Reliability of estimates

Measures of sampling error

The survey estimates of occupational injuries and illnesses are based on a scientifically selected probability sample, rather than a census of the entire population. These sample-based estimates may differ from the results obtained from a census of the population. The sample used was one of many possible samples, each of which could have produced different estimates. The variation in the sample estimates across all possible samples that could have been drawn is measured by the relative standard error. The relative standard error is used to calculate a "confidence interval" around a sample estimate.

The 95-percent confidence interval is the interval centered at the sample estimate and includes all values within two times the estimate's standard error. If several different samples were selected to estimate the population value (e.g., injury and illness incidence rates), the 95-percent confidence interval would include the true population value approximately 95 percent of the time.

For example, the total injury and illness case incidence rate of 9.1 for health services (SIC 80) in 1996 had an estimated relative standard error of 3 percent. Hence, we are 95-percent confident that the interval between 8.6 and 9.6 (or 9.1 + (2 x 9.1 x 0.03)) includes the true rate for total cases for health services.

Nonsampling error

Although not measured, nonsampling errors will always occur when statistics are gathered. The inability to obtain information about all cases in the sample, mistakes in recording or coding the data, and definitional difficulties are general examples of nonsampling error in the survey. The Bureau has implemented quality assurance procedures to reduce nonsampling error in the survey, including a rigorous training program for State coders, mechanical edits that identify questionable entries, and a continuing effort to encourage survey participants to respond fully and accurately to all survey elements.

Last Modified Date: October 16, 2001