Week 6-Friedman and Kruskal-Wallis

Question 1

Define the Friedman Test

Answer 1

The non-parametric version of the One-Way Repeated-Measures ANOVA

Question 2

What are the assumptions of the Friedman test?

Answer 2

■ 3+ conditions
■ Within-subjects design
■ You can use Ordinal, Interval, or Ratio data HOWEVER, you’d only use Friedman if it violated the
assumptions for ANOVA.

Question 3

What’s the Friedman test? (further elaboration)

Answer 3

■Sometimes called Friedman’s ANOVA
■Tests the differences between 3 or more related (within) samples of scores.
■Based on ranking the data and comparing the mean rank* of each condition.
*Despite the underlying theory working on mean ranks, you
should report medians (of the raw data; not median of ranks)

Question 4

What are Omnibus Tests?

Answer 4

■Friedman and Kruskal-Wallis are both ‘Omnibus tests’.
■These are tests which investigate if there are overall differences between several conditions.
■BUT it does not tell us exactly where the difference lies.

Question 5

What’s the issue with using more than one test to compare more than 2 groups?

Answer 5

■We have an increased likelihood of making a type 1 error (false +ve).
■So omnibus tests reduce the
familywise error rate.

Question 6

What are the 3 upsides of omnibus tests?

Answer 6

1.One test
2.One p-value
3.Reduced FWE

Question 7

What are the 2 downsides of omnibus tests?

Answer 7

1.We don’t know where the difference actually is (yet)
2.Two groups could significantly differ from each other but the whole test may not show this.

Question 8

Define a type 1/type 2 error

Answer 8

type 1=false positive
type 2=false negative

Question 9

How can we limit the chances of committing a type 1 error? aka Bonferroni correction

Answer 9

■We should correct the number of statistical tests we conduct.
■An easy way to do this is to change your alpha level accordingly. So that you no longer accept results to be significant as p<.05. Instead you divide
your alpha level (.05) by the amount of tests you are going to conduct.
■So I will no longer accept a finding to be significant unless it is below that number

Question 10

What are post-hoc tests?

Answer 10

tests conducted after a significant omnibus test to find where the effect is/differences between conditions

Question 11

Define pairwise comparisons

Answer 11

tests conducted for every pair of conditions

Question 12

Define planned comparisons

Answer 12

tests you had pre-planned (pairwise, planned and post-hoc tests can be used interchangeably)

Question 13

What’s the kruskal-wallis test?

Answer 13

The non-parametric version of the One-Way (Between-subjects) ANOVA

Question 14

What’s the ANOVA test?

Answer 14

A statistical test used to analyse the difference between the means of 2+ groups. A one-way ANOVA uses one independent variable, while a two-way ANOVA uses two independent variables

Question 15

What are the assumptions of the kruskal-wallis test?

Answer 15

■Used when we have 3+ unrelated (between-subject) conditions.
■When data is ordinal or scale (but non-parametric).
■Ideally distributions across conditions should also possess the
same shape…but again we’re not going to go into that!

Question 16

What does eta-squared tell us about our results?

Answer 16

■It tells us how much variance (in the form of a percentage) in our
results is accounted for by our IV.
Some suggested cut offs for eta squared are:
■ Small = .01
■ Medium = .06
■ Large = .14

Question 17

True or false: Post-hoc tests / pairwise comparisons are used to assess differences between conditions ONLY if the omnibus test is significant

Answer 17

True

Question 18

What did Bennett et al (2009) find with type 1 errors?

Answer 18

-put a dead salmon under an fMRI scan and found a lot of brain activity
-found this because they ran thousands of comparisons hence why it showed up as brain activity (despite being dead)

Question 19

Define familywise error rate

Answer 19

the type error 1 rate/false positives we’re going to find