12. Comparison of Several Means 2 Flashcards
assumptions of an ANOVA
- normally distributed data
- homogeneity of variance
- independence
- DV should be measured on interval scale
ANOVAs and independence
applies to independent ANOVAs, not repeated measures designs
How is the ANOVA a robust test?
it is still accurate even when certain assumptions are violated
violations of normality: groups not equal
accuracy of f-stat affected by skew
violations of normality: group sizes equal, have at least 21 subjects per group, low drop out rate
f-stat not affected by skew
violations of normality: group sizes equal, at least 40 subjects per group, higher drop out rate
f-stat not affected by skew
violations of normality: homogeneity of variance
when sample sizes are equal, ANOVA is robust to violations of homogeneity of variance
planned contrasts aka
planned comparisons
When do you used planned contrasts
- when you have an apriori directional hypothesis that you want to test (1 tailed)
- only needed if the f-stat is significant
What do planned contrasts allow you to do?
know where the significant difference lies
When are post-hoc tests used?
- 2 tailed test
- only needed when f-stat is significant
What do post-hoc tests consist of?
pairwise comparisons
What are pairwise comparisons?
- performing multiple t-tests to determine where significance lies
- correcting for inflated type I error rate
What happens as type 1 error rate decreases?
type II error increases
What does an increase in type II error rate = ?
decreased power
What are the post hoc tests?
- REGWQ test or Tukey test
- Bonferroni correction
- Gabriel’s test
- Hochberg’s GT2
- Games-Howell test
What post hoc test would you use if you have equal sample sizes and group variances?
REGWQ test or Tukey test
What post hoc test would you use if you want guaranteed control over type I error rate?
Bonferroni correction
What post hoc test would you use if sample sizes are slightly different?
Gabriel’s test
What post hoc test would you use if sample sizes are very different?
Hochberg’s GT2
What post hoc test would you use if group variances are not homogeneous?
Games-Howell test
