Week 8 day 2 Flashcards
What is the Bonferonni correction test and why do we need to comparison corrections when doing multiple comparisons?
The Bonferonni correction is the original p-value multiplied by the number of tests done.
The reason we need to do corrections to our p-value is because we are doing multiple tests that all have type I error associated with them. We want the ‘family wise’ type I error rate to 5%, not each test to have a 5% type I error rate - this would increase the family type I error.
What are the limitations of Bonferonni correction test?
The Bonferonni correction is very conservative. This is good for trying to avoid type I errors, however, it increases type II errors.
What is the Holm correction and why is it superior to the Bonferonni test?
The Holm correction retains the conservative type I error adjustment as the Bonferonni, but also decreases type II error.
When reporting post-hoc t-test results in an ANOVA, do you need to report the t-stat and degrees of freedom?
No, just the p-value.
Make sure to report what correction you have used.
What are the assumptions for doing a one-way ANOVA?
- Normal distribution of residuals (the residuals are the difference between the group mean and the values for data in that group, i.e. the within groups variance).
-check using the Shapiro-Wilk test.
-If violated use the Kruskall-Wallis test. - Homeogeneity of variance across all groups.
-check using Levene’s test.
-if violated, using the Welch one-way ANOVA. - Independence of data.
What test should be used if the normal distribution of residuals and homogeneity of variance assumptions are violated?
What if just homogeneity of variance assumption is violated ?
Kruskall-Wallis test - Kruskall-Wallis test does NOT assume either of these things.
Use Welch one-way ANOVA.
Does eta-squared assume normality?
Yes. We therefore need to use another effect size if our residuals are not normally distributed.
This is the Kruskall Effect Size.
If we do a Kruskall-Wallis test, what effect size do we use and why don’t we use eta-squared?
We use the Kruskall Effect Size.
If we are doing a Kruskall-Wallis test because the residuals are not normally distributed, then we cannot use eta-squared because eta-squared also assumes normal distribution of residuals.
Does the Kruskall Effect size have the same interpretation as eta-squared? If so, what is this interpretation?
Yes. Just like eta-squared, the Kruskall Effect size can be interpreted as the percentage of variance of the outcome variable explained by the grouping variable.
How do we test whether groups have homogeneity of variance?
We do a Levene test.
If we get a significant p-value for a Levene test then there is NOT homoegenous variance across the groups.
What test can we do if there are multiple grouping variables (factors)?
We can do a two-way ANOVA.
For two-way ANOVAs do they need to in long form in R?
Yes.
Why do a two-way ANOVA and not multiple one-way ANOVAs when looking at factors/predictor variables that are likely influencing the outcome variable in different ways?
Something about residuals.
What is the sum of squares for residuals in a two-way ANOVA?
What are some of the different types of interactions discussed in this lecture?
- Cross over interactions.
- Single factor interaction.
