non-parametric tests Flashcards
parametric, interval scales and skew, test options, Mann-Whitney U test, Wilcoxon related-samples test, Kruskal-Wallis H test, Friedman test (21 cards)
what are parametric tests based on?
commonly used parameters which assume normal distribution (standard deviation)
what happens when data is heavily skewed?
step size between points on the scale probably not constant (i.e. scale may only be ordinal)
what happens when data is ordinal?
may not have a normal distribution and parameters like SD, SEM, variance may no longer capture the data
at simplest level, should probably be using a non-parametric test
what are non-parametric tests usually based on?
ranking scores
what is the non-parametric equivalent of Pearson’s r (correlations)?
Spearman’s Rho
what is the non-parametric equivalent on paired-samples t-test (two-samples)?
Wilcoxon T
what is the non-parametric equivalent of the unpaired-samples t-test (two-samples)?
Mann-Whitney U
what is the non-parametric equivalent of repeated-measures ANOVA (K samples)?
Friedman’s
what is the non-parametric equivalent of between-groups ANOVA (K samples)?
Kruskal-Wallis
what is the Mann-Whitney U-test?
two unpaired samples
between groups design
what is the rationale behind the Mann-Whitney U test?
if arrange all the data into ascending sequence of scores
when null hypothesis true, would expect group labels then to be randomly distributed
when null hypothesis false, would expect scores of two groups to be clustered at either end of sequence
what is the test statistic in Mann-Whitney U test?
U
what is the significance in Mann-Whitney U test?
test statistic allows us to determine significance by comparing U to some critical value in look-up table
what is the Wilcoxon related-samples test?
Wilcoxon signed-rank test
two paired samples
repeated measures design
what is the rationale behind the Wilcoxon related-samples test?
test is based on difference between the two scores for each subject
uses the direction (which score is greater) and magnitude of the difference (how much greater)
if H0 true then differences in one direction will be as large as differences in the other - that is what the test statistic will measure but it will rank magnitudes rather than the actual magnitudes so it can be used on skewed data
what happens when there are tied ranks?
whenever we have tied ranks (in the unpaired or paired test) we take the average of the range of ranks that the ties cover and allocate this to the value of the ties
do not change the other ranks
what is the Kruskal-Wallis H test?
multiple levels with between-groups design
what is the rationale behind the Kruskal-Wallis H test?
when null hypothesis is true, expect random distribution of ranks across groups - average ranks of levels in the design should be equal
when null hypothesis is false, expect a systematic distribution of ranks across ranks across groups - average ranks of levels in the design should be different
how do you test differences in the Kruskal-Wallis H test?
only tells that the three conditions differ but not which ones
can use multiple Mann-Whitney tests and apply Bonferroni correction (multiply p value by the possible number of tests)
what is the Friedman test?
multiple levels with repeated-measures design
rank within each subject
how do you test differences in the Friedman test?
doesn’t tell us which conditions different
can use multiple Bonferroni-corrected Wilcoxon tests
