Please could someone help me to calculate the sample size I would need for Moderation Analysis (Hayes' model 2) - I am interested in considering how attachment style (avoidant vs anxious) moderates the relationship between cognitive behavioural therapy and outcome measure (PHQ-9).
How do I work out sample size using G-Power?
Independant variable: Mode of delivery (online versus remote delivery)
Moderation variables: 1) Anxious attachment style and 2) Avoidant attachment style
Dependant variable: Score on Patient Health Questionnaire
There are no co-variates.
A similar question has been posted here:
https://stats.stackexchange.com/questions/10137/power-analysis-for-moderator-effect-in-regression-with-two-continuous-predictors
The answers make good sense to me, except that with regard to G*Power, I would choose a larger value for power (> .8) and also examine really small effect sizes for the interaction to be on the safe side (interaction effects in moderated regression are often tiny and therefore difficult to detect--you may need a rather large sample for sufficient power).
Lindsay Young To compute the sample size using GPower for moderation (Hayes' Model 2), use the following values:
1. Select "Multiple regression: Moderation (Model 2)" as the statistical test.
2. Effect size: Determine the proper effect size based on the study question to detect the smallest relevant difference in the moderating effect.
3. The alpha level is often set to 0.05, implying that you wish to keep the Type I error rate (the chance of rejecting the null hypothesis when it is really true) to 5%.
4. Power (beta error probability): Typically, the power is set to 0.8, indicating that you wish to have an 80% chance of detecting the stated effect size if it occurs.
5. The number of independent variables in this example is three (mode of delivery, anxious attachment style, and avoidant attachment style).
6. In this scenario, there are two moderators (anxious and avoidant attachment styles).
7. Connection between independent variables: Typically, the default value is set to 0, indicating that there is no correlation between the independent variables.
Once these parameters are set, GPower will calculate the sample size required to produce the target power with the selected alpha and effect size. It is crucial to remember that sample size estimates are based on assumptions and do not guarantee the intended findings, and additional considerations such as measurement error, data variability, and demographic heterogeneity should be addressed while organizing the study.
I recommend the attachment here, the paper is available for download. The sample size calculations would be the same as any other regression. G*power should work for you. Best wishes David Booth
- Badges
- Science topic
- Similar topics
- Statistics
- Survey