Significance Testing Flashcards
When to use 1-sided p values
When difference in one direction is impossible
Non-parametric unpaired binary data test + H0
Fisher’s exact test
H0: The frequency table is as you would expect for no association
Numeric parametric paired data test + H0
Paired t-test
H0: Paired differences are on average 0
Parametric numeric unpaired with 2 groups data test + H0
Two sample t-test
H0: Mean difference of 0
Parametric numeric unpaired with more than 2 groups data test + H0
ANOVA
H0: means are the same across all groups
Non-parametric binary paired data test + H0
McNemar’s
H0: Marginal frequencies are the same
Non-parametric paired numeric data test + H0
Wilcoxon signed rank/ sign test (but sign is not a very good test)
H0: Median of paired differences is 0
Non-parametric numeric unpaired 2 group data test + H0
Mann Whitney U
H0: No differences in the ranks of groups
Non-parametric numeric more than 2 groups data test + H0
Kruskal Wallis
H0: No difference in the ranks of groups
p-value definition
Probability of getting result or more extreme if null hypothesis is true
Parametric unpaired binary data test + H0
Chi squared test
H0: No difference in % across groups
One sample parametric numeric + H0
One sample t test
H0: Hypothesised mean is true
Chi square assumptions
> 20 per group
No extreme %s
At least 5 in each box of table
