Bootstrapping:

Question 1

What is the main advantage of resampling approaches over parametric tests?

Answer 1

The underlying data distribution do not have to meet assumptions

Question 2

How does randomisation work?

Answer 2

1. Take the mean - test statistic
2. Pool data together
3. Take random samples of two groups with the original sample size and calculate the difference in means
4. Repeat this process many times
5. This builds the null distribution -> what you would expect by chance

Question 3

How does parametric bootstrapping work?

Answer 3

1. Take the difference in means between the two groups
2. Generate bootstrap samples by taking samples from the same groups and store the difference in means
3. Repeat this many times to create the parametric bootstrap distribution
   - If the CI does not include 0, you can be 95% confident that there is a real difference between the groups

Question 4

How do you do resampling with correlation coefficients?

Answer 4

Shuffle one of the variables so that it removes any association between the two variables, that way we can create a randomisation distribution.

Question 5

How do you do a randomisation test for correlation?

Answer 5

1. Calculate the observed correlation
2. Break the association between X and Y by randomly shuffling one
3. Calculate the correlation between the shuffled X and non-shuffled Y
4. Do this many times
   This is what you would get is Y and X were paired randomly