The following is the US Consumer Product Safety Commission (CPSC) Product Summary Report issued with every annual report of the injury estimates for the 15,000 products the CPSC monitors. It has remained virtually the same since it was revised in 1990 when the CPSC changed their sample.

The CPSC sends this out with their reports and often with any requests for data from these reports. NEISS has a number of other documents that describes how this sytem was developed and how they account for the variations in their estimates.

PRODUCT SUMMARY REPORT

Injury Estimates for Calendar Year 199#

National Electronic Injury Surveillance System

National Injury Information Clearinghouse

U.S. Consumer Product Safety Commission

Washington DC 20207--0001

NEISS data and estimates are based on injuries treated in hospital emergency rooms that patients say are related to products. Therefore it is incorrect when using NEISS data, to say he injuries were caused by the product.

Overall, NEISS estimates for 1990 are 18 percent higher than estimates for the past few years. This reflects an improved sample, updated to represent a more current universe of hospital emergency rooms in the U.S. The 1990 NEISS sample update takes into account changes between the 1975 and 1985 sampling frames. The changes include the opening of new hospital emergency rooms, the closing of some others, and substantial changes in caseloads in many hospital emergency rooms

Because of this sample update and overall increase in the NEISS estimates, caution is suggested when analyzing NEISS data for trends. If you have any suggestions on the use of these data, please contact the Clearinghouse.

The Product Summary Report is a compilation of information derived from product-associated injuries treated in hospital emergency rooms participating in the National Electronic Injury Surveillance System (NEISS). The NEISS estimates are calculated from a sample of hospitals which are statistically representative of institutions with emergency treatment departments located within the United states and its Territories.

1. Some individual estimates may be refined by the staff at a later date to include follow-up data obtained through further analysis and special studies. Thus, they could possibly differ from individual estimates in this report.
2. The data are specie to product codes and arc presented in percentage of estimate by age group and disposition.
3. The column titled "Sample Count" is the actual number of injury cases reported through to NEISS emergency rooms for the specific product code. The "National Estimate" is a projection to the total U.S. population treated in emergency departments based upon the number in the sample.
4. The Commission recognizes that estimates based on small samples if carefully interpreted -- are useful. However, the Commission policy is to disseminate only national estimates that meet all three of the these criteria: (1) coefficients of variation (CV) that are 33 percent or smaller; (2) national estimates that are 1200 or greater; and (3) national estimates that are based on a sample count of 20 or greater. Estimates which do not meet one of these criteria are shown with an asterisk,

There are two types of variation possible in an estimate based on a sample survey: sampling and nonsampling. Sampling errors occur because observations are made only on a sample, not on the entire population. The standard error is primarily a measure of variability; -- that is, the variations that might occur by chance because only one of a large number of all possible samples of a given size was surveyed. If the survey was repeated, the chances are about 68 out of 100 times that an estimate from the sample would differ from the average result of alt possible samples of the same size by less than the standard error. The CV (coefficient of variation) listed in the second column of the attached table represents the value of the standard error of the estimate divided by that estimate.

For example, a product code might have an estimate of 5790 injuries and a CV of 0.18. These numbers multiplied produce a standard error of plus-or-minus 1042 injuries. The range of 4748 (5790 minus 1042) to 6832 (5790 plus 1042) represents a 68 percent confidence interval. However, it is customary to use a 95 percent confidence interval. For the customary 95 percent, multiply the standard error by 1.96, which in this example, produces a range of 5790 plus-or-minus 2042.

Nonsampling errors can be attributed to many sources: for example, inability to obtain information about all cases in the sample, definitional difficulties, mistakes in recording or coding the data obtained, and other errors of collection, response, processing, coverage, and estimation of missing data.

When the NEISS sample was redesigned in 1978, Westat Inc. produced generalized sampling error estimates for NESS data by applying the curve-fitting procedure used at the Bureau of the Census. The generalized sampling errors are smoothed values that are derived from fitting a curve through points determined by estimates and calculated sampling errors for a defined set of product groups. This procedure has been used with current annual NEISS injury estimates of 1200 or more to produce the table which follows. The reader should bear in mind that, because the table values are generalized estimates of relative sampling errors, most likely they will differ from the calculated values for a given estimate.

The table presents generalized estimates of relative sampling errors for NEISS injury estimates of 1200 or more. The estimated relative percent error shown is a measure of the estimated sampling error of the injury estimate expressed as a percent of the estimate. Often this is referred to as the coefficient of variation (c.v.) of the estimate.

Example of use of table: an aggregate NEISS estimate of 50,000 injuries has an estimated relative sampling error (standard error) of 12 percent, or an estimated sampling error of 6000 (12 percent of 50,000).

A 68 percent confidence interval is that interval within one sampling error of the estimate. For an estimate of 50,000, the range 44,000 (50,000 minus 6000) to 56,000 (50,000 plus 6000) represents a 68 percent confidence interval, The chances are 2 out of 3 that the interval includes the average result of all possible samples of the same size.

For a 95 percent confidence interval, the estimated sampling error is multiplied by 1.96. In that ease, the chances are approximately 19 out of 20 that the interval includes the average result of a possible samples of the same size.

-END-

Home - Mighty-Mite - Precocious - AFN - Skyrockers - 4 Sale - FAQ - WARNINGS - Request info

© April 1997, Charles P. Weeth ^™ , La Crosse, WI USA. All rights reserved worldwide.

Revised June 6, 1997