ppt课件-anova-universityofcalifornia,santabarbara.pptVIP

ANOVA ESM 206 11 February 2002 Metal toxicity in fish Clark Fork River (Montana) contaminated with mixture of toxic metals Can trout develop “resistance” to metals toxicity by being exposed to low concentrations of metals? Data Three groups of fish (about 60 each): Hatchery brown trout Hatchery rainbow trout Clark Fork River brown trout Two treatments (half of each group): Control: keep fish in clean water for 3 weeks Treatment: keep fish in weakly contaminated water for 3 weeks All fish survived this stage Measurement Expose fish to highest Clark Fork River concentrations Measure time to death in hours Multiple Comparisons We have established that independent variable doesn’t have the same mean across all levels of the independent variable… …but which means are different? With n levels, there are n(n-1)/2 comparisons Do a bunch of t-tests… …but the more tests we do, the more likely we are to find a small P-value Bonferroni correction Comparison-wise error rate: Probability of type-I error in a given test Family-wise error rate: Probability of at least one type I error in all the tests Also called experiment-wise error To obtain family-wise error rate a across m tests, set comparison-wise rate to Two-factor ANOVA Assumptions of ANOVA Metals fish The factors (independent variables) are Species/source of fish Acclimation treatment yes/no The variances are very different among groups Range from 20 to 1200 Log transform greatly reduces this Splus ANOVA table Multiple comparisons Control vs. treatment, for each fish (3) Control fish vs. one another (3) Treatment fish vs. one another (3) Treatment effect per fish, vs. one another (3) For experiment-wise a of 0.05, need to set test-wise a to 0.004 * * MSE IJ(m-1) SSE Error MSAB/MSE MSAB (I-1)(J-1) SSAB Interaction MSB/MSE MSB J-1 SSB Factor B MSA/MSE MSA I-1 SSA Factor A F Mean square df Sum of squares Source of variation Biased, Wrong inf. True model nonlinear Unbiased, extra noise Included Regressors Biased Omitted Vari