non-parametric statistics

Question 1

non-parametric statistics

Answer 1

don’t make assumptions about the underlying population (as parametric ones do and assume them to be normally distributed hence…)
- ‘distribution free statistics’ (no normality assumption)
- less power than parametric
- higher type II error risk
seen as a ‘back up’

Question 2

type II errors

Answer 2

when there is a relationship, but we fail to reject null hypothesis

Question 3

independent t-test non-parametric equivalent

Answer 3

Mann-Whitney U test
- 1 IV with only 2 levels

Question 4

paired t-test non-parametric equivalent

Answer 4

Wilcoxon t-test
- 1 IV with only 2 level

Question 5

1-way independent ANOVA non-parametric equivalent

Answer 5

Kruskal Wallis Test
- 1 IV with more than 2 levels

Question 6

1-way repeated measures ANOVA non-parametric equivalent

Answer 6

Friedman test
- 1 IV with more than 2 levels

Question 7

how can normality assumption be tested

Answer 7

Shapiro-Wilk test
-checks to see if assumption of normality has been violated, if so then use non-parametric

Question 9

non-parametric equivalent for Pearson’s correlation coefficient when N > 20

Answer 9

Spearman’s rho

Question 10

non-parametric equivalent for Pearson’s r where N < 20

Answer 10

Kendall’s Tau

Question 11

continuous vs ordinal variables

Answer 11

for continuous variables e.g. age or salary use parametric

for ordinal best to use non-parametric (especially if concerned about intervals being equivalent between measurements) e.g. age (contin) and position of people in a running race (ordinal) (diff between 1st place time and 2nd would not be even)

• likert is an exception if there is 7 or more it is treated as continuous

Question 12

analysis of categorical data tests

Answer 12

no parametric equivalent
- one-variable chi-squared
- chi-squared test of independence