Stats 2 Third Exam Flashcards
What does Repeated Measures ANOVA / “Within Subjects Design” do?
Test to see if responses change over time OR Test significance between multiple simultaneous responses on the same item
What is most common within subjects design?
Pre-Post Design: O X O
Pre-post statistical test sometimes called
“Dependent t test” This is inaccurate.
Called “Dependent Test of Means”
What are the issues with within subjects designs?
- Same people responding – impacts our measurements
- Individuality differences
- Start separating out error terms
What comprises a Repeated Measures ANOVA Model?
- 1 fixed categorical (dependent) IV with 2 levels that will be repeated
- 1 continuous random DV measured multiple times
What are assumptions of Repeated Measures designs?
- R - random selection (not random assignment)
- N - Normality of error term
- H - Homogeneity of variance
- D - Dependence (not independence)
What are potential methods of analysis for within subjects designs?
- Multiple dependent t tests
- Gain scores
- ANCOVA
- Univariate repeated measures
- Adjusted univariate repeated measures
- Multivariate repeated measures
Defining features of gain scores approach to analysis in within subjects design?
- Frequently used in pre-post
- Issues:
a. Reliability issues
b. Alpha issues
c. However, Observed score is true score + error
d. Assumes same reliability across measurements
Defining features of ANCOVA approach to analysis in within subjects design?
- No test of time
2. Violates assumption of independence (it’s the very thing we want to test)
Defining features of Univariate repeated measures approach to analysis in within subjects design?
Requires new assumption: Sphericity (symmetry)
What is assumption of sphericity?
- Deals with correlations among variables
2. Assumes correlation matrix will have equivalent values within it
Defining features of Adjusted univariate repeated measures approach to analysis in within subjects design?
- Adjusts for symmetry (through degrees of freedom)
2. Employs multiple tests (experimentwise alpha considerations)
Defining features of Multivariate repeated measures approach to analysis in within subjects design?
- Allows for non-symmetry
- Controls alpha
- One test
What are implications of testing the assumption of syphericity (Mauchly test)
- If meets symmetry, univariate approach
2. If does not, multivariate approach
What are three types of repeated measures?
- Time
- Response mode
- Persons unit (e.g., family)
What is a twice repeated design?
1) Pre-post-follow up.
2) Unit pre-post
In repeated measures, what are implications of unequal n?
Cannot work with unequal n. Excluded participants with missed observations.
What are defining features of repeated measures error terms?
Every factor and every interaction has its own error term.
Class example of a 2x2 fully repeated design
Child-parent, pre-post
What are defining features of a mixed model design?
- 1 fixed categorical IV (between subjects)
- 1 fixed categorical repeated IV (within subjects)
- 1 continuous random DV
In an # x # mixed model design, 1st number represents
Non-repeated factor
In a # x # mixed model design, 2nd number represents
Repeated factor
In a # x # x # mixed model design, 2nd number represents
- 1st refers to non-repeated factor
- 3rd refers to repeated factor
- Nothing in our language denotes what middle number represents
What are assumptions of mixed model designs?
- Homogeneity
- Independence for some IVs
- Dependence for some IVs
Effect size for mixed model design?
Eta squared or partial eta squared
Eta squared
SSH/SST
Partial eta squared
SSH/SSH+SSE
What is the relation between eta squared and partial eta squared?
Partial eta squared gives higher value and thus overestimates strength of association.
Eta squared is a more conservative estimate.
What assumptions do you check in a mixed model design?
Homogeneity and sphericity first.
If sphericity assumed
Look at sphericity assumed
If mauchley test is significant
Look at multivariate or greenhouse geisser
What post hocs should one examine in a mixed model design?
ONLY for main effects.
When would one use a non-parametric test?
- Non-continuous DV
- Ordinal data
- Measurement reliability less than .60
Does using non-parametric tests resolve issues associated with a small sample?
No. Still requires power analysis.
What assumptions do non-parametric tests use?
R, I, D. Not N or H.
What are defining features of Kreska Wallace test?
An analysis of mean ranks.
What are defining features of Friedman test?
Used for pre-post-follow up or participant unit over time for non-parametric data.
When should one use an Independent t test?
Fixed categorical IV with 2 levels (can also use 1-way ANOVA)
When should one use a Dependent t test
Fixed categorical repeated IV with 2 levels
When should one use a 1-way ANOVA?
Fixed categorical IV with 3 or more levels
When should one use a 1-sample test of means?
Comparing sample to larger population
When should one use a 2-way ANOVA?
2 fixed categorical IVs
Number of hypotheses is determined by…
Number of IVs, not number of levels
When should one use a Helmert contrast?
To compare 1st group vs everything else
When should one use a Difference contrast?
To compare last group vs everything else
When should one use a simple contrast?
To conduct pairwise comparisons
When should one use a Polynomial contrast?
To test shape of lines and trends (e.g., linear, quadratic, cubic)
When should one use a Mixed Model design?
1 or more fixed categorical IV, repeated
1 or more fixed categorical IV, non-repeated
SSbetween =
Column mean - grand mean OR row mean - grand mean
MS =
SS/df
F =
MS/MS
