PS598 Unit 9 Data Analysis Plan Example

One Test per Prediction: A PS598 Unit 9 Data Analysis Plan for a Self-Compassion Writing Experiment Before Anatomy Exams

[Student Name]

Purdue University Global

PS598: Master's Capstone in Psychology

Unit 9 Data Analysis Plan

[Instructor Name]

[Date]

The writer, the college and the planned study are composite and illustrative; the sources cited are real.

Data and Variables

The data set will contain one row per participant with the following variables: writing condition (coded 0 for neutral and 1 for self-compassion), baseline self-compassion (Self-Compassion Scale Short Form total), cognitive test anxiety (Cognitive Test Anxiety Scale total), exam score (percentage), term (fall or spring), section and an engagement code from the writing sample. The planned sample is about 260 complete cases.

Preliminary Analyses

Before testing hypotheses, I will check that randomization worked by comparing conditions on baseline self-compassion, term and section with t tests and chi-square tests. I will report means, standard deviations and correlations for all continuous variables and internal consistency for both scales in this sample. A baseline difference between conditions would not invalidate random assignment but would be reported and included as a covariate.

Hypothesis 1: Cognitive Test Anxiety

H1 predicts lower cognitive test anxiety in the self-compassion condition. I will test it with an analysis of covariance, with condition as the factor, cognitive test anxiety as the outcome and baseline self-compassion and term as covariates. Including baseline self-compassion increases precision by accounting for variance unrelated to condition. The test of the condition effect is one-tailed in direction but will be reported two-tailed to be conservative. I will report the adjusted mean difference with its 95 percent confidence interval and Cohen's d.

Hypothesis 2: Exam Score

H2 predicts higher exam scores in the self-compassion condition. The same analysis of covariance will be used with exam score as the outcome. Because exam difficulty may differ slightly between fall and spring, term is especially important here. I will also report the result without covariates, so readers can see the raw difference.

Hypothesis 3: Moderation

H3 predicts that the reduction in cognitive test anxiety will be larger for students low in baseline self-compassion. Hayes (2018) describes moderation as a question of whether the effect of one variable on an outcome depends on a third, tested by including the product of the predictor and moderator in a regression. I will regress cognitive test anxiety on condition, mean-centered baseline self-compassion, their product and term. A significant negative interaction term would support H3. Following Hayes, I will probe the interaction by estimating the condition effect at low, average and high values of self-compassion, set at the sample average and a standard deviation on either side of it, and plot the conditional effects.

A significant interaction will not be the end of the story; the plan shows exactly where along the self-compassion scale the exercise helps.

Assumptions

Analysis of covariance assumes that residuals are approximately normal, that variances are equal across groups and that the relationship between covariates and the outcome is the same in both groups. I will examine residual plots and Levene's test and test the condition by covariate interaction for homogeneity of regression slopes. If slopes differ, the moderation analysis will be treated as primary for that covariate. If residuals are badly skewed, I will report robust standard errors.

Effect Sizes and Power

Cohen (1992) argued that effect sizes, not only significance, should guide interpretation, and the power analysis in Unit 7 was based on an expected d of 0.35 using G*Power (Faul et al., 2007). I will report d for both main comparisons and the change in explained variance for the interaction. A sensitivity analysis in G*Power will show the smallest effect the final sample could detect with 80 percent power.

Missing Data

Some students will consent but miss the exam or skip the post-exam scale. I will report attrition by condition. If less than 5 percent of data are missing, I will use complete cases. If more, I will use multiple imputation with condition, baseline self-compassion and term as predictors, and report both sets of results.

Engagement Analysis

Writing samples will be coded for engagement with the prompt. The main analyses will include all randomized students, following an intention-to-treat approach. A secondary analysis will repeat the tests among students coded as engaged, using a coding guide with two trained raters and a check of their agreement, with the caution that engaged students are a self-selected subgroup.

Alpha and Multiple Tests

Each of the three hypotheses will be tested at alpha .05. Because they are separate predictions stated in advance, I will not correct for multiple comparisons among them, but exploratory analyses, such as differences by section, will be labeled as exploratory.

Conclusion

The plan matches each hypothesis to a test: analysis of covariance for cognitive test anxiety and exam scores, and regression with an interaction term, probed as Hayes recommends, for moderation by baseline self-compassion. Assumption checks, effect sizes guided by Cohen, a sensitivity analysis in G*Power, clear missing-data rules and an intention-to-treat approach complete the plan.

References

Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155-159. https://doi.org/10.1037/0033-2909.112.1.155

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175-191. https://doi.org/10.3758/BF03193146

Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (2nd ed.). Guilford Press.