Week 5 Flashcards
What are two benefits of using repeated measures?
- Economy of subject numbers
2. Each subject acts as their own control
In a repeated measures design, each subject acts as their own control. What does this in turn do?
Reduces the error variance, making the test more sensitive.
ie., a smaller difference between means will be sufficient to produce a significant F ratio.
What are some disadvantages of repeated measure design
- order effects (due to learning or fatigue)
- differential carry over effects
How can we compensate for order effects in repeated measure designs?
We can counterbalance
What is counterbalancing? Give an example?
(A1, control) –> (A2, experimental)
and
(A2, experimental) —> (A1, control)
These two results will not be the same and can cause differential carry over effects.
What are differential carryover effects?
Occur when counterbalancing does not balance out order effects. One particular ordering, either treatment than control or the control then the treatment, creates a reaction to the DV for that ordering only.
what is an additional assumption in a within subjects design?
sphericity.
What is sphericity?
compound symmetry of the covariance matrix. In other words, homogeneity of variance within treatments, and homogeneity among treatments (variances of the differences between treatments).
When is the assumption of sphericity always met?
Always met when there are only 2 levels of within subjects IV (so long as you have homogeneity of variance)
When is the assumption of sphericity usually violated?
Usually violated when there are more than two levels.
What happens if the data doesn’t meet the assumption f sphericity?
The ANOVA becomes more generous at calling a result statistically significant.
Are there tests to check the assumption of sphericity?
Yes but they are unreliable so we don’t really use them
How do we correct for the “overly generous” results of a repeated measures ANOVA?
One way is to adjust the dof in the F ration by a correction factor (e, epsilon).
What are two methods for changing the epsilon when the assumption of sphericity is violated in a repeated measures ANOVA?
- Huynh Feldt
2. Greenhouse Geisser
What is the specific test you can do to see whether or not you need to make an epsilon correction or not?
The Mauchly test
What is a generalised eta squared and why would we use it?
measure of effect size
Can use in repeated measures as it takes sample into account
Why is the difference between eta squared and partial eta squared in repeated measures?
eta squared reflects IV as a proportion of the variability in the entire study
partial eta squared (in the case of one variable) reflects how much variation the repeated measures variable explains after we have separated the effects of the differences between participants
How do we calculate Cohen’s d?
Difference between means divided by some measure of pooled SD
What is an issue with counterbalancing when we have multiple levels?
We increase the amount of orders that we need to counter balance, which almost defeats the purpose of the benefit in repeated measures, economy of number.
e.g., 4 levels = 4X3X2X1 = 24 different possible combinations.
What is a more economical way to counterbalance?
using a latin square
Rather than using ALL possible orders, latin square designs use a _____ of possible orders.
_____
What are the two conditions of a latin square?
- each treatment occurs equally often in each position in the treatment order
- each treatment immediately precedes and immediately follows each treatment exactly equally often
most people just follow condition 1
What is a technological way to follow conditions in making a latin square?
a latin square generator
What type of puzzle are latin squares like?
sudoku
What is one assumption you need when doing latin squares?
sphericity - if you don’t have this then you probably shouldn’t do a latin square
If sphericity is violated, you need to adjust the ANOVA with an _____, otherwise, the analysis will be too liberal (generous).
Epsilon correction
Counterbalancing ______ over come differential carryover effects.
WILL NOT
What does a factorial design mean?
That there are two or more IV’s
What are fixed factors?
IVs where the levels of each factors are chosen by the experimenter, if replicated, these would stay the same
What are random factors?
Factors in which the levels if the IV are randomly selected from a larger population of potential levels. They would differ from one replication to the next.
What is the “practical upshot” of having a random factor in an analysis?
it changes the way we calculate the ‘error term’ in the ANOVA (for each effect and interaction)