Flashcards in Non-Parametric Alternatives to ANOVA & Statistical Power Deck (24)
What are the parametric test assumptions?
Normally distributed data.
Homogeneity of variance (for between-subjects).
What are 3 advantages of non-parametric tests?
Can use small datasets.
Easy to calculate + interpret by hand.
What is a disadvantage of non-parametric tests?
They have a lower power than parametric tests (increase in Type II error).
When is Friedman's ANOVA used?
For repeated measures where the IV has 3 or more levels.
When is Kruskal-Wallis used?
For independent measures where the IV has 3 or more levels.
What is the main statistic for Friedman's ANOVA named in SPSS?
What is the asymptotic sig./adj.sig. in SPSS?
What is the z-value named in SPSS?
The std. test statistic.
What does R stand for in both tests' formula?
R is the sum of ranks for each condition.
How do you calculate H in Kruskal-Wallis?
You rank all scores ignoring the group they belong too (as it's independent conditions).
Add up ranks for each condition and put these into the main formula.
How do you calculate Fr in Friedman's ANOVA?
You rank scores within each participant (as it's repeated conditions).
Calculate the sum and mean of ranks in each condition and put these into the main formula.
What are the two options when normality assumptions aren't met (therefore a mixed ANOVA can't be used)?
Using several non-parametric tests.
Why do we transform data?
If we have skewness or kurtosis then transforming data could potentially stop this (e.g. log transformation) and we can then use a mixed ANOVA.
What non-parametric test is used for within-subject conditions?
What non-parametric test is used for between-subject conditions?
What are the two choices when looking at the interaction of variables in several non-parametric tests?
1. Calculate the change of score and compare changes across groups using an Independent T-Test or Mann-Whitney.
2. Calculate difference between groups' scores using a Dependent T-Test or Wilcoxon.
How do you complete a Bonferroni adjustment for the number of tests used?
Bonferroni alpha = alpha / number of comparisons.
What is a Type I error?
Incorrectly rejecting the null hypothesis. Probability of this is alpha (.05).
What is a Type II error?
Incorrectly accepting the null hypothesis. Probability of this is beta.
What is the definition of power?
The ability to detect an effect if there is one.
What value should power usually be?
What are the three influences on power?
Effect size (the larger the effect, the easier it is to detect).
Alpha level (usually p < .05).
Sample size (larger samples are more representative = less error, more power).
Power, effect size, alpha level and sample size are all linked. Which one is useful to be able to calculate?
The estimated sample size needed to achieve adequate power.