I am working on a policy capturing study and discovering hierarchical linear models. My model has 3 x 4-level factors and 2 x 2-level factors. There are 108 plausible scenarios. After considering D-efficiencies (Kuhfeld, 2010; Auspurg & Hinz, 2015), I am currently collecting data for 60 scenarios. They are blocked in 6 decks of 10 scenarios each, so each participant evaluates only 10 scenarios.
How do I determine an appropriate sample size?
Thank you.
You may want to look at the resources here under 'Power'
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#datadesign
Diana, there was another thread on this topic on ResearchGate:
https://www.researchgate.net/post/How_can_I_calculate_power_for_multi-level_analysis
There you will find some useful references.
Regards, Karin
I am trying to figure out the number of levels for my HLM model. I am working my way through the modules in LEMMA and find them very instructive.
Am I correct that the number of levels has an effect on my optimal sample size (e.g., a two-level vs. three-level model)?
Thank you.
Thanks, Tae-Yeol.
If you have a reference for determining level size, it would be very helpful.
Hi Diana,
Have a look at this book. It is about sample size requirements for HLM models and have some good, clear examples.
http://www.ierinstitute.org/fileadmin/Documents/IERI_Monograph/IERI_Special_Issue_1.pdf
I hope it helps!
Andres
That looks very useful.
The MLpowsim software allows you to undertake these 'experiments' for your particular problem and is not limited to two level models. The manual that goes with it is really helpful in talking through the issues.
http://www.bristol.ac.uk/cmm/software/mlpowsim/
Thanks Andres and Kelvyn for sharing these resources.
For now, I believe that at least 2 different models are required for my analysis. Each respondent evaluates 10 scenarios with 3 questions per scenario (my dependent variable is measured by 3 items). To simplify the discussion, I will assume that there is one response per scenario. I have 5 independent variables, 3 of which moderate the 2 main hypotheses.
Model 1: A 2-way cross-classified model of the 10 responses of each respondent.
Model 2: A 4-level multilevel model with the 10 responses of each respondent at the lowest level.
Currently I have 100 respondents, i.e., 1,000 responses based on the above assumption.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size