ANOVA's Flashcards
One-Way ANOVA definition
One factor with 2 or more levels.
H0 and H1 of one-way ANOVA
H0: μ1 = μ2 = μ3.
H1 = not all μ’s are the same.
Assumptions of a One-way ANOVA
1) Population normal within each group.
2) Homogeneity of variance of population.
3) Independence of observations.
What are the 2 sources of variance in a One-Way ANOVA?
The variance explained by the model (MSM) & The variance within groups, or the residual variance (MSR)
MSm refers to…
Variance between groups that is due to the IV.
MSr refers to….
The random variation in the scores for the subjects within each group.
If there’s a large difference between groups, we have a ___ MSm and a ____ F ratio.
High, high.
F = (df___, df____)
dfm, dfr
With only two groups, either a t-test or an F-test can be used for testing for a significant difference between means. What equation represents this?
F = t**2
What are the two size measures that can be used?
eta squared (η2) & omega squared (ω2)
Which size measure (between η2 & ω2) is biased and which is unbiased?
eta squared (η2) is positively biased.
omega squared (ω2) is an unbiased estimator.
What’s a small, medium and large effect?
Small: 0.01
Medium: 0.06
Large: 0.14
What happens if assumptions are violated in a One-way ANOVA?
F ratio no longer follow F-distribution.
5 Assumptions in between-subjects ANOVA
1) Independence of observations (1 person doesn’t affect another person’s performance)
2-3) Identical distribution (within and between groups)
3) Homogeneity of variance (homoscedasticity)
4) Normal distribution
How can we violate the “independence of observations” assumption & what are the consequences of its violation?
1) Sampled from the same natural class (e.g friend)
2) Underestimation of true variability (MSr), Increase Type 1 error -> Sample size doesn’t solve problem
What is needed for the “identical distribution within groups” assumption? What are the consequences of its violation?
Needed: We don’t know more about 1 participant’s score than we do about others. No subtypes of population in the same group.
Consequences: Not accurate representation of population of interest - biased results. MSr inflated which reduces the power of the test.
How can we resolve the problem of “independence of observations & “identical distribution within groups”?
Random sampling
What is needed for the “identical distribution between groups” assumption?
Groups differ only by their means (same variance, skew…).
What is needed for the “homo of variance” assumption? How is it violated? What are the consequences of its violation?
Needed: Variance is the same for all groups.
Violated: groups defined by classification factor (e.g. nationality), problems with experimental manipulation.
Consequences: High type 1 error.
How can we alleviate some of the inflation of Type 1 error rate if homo of variance violated?
Higher sample size + equal groups
What are the consequences of violating the normal distribution assumption? (2)
F ratio is derived from the normality assumption.
Produce Type 1 error rate LOWER than nominal value -> Difficult to reject H0.
Hypotheses in a Two-Way ANOVA
1) Main effects (For factor A & B):
H0a : μa1 = μa2 = …
H1a : Not all μag are the same
2) Interaction effects:
H0axb : All μAgBj are equal.
H1axb : Not all AgBj are equal.
4 methods to assess normality
Skewness statistic, Kolmogorov-Smirnov test, Shapiro-Wilk test, normal quantile plot
Null hypothesis in Skewness statistic (assess normality)
H0: Skew is 0 in the population.
Which is the most powerful between the Kolmogorov-Smirnov & and Shapiro-Wilk test?
What are their H0?
Shapiro-Wilk test
H0: Distribution of the variable is normal.
Limitation of the normality tests
It is easy to find significant results (reject null hypothesis that data is normal) when sample size is large.