Relative impacts of interval and categorical factors on ordinal/interval DVs in an emotion regulation study?

Question

For a study on emotion regulation, we used the experience sampling method via a cellphone app to collect 4 survey responses per day from participants over 2 weeks. While some of our hypotheses are straightforward to test, we are asking for suggestions on a comprehensive statistical approach to test the relative influence of personality factors and contextual factors on participants’ reported emotion intensities and chosen regulation strategies.

For each participant, we have:

• personality measures (interval score for each of the “Big Five”)
• and approximately 40 survey responses per participant with:
  • multiple contextual factors:
    • time of day (categorical, 4 time periods they were contacted throughout the day)
    • life stressor (categorical, participants chose one from a list of life stressors currently affecting them)
    • goal (categorical, participants chose one from a list of motivations for regulating emotion)
    • social context (categorical, yes/no were other people involved, and if so, categorically choosing whether they were close/non-close relationships).
  • self-reported degree to which the participant experienced each of 5 emotions during the most recent time period (ordinal/interval, collected as sliders with numerical values from 0 – entirely disagree – to 100 – entirely agree)
  • self-reported degree to which the participant used one of 6 emotion strategies (ordinal/interval, also collected as sliders with numerical values from 0 – entirely disagree – to 100 – entirely agree).

Ideally, we’d like to examine the relative impact of personality and context on the self-reported emotions, and then we’d like to examine all of those (personality, context, and self-reported emotions) in influencing the chosen emotion strategy. However, there may also be relationships between personality and contextual factors – e.g., scoring high in a certain personality trait may be associated with particular life stressors, social contexts, etc.

We appreciate your advice!

Answer 1

This seems very broad. You will probably get better answers if you can ask more focussed questions. Anyway I will give it a try:

we’d like to examine the relative impact of personality and context on the self-reported emotions

I am assuming that the the self-reported emotions are: "self-reported degree to which the participant experienced each of 5 emotions during the most recent time period (ordinal/interval, collected as sliders with numerical values from 0 – entirely disagree – to 100 – entirely agree)"

With a range of 0-100, you can probably start out by treating these outcomes as numeric/continuous. You will want to take a look at the residuals and perhaps consider more complex models later.

So you have 5 outcomes. Presumably you don't want to combine them. So you either need to fit seperate models with each of the 5 outcomes or use a multivariate model.

You have repeated measures within participant and it is a longitudinal model, so one approach will be to fit random intercepts for participant using a mixed effects model with continuous time as a fixed effect along with time of day as categorical. If you measure other demographics such as location then you may want to consider nested random effects.

there may also be relationships between personality and contextual factors – e.g., scoring high in a certain personality trait may be associated with particular life stressors, social contexts, etc.

This implies you will be interested in fitting interaction terms for the relevant variables.

So the most ambitious approach is a multivariate mixed effects model, possibly a generalised multivariate mixef model to deal with the interval and bounded scale. A set of linear mixed effects models, one for each emotion, is probably the best way to start.