What is the non-parametric equivalent of a one-way ANOVA? RM ANOVA?
Kruskal-wallis; Friedman
Why are non-parametric tests named so?
They do not involve the estimation of any population parameters
What is special about non-parametric tests (how they are calculated)? 3
- They dont involve estimating population parameters 2. Distribution free - dont assume SND shape 3. Dont assume interval or ratio scale - can use ordinal scale
What are advantages of non-parametric tests? 4
- Dont assume normality 2. Dont assume homogeneity of variance 3. Can analyse small samples 4. Ranks reduce outlier effects
What are disadvantages of non-parametric tests?
- Less power when population are normally distributed 2. Increase type II errors 3. Require larger N than SND 4. Scales less sensitive than interval/ratio
What does it mean to reject the null in a non-parametric test?
The populations differ, perhaps not just on the basis of central tendency - they are not identical!
Which rank sum is the obtained Ws?
The sum of the smaller group (or the smaller if equal sized)
What alpha is used in Ws tests and T signed-ranks tests?
.05/2=.025 for a two-tailed test
When is Ws significant?
If less than or equal to critical value (opposite of other tests where we try to exceed the cutoff)
If Ws is larger than the critical what happens?
Retain Ho
What if the smaller group has a larger Ws than the bigger group?
Convert to W’s
Would you use rank sum (or signed-ranks) for sample size >50?
You can use a normal approximation method (z test)
What does the null state for signed-ranks test?
The distribution of difference scores is symmetric around 0
How do you rank difference scores of 0?
You dont, just ignore them
What is N in a sign-ranks test?
The number of non-zero difference scores
