Final session 10b Flashcards
How nonparametric tests (presented in this lecture) work
Step 1: Transform the original data to sign (nominal) or rank (ordinal) data
Step 2: Use parametric test or a new statistic on the transformed data Step 3: Make the decision whether to reject the null hypothesis under the test
what is The median test (k = 2)
This is a sign test for two independent samples
It compares the medians of two independent samples
what are the hypotheses for the median test
This is a sign test for two independent samples
It compares the medians of two independent samples
The median test can be extended to cases when k _____
> 2
what is the Wilcoxon rank sum test
This is a rank test for two independent samples
the Wilcoxon rank sum test is equivalent to what
Mann-Whitney U test
Similar to a two sample t-test but doesn’t require the data to be normally distributed
what is the hypotheses of Wilcoxon rank sum test
H0: Two samples come from populations with the same continuous distribution
H1: Samples come from populations with different continuous distributions (H0 is not true)
stes for Wilcoxon rank sum test
Step 1: Transform original data to rank (ordinal) data
Step 2: Apply z-statistic
Step 3: Make a decision
how to deal with ties in ordinal values (Wilcoxan rank sum test)
Use average rank of tied scores
what is the Kruskal-Wallis H test
This is a rank test for k independent samples
A generalization of the Wilcoxon rank sum test to k groups Similar to a one-way ANOVA but does not require the normality assumption
what re the hypotheses for Kruskal-Wallis H test
H0: k independent samples come from the same population (no difference in the DV)
H1: At least one sample comes from a different population
what are the steps for Kruskal-Wallis H test
Step1:JointlyrankallN=N1+N2+···+Nk observations
Step 2: Compute the sums of the ranks of the k samples: R1, . . . , Rk Step 3: Calculate the H statistic:
what is the Kruskal-Wallis H test df
The distribution of H approximates the chi-square distribution with
df = k − 1
The approximation is reasonably accurate when Nj ≥ 3 and k ≥ 3
do you do Effect sizes calculations with Kruskal-Wallis H test
None that are straightforward to compute
do you need Follow-up tests for Kruskal-Wallis H test
Pair-wise comparisons are possible
Analogous to post-hoc tests from one-way ANOVA
Some adjustment of p-values ought to be made to control Type I error
