Chapter 14: Repeated-Measures Flashcards
Repeated Measures Design (GLM4)
- when the same entities participate in all conditions of an experiment or provide data at multiple time points
What assumption is violated in Repeated Measures Design?
- assumption that scores in different conditions are independent is violated
- scores taken under different experimental conditions are likely to be related because they come from the same entities
—> As such, conventional F-test will lack accuracy
What is the new assumption in Repeated Measures Design?
- new assumption: the relationship between pairs of experimental conditions is similar [sphericity]
Mixed Design: Sphericity Assumption
- can be linked to homogeneity of variance in between-groups ANOVA
- denoted by ε
- sometimes referred to as circularity
- general condition (less restrictive form) of compound symmetry
- assume that the variation within experimental conditions is fairly similar and that no two conditions are any more dependent than any other two
- compound symmetry not a necessary condition
- if you were to take each pair of treatment levels, and calculate the differences between each pair of scores, then it is necessary that these differences have approximately equal variances
- at least 3 conditions for sphericity to be an issue
Example:
Condition A
Condition B
Condition C
Variance (A-B)~Variance (B-C)~Variance (A-C)
Mixed Design: Sphericity Assumption
- Assessing the severity of departures from sphericity
- use Mauchly’s test
Mauchly’s Test
- tests hypothesis that variances of differences between conditions are equal
- if Mauchly’s test statistic is significant (p
What is the effect of violating the assumption of sphericity?
- sphericity creates a loss of power and a test statistic that doesn’t have the distribution that it’s supposed to have (F- distribution)
- complications for post hoc tests
—> Bonferroni method: most robust of the univariate techniques, especially in terms of power & control of Type I error
—> When sphericity is definitely not violated, Tukey’s test can be used
What do you do if you violate sphericity?
- adjust the degrees of freedom for any F-ratios affected by violation
- if data spherical, sphericity: 1
- if data non-spherical, sphericity: less than 1
- multiply df with sphericity estimate
- Smaller degrees of freedom make p- associated with F-ratio less significant
- by adjusting df, we make F-ratio more conservative when sphericity is violated
What to use to adjust the degrees of freedom when sphericity is violated?
- Greenhouse–Geisser estimate varies between 1/(k − 1) and 1
- denoted by epsilon with cap
- k: no. of Repeated measures conditions
- lower bound = 1/(k-1)
- when Greenhouse-Geisser estimate is greater than .75 the correction is too conservative
- when estimates of sphericity are greater than .75, Huynh–Feldt estimate should be used
- when Greenhouse–Geisser estimate of sphericity is less than .75 or nothing is known about sphericity at all Greenhouse–Geisser correction should be used
Other option when sphericity is violated
- multivariate test: MANOVA (not dependent on sphericity)
- tradeoff in power when using multivariate tests
- complex way: multilevel model
Power in ANOVA and MANOVA:
- when you have a large violation of sphericity (ε < .7) and your sample size is greater than a + 10 then multivariate procedures are more powerful
- a: no. of levels in repeated measures
- with small sample sizes or when sphericity holds (ε >.7) the univariate approach is preferred
Theory of One-Way repeated-measures ANOVA
- within participant
- within-participant variance will be made up of two things: effect of manipulation & individual differences in performance (SSR)
- use an F-ratio that compares size of variation due to our experimental manipulations to size of variation due to random factors
Repeated Measures: SST
- same formula as in independent ANOVA
- SST=grand variance x (N-1)
- dfT= N-1
- Treat data as a single group
Repeated Measures:
- Within-participant Sum of Squares [SSW]
- represents individual differences within participants
- SSW=s^2 (entity1) x (n1-1)+…s^2 (entity i) x (ni-1)
—> n: number of repeated conditions
- dfW= N x (n-1)
Repeated Measures:
- The model sum of squares [SSM]
- same formula as independent ANOVA
- SSM=sum[nk(condition mean - grand mean)^2]
- dfM=k-1
- k=no. of conditions
Repeated Measures:
- Residual sum of squares [SSR]
~ SSR= SSW- SSM
~ dfR= dfW- dfM
Repeated Measures:
- Mean Squares
- eliminate bias
- MSM: average amount of variation explained by model (systematic variation)
—> MSM= SSM/dfM - MSR: average amount of variation explained by extraneous variables (unsystematic variation)
—> MSR= SSR/dfR
Repeated Measures:
- F-ratio
- F=MSM/MSR
- F>1
- F(dfM, dfR)
Repeated Measures:
- Between-participants sum of squares [SSB]
- represents individual differences between cases
—> e.g: different participants have different tolerances - SST = SSB + SSW
- SSB = SST - SSW
Repeated Measures: Assumptions
- Linearity
- Normality
- Independence of Errors
- Sphericity
- if violated: one-way repeated ANOVA: use Friedman’s ANOVA
- no non-parametric version for factorial repeated measures
- follow ch.5 for correcting for bias
One-way repeated-measures ANOVA: SPSS
I. Analyze
II. General Linear Model
III. Repeated Measures
- Within Subject Factor Name: (no spaces in the name)
- Number of levels
- Click Add
- Click Define
- Transfer levels to ‘Within Subjects Variables’
One-way repeated-measures ANOVA: SPSS
- Defining Contrasts
- not possible to define planned contrasts for repeated measures
- use standard contrasts
One-way repeated-measures ANOVA: SPSS
- Post-Hoc tests
- When sphericity is definitely not violated: Tukey’s test can be used
- if sphericity can’t be assumed: use Games–Howell procedure
- sphericity-related complications: post hoc tests seen for independent ANOVA not available for repeated-measures analyses
- Some Post-Hoc tests can be done by Clicking ‘Options’
One-way repeated-measures ANOVA: SPSS
- Post-Hoc tests using ‘Options’
- Click ‘Options’
- Drag repeated-Measures variable to ‘Display Means for’
- Compare main effects
- Confidence adjustment: Use Bonferroni or Sidák