Lecture 14 - Non parametric tests

Question 1

When do we use non parametric tests?

Answer 1

• when assumptions for parametric tests are violated
• when we have nominal or ordinal data
• if your data is not or cannot be normally distributed

Question 2

What is a non parametric test?

Answer 2

• a family of statistical procedures that do not rely on the restrictive assumptions of parametric tests, in particular they do not assume that the sampling distribution is normally distributed

Question 3

What is the aim of a non-parametric tests?

Answer 3

• to turn the data from raw scores to rank scores
  ->the lowest score will be the first rank and the highest score will be the last rank
  -> if there are tied values we have to calculate the mean ranked score
• should use the median if data is ordinal or if data violated parametric assumptions

Question 4

Parametric vs non-parametric tests?

Answer 4

• non parametric tests are less powerful
• a parametric test is more likely to find a genuine effect in data than a non-parametric test as they have a greater sensitivity to data

Question 5

When do we use the Mann-Whitney test?

Answer 5

if we want to run an independent t-test but have violated the assumptions

Question 6

When do we use the sign test and the Wilcoxon test?

Answer 6

• when we want to compare the data for 2 conditions within the same group of p’s
• is based on the analysis of the rank data
• sometimes if you carry out the sign test, you may not get a significant result, but then carry out the Wilcoxon you do get a significant finding
• this is because the Wilcoxon takes into account the magnitude of the differences and not merely whether there is a positive or negative difference

Question 7

Non parametric test for 1 sample?

Answer 7

• Just like for one-sample t-test here we have only one set of data, which we want to compare to a reference value
• Just like for the repeated measures design, you can use either the sign test or the Wilcoxon
• The calculations are the same, except that instead of doing the difference between two sets of score, you calculate the difference between your sample and the reference