Overspecification and confederate priming
This week we are looking at Loy & Smith (2021), which is paper by my former postdoc Jia Loy and me. This paper reports a bunch of experiments (5 total) where participants interact with a partner on a picture-description task where they take turns describing pictures. Unbeknownst (usually) to the genuine participants, the person they are interacting with isn’t actually a real person doing the experiment - instead, what they hear from their partner is a sequence of pre-recorded audio files designed by us. This method (sometimes nown as “confederate priming” - the confederate is the accomplice of the experimenter) allows us to control the structure of the descriptions produced by one interlocutor and see how that affects the descriptions used by the other interlocutor, the genuine participant. The standard result in these paradigms is that participants are influenced to produce descriptions which match those of their partner. We were interested in whether this effect would be modulated by the efficiency of the partner descriptions - do people follow this tendency to copy the form of their partner’s descriptions even if the partner’s labelling behaviour is inefficient (i.e. includes adjectives which are unnecessary for the partner to select the correct picture)? We started running this series of experiments in Covid times, hence the switch from lab to online data collection part-way through the paper.
As usual, in this week’s practical you’ll get a chance to look at a similar experiment in jsPsych, which will involve audio stimuli again but also the infrastructure to record audio from your participants via their browser.
Read:
All aspects of this work are licensed under a Creative Commons Attribution 4.0 International License.