ANOVA 2

Question 1

Which tests do you use after ANOVA after P-value is significant so that there is evidence that at least one group has a mean different to another group, but you want to know which groups are different?

Answer 1

DO NOT do multiple t-test= Will increase probability of making a type I error= 0.95^3= 0.857

Dunnett’s Test: Have a control and one or more treatments- Want to compare each treatment to the control

Tukey’s Test: When none of the groups can be a control, so want to look at all possible pairwise comparisons

Question 2

What is an example of how to layout the explanation of results of ANOVA?

Answer 2

1) There is/no evidence that the mean…..
2) (ANOVA; F2degrees of freedom= , P= )
3) Then state the means of each group
4) Then talk about results from Dunnet/ Tukey’s test: (Test; Each combination of groups, P= )

Question 3

What are the assumptions needed for the ANOVA analysis to be valid?

Answer 3

1) The variation about the group means is NORMALLY DISTRIBUTED

2) The variation about the group means is the SAME for each group
Check graphically, check that standard deviations of each group are similar
If you’ve got one group that has very little variation and another with large= Test is not very reliable

3) The data are independent
IF NOT: Perform a repeated measures ANOVA- improves ability to detect true effects + consistent differences from the mean for repeatedly measured subjects

Question 4

Which test can be used to see if data is consistent with a normal distribution?

Answer 4

The Kolgomorov-Smirnov test

Null hypothesis: Distribution of points is not different from the normal distribution

If reject= Not a normal distribution

Question 5

How do you transform non-normally distributed data?

Answer 5

Positively skewed= Taking the natural log
y= ln (y)

Transformed data= More symmetric about the mean

Question 6

Which test do you use when wanting to see where the differences lie in NOT INDEPENDENT data?

Answer 6

Bonferroni Test

Question 7

What is the difference between parametric and non-parametric tests?

Answer 7

Parametric: Normally distributed or is easily parameterise normal distribution (only need to know its mean and variance) and equal variances

Non-parametric: Not consistent with normal distribution and/or unequal variances

Parametric: T-test
Non-parametric: Mann-Whitney U Test

Parametric: ANOVA
Non-parametric: Kruskal-Wallis Test

Parametric: RM ANOVA
Non-parametric: Friedman’s Test

Question 8

What does the Kruskal-Wallis test compare?

How do you carry out the test?

Answer 8

Compare medians of 2 or more unrelated groups

Very similar to Mann-Whitney test but lets you compare among more than 2 groups

Null hypothesis: All groups have the same median
Alternative hypothesis: At least one group’s median differs

1) Calculate ranks
2) Sum ranks for each of the M groups
3) Calculate the test statistic (H)
4) Compare H with Chi-square distribution with M-1 degrees of freedom