Hello everyone!
I've encountered a problem when trying to figure out the methodology of my dissertation. The study has multiple variables and will be, in general, rather complicated, but my only real problem pertains to the dependent variable.
In the study, the participants view short videos with emotional expressions, and are asked to evalaute to what extent they see 8 different emotions in a video, on a scale. In other words, for each stimulus they provide scores for 8 separate Likert scales (happiness, sadness, disgust, etc.). All participants are exposed to all faces, and as such the design is within-subjects. We want to see if different faces are perceived differently or if individual differences affect emotion perception. This way, we can see for example that people assign higher anger scores to male faces that female ones, or that gamers assign lower scores on dimension of happiness, thus they perceive faces to be less happy. So, variables such as gender or gaming experience would be independent, but what would be a dependent variable here? Is it eight different dependent variables, that is the eight emotion dimensions? Or is it one variable with eight levels?
I am at a loss as to how to approach this, and what analysis to use. Currently, I believe the right approach would be to treat this as eight separate dependent variables and use a MANOVA for the analysis. Would that be the right approach?
I would greatly appreciate any input, tips on where to look for an answer, etc. Thank you in advance :)
Hello Aleksandra,
Yes, sex of the face would be an IV (independent variable). Any other conditions that you varied systematically among the videos would also be IV(s). The DV is the rated presence of emotion, but you have 8 DVs. So, it's a multivariate study. The fact that the IVs would be within-subjects (aka, repeated measures) factors would make this design what is sometimes called a "doubly multivariate" design.
The chief question is whether the rated levels may confidently be treated as interval strength scores (or better), as opposed to ordinal scores. If interval strength, then a repeated-measures manova could be used, as could a multivariate regression model. If ordinal, then your choice is a non-parametric counterpart to a repeated-measures manova, or a multivariate ordinal regression.
You could run 8 separate analyses, each evaluating a single emotion rating, but doing so would mask the likelihood that the emotional ratings are likely inter-correlated.
I strongly suspect that your university has a number of folks on faculty who could further assist you in formalizing your analytic plan, especially after discussing your specific variables, quantification plan for each, and research questions.
Good luck with your project.
The semantic differential or the bipolar system of the scale should be considered. By so doing, you would understand the coding system for the scale afterwhich you can apply Spearman Rank to analyze it.
- Badges
- Science topic
- Similar topics
- Psychology
- Emotional Psychology
- Happiness