A criterion (referred to as "residual flame time") was developed for molten material that continued to burn on the floor of the test cabinet, limiting burn times of such material to 10 seconds. This was incorporated in the standard for children of ages 0-5 years (sizes 0-6X), because it was noted that flaming material falling from a garment represented two additional risk factors. The potential existed for the falling material to cause relatively deep localized burns and increased the likelihood of secondary ignition of other materials thereby increasing the extent of injury. However, there is a difference between hazard and an unreasonable risk. These flaming drips can be considered an unreasonable risk to children that are relatively immobile (i.e. sizes 0-6X covered by DOC FF-3-71), but are less of a risk to older more mobile children (i.e. sizes 7-14). The proposed acceptance criteria for children's sleepwear sizes 7-14, therefore, differ from the criteria found in DOC FF-3-71. For children's sleepwear sizes 7-14 (PFF-5-73), the residual flame time criterion is deleted. In DOC FF-3-71, the most critical criterion is that the average char length shall not exceed 7 inches for a set of 5 replicate specimens. Secondary criteria are that no single specimen have a char length of 10 inches (specimen size: 3.5" X 10") and that no single specimen exceed the 10 second limit on residual flame time. The latter two criteria are secondary only in that a failure of average char length results in a unit rejection under the accompanying sampling plan. The same is not true of the single specimen criterion, which allows for sample retesting before unit rejection occurs. 4. CONCLUSIONS A description of the methodology used in developing the children's sleepwear standards clearly establishes the need for a synthesis between accident data and laboratory experimentation. The basic approach taken has application in areas other than flammability. Any consumer oriented product safety standard must consider the best available end use and accident data and, at the same time, represent state-of-the-art technology with regard to repeatability and reproducibility. The continued surveillance of accidents should, in several years, indicate the success or failure of this approach. [1] The Flammable Fabrics Act, 81 Stat. 568, 15 U.S.C. 1191, as amended and revised Dec. 14, 1967. [2] [3] [4] Children's Sleepwear Sizes 7 Through 14, Notice of Proposed Flammability Standard, DOC PFF-5-73, Federal Register, 38, no. 47 (Mar. 12, 1973), 6700-6710. The Standard For the Flammability of Children's Sleepwear, DOC FF-3-71, [5] The Flammable Fabrics Program 1970, Nat. Bur. Stand. (U.S.), Tech. Note 596 (Sept. 1971), 28. [6] [7] U. S. Bureau of the Census, 1960 Census of Population, United States, [8] [9] [10] [11] Mattresses, Notice of Finding that a Flammability Standard May Be Needed, McCarter, R. J., The Cause of Anomalous Behavior in the Vertical Flamma- Proposed Standard For the Flammability of Children's Sleepwear, DOC NATIONAL BUREAU OF STANDARDS SPECIAL PUBLICATION 411, Fire Safety Research, Proceedings of a Symposium Held at NBS, Gaithersburg, Md., August 22, 1973, (Issued November 1974) SAMPLING PLANS IN MANDATORY STANDARDS Paul Gottfried1 Consumer Product Safety Commission, Washington, D.C. The Children's Sleepwear Standard for sizes 0-6X provides an example of the necessary interplay between sampling theory, industry capabilities and consumer safety assurance. The sampling plan imposes requirements for the fabric, for garment design and for garment production. These requirements address the differences between hazards due to design and those due to production_errors. The requirements also interact to provide improved assurance of safety by limiting the potential for severe injury. Key words: Flammable fabrics; product safety; sampling; standards. 1. THE NEED FOR SAMPLING The Flammable Fabrics Act and other product safety legislation administered by the Consumer Product Safety Commission do not in themselves require product testing; they require only that products comply with applicable standards developed under such legislative authority. In the past, standards have been enforced on the basis of compliance testing done by a regulatory agency. A failure discovered by the agency would result in an action against the manufacturer or his product, causing him to cease production of that particular product and withdraw that product from the market. Under such procedures, it is commonly found that failures are not detected until large quantities of the product are in the hands of retailers and and consuming public. Demands for "one hundred percent compliance" in this "post-mortem mode" can protect the consumer only after the fact and can cause severe hardship to manufactuers as well. A more satisfactory way of protecting the consumer and limiting the risk to be accepted by the honest manufacturer is to require testing prior to marketing. When defective products are prevented from reaching the market, the consumer is protected more effectively and the risks and costs of product recall are decreased. At the same time, the amount of testing that can be done is increased because industry resources as well as Commission resources can be put to use. When testing is destructive in nature, as it generally must be in the flammability context, it obviously is necessary to resort to sampling. Even in other situations, sampling may be the preferred approach because one hundred percent inspection can be prohibitively expensive. Furthermore, one hundred percent inspection is generally found to be much less than one hundred percent effective; effectiveness on the order of 70 to 80 percent is observed commonly. Finally, any regulatory agency is inherently limited to sampling; it could not obtain all of each product even if that were permitted by its resources. 