Statistical Engineering Division, ITL
Fire Science Division, BFRL
Cigarette ignition of soft furnishings (upholstered furniture and bedding) has long been the leading cause of fire deaths in the United States. In 1984 and 1990, Federal legislation directed research efforts to determine whether the potency of cigarettes as an ignition source could be moderated. Under this legislation NIST produced a test procedure known as the Mock-up Ignition Test Method, which was designed to distinguish the propensity of different types of cigarettes to ignite soft furnishings.
Subsequently, the Mock-up Test has been criticized for the use of a test fabric, cotton duck, that is not indicative of the performance of fabrics used in the manufacture of upholstered furniture. A joint venture of cigarette industry firms purchased approximately 500 upholstery fabrics and used them to test 4 experimental cigarettes. They concluded that most fabrics ranked cigarettes differently from the cotton duck used in the Mock-up Test.
SED statisticians were called upon to re-analyze the data from the industry-sponsored study. Using several parallel modeling procedures, we demonstrated that there was an interaction between the fabric and the ignition propensity of the cigarettes. However, we distinguished two types of interactions. In the first type, the relative magnitudes of the ignition propensity vary among the fabrics, but the rankings of the cigarettes do not. In the second type, the rankings also vary. The first type of interaction would not invalidate the Mock-up Test, whereas the second one would.
In order to determine which of the two interactions existed, we defined a consistency score that measured the agreement with the cigarette rankings from the Mock-up Test. Positive scores indicate agreement and negative scores indicate disagreement. Most of the results from the 500 fabrics were not applicable to the analysis because the fabrics either always ignited or never ignited during pre-testing or testing. Owing to the power of the study, only 41 of the 79 the applicable fabrics statistically distinguished the cigarettes.
The figure displays a histogram of the positive scores (given in the top portion of the plot) and negative scores (given inverted in the bottom portion of the plot). The 41 fabrics that significantly discriminate among the 4 cigarettes at the 5% level are emphasized with a filled square. Note three significant patterns in the figure: (1) there are substantially more fabrics with a positive score than a negative score; (2) the positive scores tend to be larger in magnitude than the negative ones; (3) considering only the fabrics that significantly discriminate, the first two notes are still true. We concluded from this analysis that, although there is an interaction between fabrics and cigarettes, for most fabrics the rankings of the cigarettes are consistent with the Mock-up Test.
Figure 17: Consistency scores.
Date created: 7/20/2001