I'd be ever so grateful for anyone who can help guide me with which statistical method is most appropriate for my data.
Our Variables
Age (continuous)
Food rejection (continuous)
4 x Performance score (continuous 0-1)
1. condition A
2. condition B
3. condition C
4. condition D
Our hypothesis is that age and food rejection predict performance scores.
The independent variables(age and food rejection) are both continuous.
The dependent variables(score A, score B, score C, score D) are continuous, but highly collinear within each child.
Should I run a mixed effects model, with condition, age, and food rejection as fixed effects, and subject as a random effect?
Or is there a more appropriate test?
Are there 4 sets of measurements (Performance scores) for each participant? Are there separate food rejection variables for each of these 4 time points? In other words, is this a repeated measures/longitudinal design? If so, mixed effects can be used for this, and offer some advantages over traditional designs. THis will be covered in most longitudinal methods courses, so just go through the materials for that at your university. If the design is something else please be more specific about this. It is often useful to write out the equations for the models that you wish estimated.
Thank you for the additional information, it is incredibly helpful.
Daniel Wright The food rejection variable remains constant for all four condition scores. Each participant has four scores, but the four scores refer to 4 types of knowledge (a. vegetables, b. fruits, c.meat, d.sweets). I thought of running linear regressions multiple times then comparing them, but I think this means I would inflate the familywise error rate. That's why I considered a mixed-effects approach, with a model:
Score = Age + Food Rejection + Condition
Sorry if that doesn't make sense, I can't seem to get my head around what makes sense in terms of my hypothesis.
Rani P Ramachandran
am I right in thinking that since I have four scores as outcome variables (each condition score) I should apply a Multivariate approach/MANOVA instead of mixed-effects model?
Abigail Pickard , you would not do four separate ones for among other reasons you likely want to compare the food types. If I understand then, it is four food types (presumably counterbalanced for time). For the equation you list you would add food type and presumably some interactions (and depending on things like how much variation there is with age decide whether it is linear ... similar with food rejection, but I don't know how that is measured). You could treat this as multivariate or mixed. The mixed software packages (e.g., lme4 in R) are probably more flexible. As well as havng a random interccept, you could also have, for example, the differences in food type effects to vary randomly. You could include this just with more dummy variable (the product of the person dummy variable and 3 of the food type dummy variables) but this would mean a lot is being calculated (and several people argue for the random effects approach here anyway).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis