Measure that the first speaker actually speaks first on each topic

[JamesPHoughton]

How do we operationalize manipulation check?

We have participants' audio volume over time, and we can discretize this as "speaking" or "not speaking", and record "start speaking" and "end speaking" blocks.

• at 90 seconds, the first speaker should have at least 60 seconds of talk time?
• at 90 seconds, the first speaker should have the most talk time?
• at 90 seconds, the non-first speakers should have less than 30 seconds talk time?

[xehu]

Is there any way to enforce it in a more 'hard' way --- e.g., literally disable the microphones of all other speakers such that only one person is allowed to speak?

[willschulz]

I see two questions here:

1. How to operationalize an empirical manipulation check? The first step is to featurize the audio to identify when people are talking, and construct a talk time measure based on that. Then we can develop a few operationalizations of the manipulation check along the lines James suggests, verify that they all correlate, and if not, investigate why and pick the best one based on face validity.
2. If the manipulation check shows non-compliance (first speaker refuses to speak, other people are speaking), what should we do about it? I really like @xehu 's suggestion of enforcing by muting. @JamesPHoughton could you and your colleagues implement muting-on-a-timer? Additionally we could try to fix non-compliance by making instructions more salient (perhaps by delivering verbal instructions as I've suggested elsewhere).