2. EFFECTIVENESS OF SAMPLING Few, if any, manufacturing processes are capable of turning out a consistently perfect product, just as few, if any, test procedures yield results of absolute accuracy and reproducibility. Sampling has similar limitations, and these arise from theory as well as from reality. Every sampling plan carries 1The author was a consultant to the Consumer Product Safety Commission at the time this paper was presented. The opinions expressed are those of the author and do not necessarily reflect the views of the Commission. with it a finite probability of accepting a "bad" lot of products and of rejecting a "good" lot of products. Increasing the sample size can reduce these probabilities, but with diminishing returns at essentially linear increases in cost. In quality control, the conventional buyer-seller relationship has led to the development of several concepts closely related to the problems of "error" in acceptance and rejection. Among these concepts are: Concentration on the "consumer risk" or "lot tolerance percent defective" -- the probability of erroneous acceptance and the corresponding fraction defective when isolated lots are sampled. - The desirability of reduction of both risks, at the expense of increased sample size, for large lots in view of the disproportionately increased severity of the consequences of error. The notion of the "average outgoing quality limit" or AOQL arising These conventional concepts have virtually no applicability in the context of product safety and destructive testing. The notion of AOQL is obviously inapplicable when one hundred percent inspection is impossible. From the consumer standpoint, lot size is irrelevant because no individual consumer acquires more than a very small fraction of any lot. Downgrading of lot tolerance percent defective concepts arises from less obvious considerations and hence commands more of our attention. The costliness of destructive testing discourages the use of large samples and this in turn makes it difficult to limit comsumer risk in the usual ways. Fortunately, the physical phenomena that lead to destructive testing also lead to strong economic incentives on the manufacturer: in general, it is at best very difficult to rework a rejected lot and alternate markets are virtually nonexistent. The consequence is that the manufacturer must achieve a very high probability or frequency of lot acceptance in order to survive as a producer. We estimate we think conservatively that at least a 95% acceptance probability must be maintained routinely; that is, no manufacturer is expected to survive if more than one of every twenty lots is rejected. This means that ordinarily only the behavior of the sampling plan at low levels of "defectiveness" is of major concern; of course; we will still want to limit the chances that an occasional defective lot will slip through, but this is not as compelling a consideration as it is in conventional situations. - - 3. CHOOSING A TARGET - Having established that the manufacturer will need to control his process so as to achieve at least 95% acceptance probability leaves open the question of what level of quality -- what process average should correspond to this target value. Here, we enter an area demanding many tradeoffs which often must be made with incomplete information, largely because the response of an industry to the challenges posed by a mandatory standard is not entirely predictable. There is a better than even chance that any standard, especially when flammability is involved, will drastically reduce available choices among materials and designs. This implies reduced consumer freedom-of-choice and increased costs that ultimately impact on the consumer. When a product is a necessity, as is generally the case for basic wearing apparel, a further major consideration is the need to assure adequate availability of the product. (Nudism is a sure cure for garment flammability problems, but might introduce some health problems of its own.) Large cost increases or availability reductions for other textile-related products whose purchase is more discretionary, such as a replacement mattress, may have effects that are counterproductive from the standpoint of consumer safety. The delayed replacement of existing hazardous products in the interests of introducing a nearly perfect product rather than a merely superior one can mean avoidable casualties in the interim. A prospective standard, including those aspects of stringency that arise from its sampling plan, therefore must be scrutinized carefully from the standpoint of its total impact on the public and the industry. 4. AN EXAMPLE: CHILDREN'S SLEEPWEAR STANDARD (SIZES 0-6X) The standard for the flammability of children's sleepwear (sizes 0-6X), which went into full effect recently, provides illustrations of a number of the points in this discussion. It also provides an example of the way in which a multi-tiered approach can reduce the burden of consumer product safety improve ment. In many products, it is possible and appropriate to distinguish between hazards associated with the design characteristics of the product and those that can arise due to production errors. These distinctions lead to two sets of tests, each conducted on a different sampling basis. Design characteristics are evaluated appropriately by relatively stringent tests conducted on prototype (preproduction or early production) items; under some circumstances, it may be desirable to require that these tests be repeated at relatively long intervals. The presence of production errors, on the other hand, requires detection by the use of representative samples from production quantities. the extent permitted by procurement practices, processes and industry structure, such samples should be drawn from internally homogeneous lots of production items. To reduce potential confusion with definitions of "lot" that may be traditional to an industry, standards now refer to these homogeneous production quantities as "production units", whose maximum size is limited by the standard. - Children's sleepwear is unusual in that a major component - Το the fabric is itself amenable to control by a standard. This standard therefore involves a sequence of fabric production testing garment prototype testing garment production testing. (Fabric prototype testing was judged to have no special value or validity, as the dominant problems are of the production error type.) In normal sampling, the production test employs ten fabric specimens. Each of two samples contains three specimens taken in one direction and two in the other, as the flammability characteristics may differ along warp and fill. The average char length over each five-specimen sample must not exceed seven inches; any violation of this requirement leads to rejection, so excessive average char length is treated as a "critical" defect. Any individual specimen fails if it chars its entire ten-inch length or if flaming of any fragments or molten material persists for more than ten seconds. If any single specimen fails, an additional sample of five specimens (all aligned in the direction experiencing the failure) must pass without further failure. If more than one specimen fails, the production unit is rejected. - It is worth noting that even a fabric production unit that is substandard i.e., has a specimen fraction defective above the target value and is passed as a result of statistical phenomena is certain to be vastly superior to fabrics in common use prior to issuance of the standard. Some of the latter fabrics consistently would burn entirely and, if tested, would certainly be rejected. Any "defective" fabric passed under the sampling plan could be defective only in the sense that it had locally variable characteristics such as might result from uneven application of flame retardants; such a fabric would |