Difference between revisions of "Statistics"
| Line 1: | Line 1: | ||
| − | =False Discovery Rate= | + | =Multiple Testing= |
| + | |||
| + | ==Bonferroni Correction== | ||
| + | |||
| + | The total number of independent tests is ''m''. The "alpha" for significant ''p''-values is adjusted by dividing alpha by ''m''. Only tests that are lower than the new number are considered significant. | ||
| + | |||
| + | The Bonferroni procedure is generally viewed as overly conservative and alternative approaches such as false discovery rate control are often favored. | ||
| + | |||
| + | ==False Discovery Rate== | ||
False discovery rate control is a procedure that is less stringent than a classical Bonferroni correction. | False discovery rate control is a procedure that is less stringent than a classical Bonferroni correction. | ||
| − | ==Benjamini–Hochberg Controll Procedure== | + | ===Benjamini–Hochberg Controll Procedure=== |
Rank the ''p''-values from smallest to largest. Begin at 1, and increase the rank: 1, 2, 3, 4, 5, ... the rank is ''k''. | Rank the ''p''-values from smallest to largest. Begin at 1, and increase the rank: 1, 2, 3, 4, 5, ... the rank is ''k''. | ||
Revision as of 07:41, 13 August 2018
Contents
Multiple Testing
Bonferroni Correction
The total number of independent tests is m. The "alpha" for significant p-values is adjusted by dividing alpha by m. Only tests that are lower than the new number are considered significant.
The Bonferroni procedure is generally viewed as overly conservative and alternative approaches such as false discovery rate control are often favored.
False Discovery Rate
False discovery rate control is a procedure that is less stringent than a classical Bonferroni correction.
Benjamini–Hochberg Controll Procedure
Rank the p-values from smallest to largest. Begin at 1, and increase the rank: 1, 2, 3, 4, 5, ... the rank is k.
Choose a false discovery rate q, for example q=0.1.
The total number of independent tests is m. If the level of gene expression for 20,000 genes is measured and tested then m=20,000.
Accept the tests with p-values lower than qk/m.
You estimate that a fraction q of these tests are false positives, due to random chance, but that the remaining ones are true positives.
Benjamini, Y., Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society, Series B. 57 (1): 289–300.
Normal Distribution
https://www.maa.org/sites/default/files/pdf/upload_library/22/Allendoerfer/stahl96.pdf
https://www.embedded.com/print/4413095
Derivation of the normal distribution from the binomial:
http://www.m-hikari.com/imf/imf-2017/9-12-2017/p/baguiIMF9-12-2017.pdf
