810 841

    (2007)
  • J.R. Hall et al.

    The national estimates approach to U.S. fire statistics

    Fire Technology

    (1989)
  • T.Z. Harmathy

    On the economics of mandatory sprinklering of dwellings

    Fire Technology

    (1988)
  • J.J. Heckman

    Econometric causality

    International Statistical Review

    (2008)
  • Cited by (0)

    View full text