Which statistical technique should I use for a within-person repeated measures study?

Question

I have collected all my data for a study and need to run my analysis but have come unstuck (I should have planned better beforehand I know).

I'm looking to see whether personality traits (five trait variables values ranging from 0-5) predict whether someone will give feedback at work (discrete outcome, yes/no) and the type they will provide (likert lots of positive feedback 1-6 or likert lots of negative 1-6)

Participants completed a survey on time one which captured their personality and then were invited to complete four additional weekly surveys (one survey each Friday for four consecutive weeks). These surveys captured data on the feedback they gave that week.

So, I have my independent variables (personality) and I'm trying to predict my repeated measures outcome variables (feedback).

I also have the extent to which the participants worked virtually as a moderator variable, collected alongside the feedback data (also repeated measures values ranging from 0-100). The research question here is whether working virtuality influences how extraverted (one of the five trait variables) participants give feedback and so will be a separate model to the one above.

What analysis do I run?

My instinct is that it should be a fixed effect model to see whether personality predicts (non-time varying variables) the feedback outcomes (time-varying variables) and a random effects, multilevel model to examine the moderating effect of virtual working (time-varying) on extraversion (non-time varying).

All analysis will be done in R.

Answer 1

Your instinct is right, you will probably need to use a fixed effects model to examine the relationship between personality traits and feedback outcomes, and a multilevel model to examine the moderating effect of virtual working on extraversion.

For the fixed effects model, you can use logistic regression to predict whether someone will give feedback at work and ordinal regression to predict the type of feedback they will provide (positive or negative). You can include the five personality traits as predictors in the model.

For the multilevel model, you can use a linear mixed-effects model to examine the moderating effect of virtual working on extraversion. You can include virtual working as a time-varying predictor and extraversion as a non-time varying predictor. You can also include random intercepts and slopes for each participant to account for the repeated measures design.

You can use the lme4 package in R to run the multilevel model. You can also use visualization techniques to examine the distribution of the generated data and evaluate the model's performance using metrics. If the generated distribution is not satisfactory, you can adjust hyperparameters or use advanced techniques like adversarial training to improve the model's performance.