Hello!
I am ding a research where I have 3 DVs(3 domains of a scale) and 3 levels of the IV(high,low,moderate). I am not very familiar with complex statistics. Can anyone suggest if MANOVA is suitable for this? and my sample is very small only 20. So non parametric equivalent to that is what i am concerned about. Also if consider variables like age and gender... can i consider them as covariates?
Anyone please help me with this! Thank you!
If your dependent variables are three sub-scales that are in running numbers (continuous form) and if your independent variable is in categorical or ordinal form (three levels, as you mentioned), you could opt for three Kruskal-Wallis tests in this manner:
IV (with three levels) with first sub scale.
IV (with three levels) with second sub scale.
IV (with three levels) with third sub scale.
I've carried out this type of test and have interpreted the results in my paper on the wellbeing of palliative care workers in India-
Article Wellbeing of Palliative Care Workers During Covid-19 Pandemi...
Additionally, I have a YouTube video on this as well. https://www.youtube.com/channel/UCvu4Q2E9aMBdkBvGGu-pWwg
Hello Mahima Das. After reading your question, I have several questions of my own. The italicized sections below are quoted from your question.
I am ding a research where I have 3 DVs(3 domains of a scale) and 3 levels of the IV(high,low,moderate).
How were the 3 categories determined? I am wondering if you used two cut-points to carve a quantitative variable of some kind into categories. If so, it is generally preferable to treat such variables as quantitative explanatory variables. Carving into categories typically results in loss of power, introduction of artificial step functions into the fitted values, and other problems. For more info, see:
- https://discourse.datamethods.org/t/categorizing-continuous-variables/3402
I am not very familiar with complex statistics. Can anyone suggest if MANOVA is suitable for this? and my sample is very small only 20. So non parametric equivalent to that is what i am concerned about.
I wonder if you are suggesting here that only large samples can be analyzed using parametric methods, and that small samples must be analyzed with (so-called) non-parametric methods. If so, take a look at this BMJ Statistics Note:
- https://www.bmj.com/content/338/bmj.a3166.long
See especially the final paragraph.
Also if consider variables like age and gender... can i consider them as covariates?
Even if you felt justified in estimating a parametric model, with n = 20, you could only use up one degree of freedom for explanatory variables, at least if you wanted to adhere to Frank Harrell's 20:1 rule of thumb (to reduce the likelihood of over-fitting). Look for 20:1 in this Author Checklist:
- https://discourse.datamethods.org/t/author-checklist/3407
Finally, whenever I see someone contemplating MAN(C)OVA, I always wonder if they have read the classic article by Huberty & Morris (1989) and the more recent article by Huang (2020).
Article Multivariate Analysis Versus Multiple Univariate Analyses
Article MANOVA: A Procedure Whose Time Has Passed?
If you have not read them, I urge you to do so. Pay attention especially to the question H&M pose in the former:
Q. Do you have one multivariate question? Or do you really have 3 univariate questions?
HTH.
Excellent answer Bruce Weaver !!
What concerned me:"I am not very familiar with complex statistics."
So, why are you doing it all on your own? I mean, there is nothing wrong with trying something out, but if you want to analyze and publish results from an actual study, there is so much that can go wrong and real data are seldom like textbook data. There are lots of pitfalls and things to consider. Therefore, if you analyze real data, I would suggest to ask a local statistician to help with it.
- Badges
- Science topic
- Similar topics
- Publishing