I'm working on my dissertation proposal and I have to pre-determine the type of statistical analysis I'm going to run. I have 2 IVs (traditional gaming group vs. virtual gaming group), and I am testing for team dynamics, critical thinking/problem solving, and politeness (3 DVs).
My sample size is 7 per IV (replication-extension case study), so I've been told a MANOVA may not yield appropriate results with a small sample size. It was recommended I use a non-parametric test, instead. The only one I've found is the Kruskal-Wallis test, but that doesn't seem to fit what I'm doing.
Would it be better to consider running paired-t for each of the DVs? The only thing I see as a problem with that is politeness could affect the other two DVs, as well (which is why I was thinking MANOVA).
Help? :D
Kruskal- Wallis is for ordinal scale. There is no separate non-parametric regression analysis so you can perform a normal linear regression analysis for your data. Ex:
A MANOVA is appropriate in the (rare) circumstances where you have three measures of the same thing. I think that you have three possible effects of your IV (you only have one – or did you forget to include the other?).
Your sample size is terribly, terribly small. Why so small?
If you have several gaming groups within each level of the IV, then you also need to adjust for clustering within each group.
Your DVs are going to fall onto scales that have constrained lower and upper bounds, which can be problematic.
What about ordinal logistic regression? This allows you to handle your IV and potential confounders.
But I'm really concerned that your sample size is so small. Have you done a power calculation at all?
Your small sample size is not encouraging for parametric or even nonparametric, you have to think carefully to increase your sample size (if possible) to get sensible statistical decisions.
The dissertation is a replication and extension study. The original study used 14 people, of which 7 consent to structued interviews.
Rather than interviews, i’m Collecting survey data after each gaming session, with a comprehensive survey after the final session. Data will be likert scale and coded open ended questions.
My IV has two levels: Virtual and Physical. The DVs can be measured individually, however, one may influence another, which is part of my concern. This is a case study, so i’m Not necessarily applying it to the greater population; I’m laying a little ground work for some future studies.
I still don't understand your sample size. What is the size of the effect(s) that you are looking for? What power have you to detect them?
There's no point in carrying out a study when there is a significant risk that you don't find the effect even though it exists.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size