Random error refers to the non-systematic reasons that estimated values deviate from the correct values. The processes that give rise to random error, either sampling or allocation across treatments, are not directly relevant to observational studies but provide the statistical framework for attempting to quantify the impact of random error. We suggest that random error should not take precedence over consideration of bias and argue against a formal interpretation of statistical significance testing. Confidence intervals provide an index of precision, and are more useful for quantifying random error in observational epidemiology. Multiple comparisons that result in identification of false positive associations due to random error can be minimized by a careful approach to data analysis and interpretation guided by subject matter knowledge. When data are explored without such guidance, as in exploratory studies examining large numbers of possible associations, interpretation of positive associations must be tempered.
Oxford Scholarship Online requires a subscription or purchase to access the full text of books within the service. Public users can however freely search the site and view the abstracts and keywords for each book and chapter.
If you think you should have access to this title, please contact your librarian.