Variabilityinmeasurement discussion paper

a rifle at a target that is not in focus. The target may correctly yield the proper direction in which one should be aiming, but the blurry picture makes it difficult to hit the bulls-eye, causing bullets to scatter all over the target. Increasing the sample size of a study or the number of measurements will yield greater precision. For example, in , a prospective study of the of cardiovascular disease in a small, rural Louisiana community, an average of six was used to characterize an individual childs blood pressure.Each child was randomly assigned to two of three trained observers who each made three independent blood pressure measurements. By taking the average of six readings, the random error was reduced, thereby improving precision.

Samplingerror

Samplingerroris a type of error that arises when obtained sample values (statistics) differ from the values (parameters) of the parent population. Sampling error is relevant to all types of epidemiologic studies: cross-sectional, case-control, cohort, or intervention.

In epidemiologic research, one wishes to make inferences about a target population without necessarily having to measure each member of the target population. The target population may be the general population of the entire United States or a specified subset (e.g., residents of California; children aged 59; African Americans; or Hmong residents of the Minneapolis-St. Paul area of Minnesota). For this reason, one typically selects samples from the target population that are of a more manageable size for study than would occur if every member of the target population was examined.

When one conducts a case-control study of colorectal cancer in the state of Utah, the study group of cases may be considered a sample of all cases of colorectal cancer in the United States. When one draws a sample from a larger population, the possibility always exists that the sample selected is not representative of the target population. Nonrepresentative samples may occur without any intention or fault of the investigators even if subjects are randomly selected. To a certain extent, sampling error may be thought of as just plain bad luck of the draw, just as there can be an unusual run of cards in poker or run of colors in a roulette game. Although there is no way to prevent a nonrepresentative sample from occurring, increasing the size of the sample can reduce the likelihood of its happening.

The validity of a study will be enhanced greatly if the data that are collected are objective, reliable, accurate, and reproducible. Even under the best of circumstances, however, errors in measurement can and do occur. For example, theBogalusa Heart Study investigators were concerned about the stability of laboratory measures over long periods of time.To determine consistency in measurement, a blind sample from randomly selected individuals was included when samples of blood were sent to the laboratory for analysis. In fact, perfect agreement was rarely achieved despite the fact that the same procedures were used and the samples were from the same individuals and collected at the same time. The lack of agreement in results from time to time reflects random error inherent in the type of measurement procedure employed.