Statistical Testing and Type I Error.
Saved in:
| Title: | Statistical Testing and Type I Error. |
|---|---|
| Language: | English |
| Authors: | White, Amy E. |
| Peer Reviewed: | N |
| Page Count: | 16 |
| Publication Date: | 2000 |
| Document Type: | Reports - Evaluative Speeches/Meeting Papers |
| Descriptors: | Hypothesis Testing, Statistical Analysis |
| Abstract: | In traditional null hypothesis testing, researchers use critical values of various test statistics in order to minimize the risk of making Type I errors. These critical values are associated with common alpha levels (e.g., 0.01, 0.05) that indicate the probability of a Type I error. Alpha values are set at conservative levels such that the Type I error is at a minimum for any given test. However, as the number of statistical tests conducted on a single sample increases, the change of making a Type I error somewhere in the study escalates appreciably, becoming much larger than the alpha level associated with any single test. An educational research data set is used to illustrate this problem, and several corrections are suggested, including better specification of research questions, increased parsimony in selection of statistical tests, and use of multivariate methods when possible. (Contains 3 tables and 16 references.) (Author/SLD) |
| Entry Date: | 2001 |
| Accession Number: | ED445080 |
| Database: | ERIC |
| Abstract: | In traditional null hypothesis testing, researchers use critical values of various test statistics in order to minimize the risk of making Type I errors. These critical values are associated with common alpha levels (e.g., 0.01, 0.05) that indicate the probability of a Type I error. Alpha values are set at conservative levels such that the Type I error is at a minimum for any given test. However, as the number of statistical tests conducted on a single sample increases, the change of making a Type I error somewhere in the study escalates appreciably, becoming much larger than the alpha level associated with any single test. An educational research data set is used to illustrate this problem, and several corrections are suggested, including better specification of research questions, increased parsimony in selection of statistical tests, and use of multivariate methods when possible. (Contains 3 tables and 16 references.) (Author/SLD) |
|---|
