Method: N/A
Analysis/Results: N/A
Conclusions: N/A
We will discuss the implications of using an inappropriate experimental unit with a concrete example (Iserbyt et al., manuscript) and provide a tutorial on how to use SAS and SPSS to conduct correct statistical analysis with intact classes as an experimental unit.
The Iserbyt et al. study involved eight intact classes: 4 classes assigned to a control condition and 4 classes to an intervention condition. For each class, eight students (male and female) were randomly selected based on skill levels (2 high, 4 medium, and 2 low). The dependent variable was percentage of correct trials that students performed during a 6-day badminton unit. Using SAS and SPSS, ANOVAs with intact classes or students as an experimental unit were conducted to analyze the data. With students as an experimental unit, the statistical model has eight factors: students, conditions, skill levels, gender, three two-way interactions between conditions, skill levels, and gender, and a three-way interaction. With intact classes as an experimental unit, the statistical model has nine factors: students nested within intact classes, intact classes nested within conditions, conditions, skill levels, gender, three two-way interactions, and a three-way interaction. The ANVOA tables, statistical results, and effect sizes from these two models were compared to demonstrate the implications using inappropriate experimental unit.
The results showed that degree freedoms for the error term for the models with students or intact classes as an experimental unit were 49 and 55, respectively. R-squares for both models were 79%. Both statistical models showed significant main effects of conditions, skill levels, and gender, and a significant interaction between skill levels and gender. The effect sizes for conditions, skill levels, gender, and a skill-by-gender interaction were .59, .52, .15, and .19 for intact classes and .33, .43, .10, and .13 for students, respectively. Both models showed a similar pattern of results, however, there were significant reductions of effect sizes when using students as an experimental unit. Using students as an experimental unit also violated design principle since the interventions were on intact classes nested within conditions. It is suggested that an appropriate experimental unit of intact classes with a nested design be used.
Our tutorial will show how to enter the data using an excel, import them into SAS and SPSS, conduct data analysis using SAS and SPSS, interpret the statistical results, calculate effect sizes, and conduct a power analysis.
See more of: Research