Lecture 4 Flashcards
ANOVA II
Video 1
What does a residual show?
The difference between the observation and the prediction
Video 2
What do you do when doing an Omnibus test?
Checking whether all means are equal.
Video 2
When are the group effects significant?
When the groups are not equal in means. The residuals are smaller if you account for a specific group membership.
Video 2
Can you convert a two-way ANOVA into regression?
Yes you can, but it is very hard
Video 3
What is an independent observation?
An observation done within or between groups (is representative of the population and do not have a small N)
Video 3
What is longitudinal data and which tests do you use for the statistics?
Longitudinal data is dependent. When there are two time points you use a paired t test. When there are more than two time points a repeated measure ANOVA is used.
Video 3
When do you used a paired t test?
When the pre-/post test scores are both from the same person
Video 3
When do you use a repeated ANOVA?
When you want to see the change within each group
Video 3
When do you use a mixed model?
When you need to handle missing or non-equidistant data (not the same time between measurements), or when there is variation between the subjects starting point and growth curve.
Video 3
When do you get longitudinal data?
When a subject is measured more than once.
Video 4
When do you get clustered data?
When the subjects are eg from the same family. When it’s not corrected for you have high risk of a type 1 error.
Video 5
What is an intracluster correlation (ICC)?
When there is eg similarity between the neurons from the same pup (it ranges from 0-1, where 1 means that all things from the same cluster give the same info)
Video 5
What is the effective sample size?
The truly unique sources of info
Video 5
What formula is used for the effective sample size?
Neff = Ntotal / (1+(n-1)ICC)
n= the sample size within each cluster
Video 5
What does sigma e stand for?
The variation within each cluster
Video 5
What does sigma u stand for?
The variation between the clusters
Video 5
What is the formula used to calculate the ICC
sigma u ^2/ (sigma u^2 + sigma e^2)
Video 5
What does it mean when the ICC is high?
The observations within the same cluster are similar
Video 5
What is the consequence of not using Neff?
The SE is underestimated, df is too large, or the p-value is too small
Video 6
What can be done to try to fix the dependency?
Take the average within the cluster (a pooled t), or take a random observation from each cluster (but you loose info/power), or you can do a multi level analysis
Video 6
What is a cluster specific formula?
Yij = boj + biGi + Eij and boj = bo + Uoj (the cluster deviation mean)
Notes
What do b0 and b1 mean when you use the t-test as a regression model?
b0 = specific data eg of the mean of female.
b1 = difference between eg male and female
Notes
When is a Lavene test used?
To test the equal variance in a group.
