Week 3 - Non-parametric tests Flashcards
Why can’t we run parametric test on non-normal data?
- Because the test will not give us an accurate p-value
Which of the following cannot be done with a non-normal distribution so we can analyse it?
PARTITIONING: partitioning involves dividing the data into subsets and does not address the underlying problem of the non-normal distribution of the data.
Non-parametric tests…
TREAT ALL DATA AS ORDINAL: meaning they use the ranks or order of the data rather than the actual data values.
To test for normality we need to…
Divide skewness and kurtosis by their respective standard error.
In repeated measures (within participants) design, to run a parametric test we need to check…..
That the score differences are normally distributed:
- For a repeated measures design, instead of testing the normality of the data in each of the two conditions, you first have to create a new variable ‘difference’, which is the difference between each person’s two scores.
- You then need to check that the distribution of this new variable is normal.
In a Mann-Whitney tests, if two groups of scores are statistically different…
The rank totals of the groups are different
- The Mann-Whitney U test is a non-parametric test that assesses whether there are differences between two independent groups.
* It does this by comparing the ranks of the scores (rather than the scores themselves) in each group. If the groups are statistically different, it means that their rank totals are different
When testing for normality in an independent samples (between-groups) desing, we need to report…
One histogram for each condition
When testing for normality in a repeated
measures (within-groups) design, we
need to report…
One histogram for the score differences
In a study with two conditions, participants only take part in a single condition and all the data is normally distributed….
we can run an independent samples t-test
We need to run a Wilcoxon signed-ranked test when…
We cannot assume normality of the score differences.
- For within-subject (or repeated measures) design we need to first calculate the difference scores
- We then need to check the normality of the score differences
- If the difference scores are not normally distributed, we would need to run a wilcoxon signed-rank test
The Wilcoxon signed-ranked W statistic is the sum of…
Either the positive or negative ranks with the smaller absolute value.
- The Wilcoxon signed-rank W statistic is
calculated by taking the sum of the ranks for he positive differences and the sum of the ranks for the negative differences, and then using the smaller of the two sums. - This test is used to determine whether there are significant differences between paired observations.
When we report the result of non-parametric test, we must also report…
The total number of participants.
- We also need to report the median, instead of the mean, as this is a better representaiton of the central tendency
