Correlated Groups Designs: Repeated Measures, Matched Groups, and Mixed Model Designs

1. Slides to accompany Weathington, Cunningham & Pittenger (2010), Chapter 14: Correlated Groups Designs

2. Objectives • Correlated groups • Repeated-measures designs • Carryover effects • Matched-group designs • Mixed-model designs

3. Why Correlated-Groups? • Sometimes the treatment groups in a study are not independent • This is an assumption underlying between-groups designs

4. Correlated-Groups Logic • Minimizing within-group variance is a constant goal (increases our power) • Homogeneous groups can help • Subject variables can also be used • Correlated-groups is another good design strategy

5. Correlated-Groups Logic • Figure 14.1 • We can partition out variance due to individual differences • This reduces MSwithin and increases the F • Two options: • Repeated measures design • Matched-participants design

6. Figure 14.1

7. Repeated-Measures Design • Data collected over multiple conditions, using the same set of participants • Testing same person over multiple levels of the IV • Could be manipulated or could be time • Figure 14.2

8. Figure 14.2

9. Repeated-Measures Results • Summary table e.g., Table 14.3 • Notice that you are able to partition out an additional chunk of the variance in the DV • Compare with Table 14.2 • This reduces MSwithin • For this type of design the groups are not independent

10. Table 14.2 vs. Table 14.3

11. Repeated-Measures Pros/Cons • Advantages • Increased power • Each participant is his/her own control • Smaller required sample size • Disadvantages • Several forms of potential carryover effects • Table 14.4

12. Reducing Carryover Effects • Several design options exist • Use between-subjects design • Especially if experiment cause irreversible change • Use a special design modification: • Solomon four-group design • Counterbalancing • Latin square design

13. Solomon Four-Group Design • Incorporates 3 control groups to account for sequence-related events • Use a 2 x 2 factorial (b-g) ANOVA • Table 14.5 and Figure 14.4 illustrate

14. Counterbalancing • Requires random shuffling of the sequence of testing for each participant • Total # of possible arrangements = k! • Sample size must allow you to have enough people to adequately test each of the possible arrangements

15. Latin Square Design • Alternative to full counterbalancing • Ensures that • Each condition occurs once in each position of the sequence • Sequence is random • Table 14.7 • Complex analysis though…

16. Matched-Group Design • Retains power of repeated-measures, but tests each participant in only 1 condition • Used when an important subject variable correlates with the DV • Each condition is a separate group of participants • Matched across conditions to control for variance due to this subject variable

17. Matched-Groups Design • Table 14.8 shows steps • Ordered pairs of participants have similar scores on the pretest • A.K.A. randomized block design • Sometimes matching is done on several subject variables at once • Because groups are matched, they are not independent

18. Table 14.8

19. Matched-Groups Pros/Cons • Advantage • Potential for high power • Disadvantage • Potential for low power if matching technique fails to match on an important subject variable • Important = significantly linked to DV

20. Yoked-Control Group • Researcher randomly pairs control participant with active participant • Both participants experience exactly the same sequence of study procedures, except the control participant is not exposed to the IV • Figure 14.6

21. Figure 14.6

22. Mixed-Model Designs • Between- and within-subjects elements combined • Between-subjects: Experimental vs. Control • Within-subjects: Multiple trials • Somewhat more complicated in terms of design, but analysis is based on same principles as we have already discussed • This chapter’s Research in Action section provides a good illustration

23. What is Next? • **instructor to provide details