1

I have this data where participants listened to each stimulus and gave a rating (1~5) for each. But each stimulus contains multiple sentence-like units (i.e., intonational phrase) within it, and for each of those units, we have a boundary tone (i.e. categorical variable, with levels like fall, rise, fall-rise etc.) - so multiple tones for each stimulus.

We want to see how and whether the types of boundary tones used affected the participants' ratings (dependent v). What makes things complicated is that what meaning/attitude/emotion each tone conveys depends on the sentence; I could code the sentence as a random effect. So it makes sense to me to treat each intonation phrase (sentence) as a separate data point, but in this case, we would be treating one response (rating) as it is multiple. Does this make sense?

What would be the best approach to deal with this data statistically?

mkt
  • 18,245
  • 11
  • 73
  • 172
mousemouse
  • 11
  • 3
    I don't think it makes sense to do this if a single rating was given to a set of intonational phrases. This sounds like a multiple regression problem - use the boundary tone of each intonational phrase as a predictor and the rating for the stimulus as the response. – mkt Jul 14 '23 at 05:00

1 Answers1

1

You have a single rating that is the outcome of not just one input (stimulus) but several (intonational phrases). Instead of building multiple models with the same response (rating) and different predictors (intonational phrases), this can simply be done with one multiple regression model. Use rating as the response (i.e. dependent variable) and the boundary tones of the intonational phrases as predictors (i.e. independent variables). You might also consider using 'participant' as a random effect because each participant rated multiple stimuli.

And if your ratings are best treated as ordinal and not continuous, which is likely, the best solution would be a mixed-effects ordinal logistic regression. This is a good thread about how to fit them:

How to use ordinal logistic regression with random effects?

mkt
  • 18,245
  • 11
  • 73
  • 172