I am trying to determine which would be the best model type to run on my data. After reading through this chapter's introduction, it seems I would be best suited to run a hierarchical regression on a multivariate linear mixed model. However, this seems like it could be difficult to interpret and am not sure that it is utilized approach throughout the literature.
Concerning my data, I have various demographic and cultural variables (IVs; let's say around 8). I have multiple categorical questions (DVs; let's say around 20). The DVs are not all regarding the same topic in the study, rather split over three topics. So, I am looking at them separately for each question (as they are categorical). I have a large dataset in which each line is a new participant (no repeated measures; ie. first 'grouping' level). I grouped the participants into 50 groups, based on location (ie. second grouping level). Given that 50 groups are too many, I grouped the second-level groups into 3 broad groups based on general location (ie. third grouping level).
I am planning to run the analysis on each DV and add in the demographics/cultural IVs in a hierarchical-fashion. I was thinking to treat the second- and third-grouping levels (specific location and broad location) as random effects. I also had IVs that were computed based on the third level groups (ie. a culture variable between broad location groups), which I would include in the regression.
These are the models that I am considering (only a linear mixed model, not multivariate):
model1a
If your dependent variable is categorial, you need to use the generalized linear mixed-effect model (GLMM). The R function "glmmPQL" under the " nlme" package is based on an approximate method for GLMMs.
However, if the dependent variable is continuous, you can use the following for three-level modelling:
model1a
Md. Belal Hossain Many thanks for your answer! Can I ask whether it is common to hierarchically add the IVs into the model when you include random effects of the second and third-level groups?
Secondly, I have categorical DVs. Can I ask why lme() from the nlme package would not suffice?
1. Yes, you can add the IVs to the model. This is a common approach to adjust for confounding bias/getting adjusted estimates.
2. Continuous DV - linear model (LM), linear mixed-effect model (LME); Categorial/Count DV - generalized linear model (GLM), generalized linear mixed-effect model (GLMM).
Note: These are just examples. There are other ways to model a continuous or a categorical DV.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running