Dear colleagues, I am in process of preparing the research methodology of the doctoral disertation. I formulated hypotheses but now I am not sure how to test some of them, specifically which statistical test to use.
Here I am going to post my hypotheses and variable in hope that someone can point me in the right direction.
H1. Students whose parents have a high level of civic competence have a high level of civic competence regardless of the type of school they attend.
Variables: student civic competence (ordinal scale); parents level of civic competence (ordinal scale); type of school (nominal scale)
H2. Students from schools where teachers have a high level of civic competence will have a high level of civic competence regardless of the type of school students attend.
Variables: student civic competence (ordinal scale); teachers level of civic competence (ordinal scale); type of school (nominal scale)
H3. In schools where the achievement of the inter-subject theme Civic Education is given high importance, students will have a high level of civic competence regardless of the type of school
Variables: achievement of the inter-subject theme Civic Education (ordinal); student civic competence (ordinal scale); type of school (nominal scale).
I wanna point out that I am not seeking quick solutions and that I am really stuck on this. Any kind of help and/or suggestion is welcome.
Thanks!
Hello Ivan,
H1: You're looking for two things: (a) a positive relationship between the IV of parent civic competence with the DV of student civic confidence; AND (b) no interaction of the IV with school type. I'd suggest ordinal regression.
H2: Similar to H1, except that the "IV" is now teachers' levels of civic competence (for the school attended by a student). While you may be able to use ordinal regression, this is now a scenario that might better be handled as a multilevel design (as I expect you'll have multiple students within a given school).
H3: Similar to H2/H1, except the the "IV" is now the level of importance for civic education. Again, it looks to me as if you might be better served considering a multilevel model.
Finally, it is also possible that all three hypotheses could be addressed within a single multilevel model, rather than looking at one target IV at a time.
Good luck with your work.
David Morse thank you for an answer. I will look into a multilevel level design.
When you say you have an "ordinal scale," does that mean that you have several separate ordinal variables that could be combined into a scale? If so, you might look at this resource:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
David L Morgan Yes, but that is not the issue that I am trying to solve. My problem is how to test inference between this composite ordinal variable and other ordinal and nominal variable. After combining them I will still have composite ordinal, not scale variable.
Combining multiple ordinal variables can indeed create an interval-level variable, and that will certainly affect the kinds of statistical tests you can do.
You have a point but the instrument I am using is standardized and its final result is indeed on an ordinal scale.
First of all, I would recommend that you later on (in your actual work) better state your hypotheses as null ones. Second, since you have one ordinal dependent variable (i.e., student civic competence) and several ordinal and nominal independent variables (i.e., parents’ civic competence, type of school, teachers’ civic competence, and civic education), you could use the ordered logit model (i.e., Ordinal Regression). You might see Williams and Quiroz (2020) for illustrations.
Williams, R. A. & Quiroz, C. (2020). Ordinal regression models. In P. Atkinson, S. Delamont, A. Cernat, J. W. Sakshaug & R. A. Williams (Eds.), SAGE research methods foundations. https://www.doi.org/10.4135/9781526421036885901
Good luck,
To follow up on @David Morse excellent suggestion I would recommend your overall approach consider ordinal logistic regression because your DV is apparently ordinal. There are many references you can find by a Google search. Best wishes David Booth
@Ivan BTW your response to @David L Morgan's initial post should , I believe, be rethought. Consider in light of the regression models that most posts have suggested. Combining scales is not what was suggested. Study the use of linear regression models.Many such texts can be found in your favorite library or online. Best wishes David Booth. PS I particularly like Kutner et al Applied Linear Statistical Models which can be found in the z-library. Best wishes for a better day, David Booth
Thank you all for your insightful and usefull answer. Since the methods you are suggesting are new to me, before making any more comments I will get familiar with them. If latter I will have any more questions I will revisit this thread.
Once again, thank you all and have a nice day!
Ivan
Mohialdeen Alotumi I will take your advice and reformulate hypotheses as null ones.
So I started looking into MLM and it all seemed fine until I read this: "Partitioning the variance in an outcome (Y) into its within- and between-group components is an important first step in determining whether a multilevel analysis is warranted. If there were little or no variation in the outcome between higher-level units, there would be no warrant for a multilevel analysis" (Heck & Thomas, 2020, p. 36).
So before starting interpretation of data with MLMI have to calculate homogeneity of variance by an intraclass correlation (ICC), and then see if the usage of MLM warranted or not?
- Badges
- Science topic