A Appendix 1: Description of unanalysed data
A.0.0.1 Experiment 0
This branch of experiments was the core of the Dates task. As such, most experimentation and piloting happened on this branch, meaning that there were many versions of the study where data were collected for which analysis is not included here.
A.0.0.2 Experiment 1A
The earliest versions contained a bug where advisors instructed to agree with a participant instead provided advice identifying the correct answer. Other versions had a bug in the staircasing code used to titrate the difficulty of the task was converging on too high a value (74% initial estimate accuracy as opposed to 71%). Once the staircasing bug was fixed, two more experiments were run, one with 60 practice trials in which participants did not quite reach the desired accuracy before the beginning of the main experiment, and one with 120 practice trials which constitutes the data analysed below.
A.0.0.3 Experiments 1B and 2B
Early versions included a bug which prevented feedback from being shown during the Familiarisation phase even to participants in the Feedback condition. These participants could theoretically be included in the No feedback condition regardless of their condition label in the data, but this is not done here.
A.0.0.4 Experiment 3A
Pilot data were collected to ensure the study functioned properly, and so the data are not analysed with regard to the hypotheses. The v1 Mixed design was conducted and run as a proper experiment in which participants learned about both advisors simultaneously (preregistered at https://osf.io/5z2fp), but there were no effects in the data.
A.0.0.5 Experiment 3B
The first version included a bug in which the advisor choice options were not recorded, making it difficult or impossible to work out which trials included a choice of advisor. Data for these participants could be included in a study that was agnostic about advisor choice.
A.0.0.6 Experiment 4A
The first version had a bug which meant that the advisors were identical in the Familiarisation phase. Both the first and second versions had a bug in which advisors who were supposed to agree with the participant’s initial estimate gave the correct answer rather than agreeing. These participants’ data could be included in an analysis which used a participant’s actual experience of advice to predict their advice-taking and advisor choice behaviour, provided appropriate care was taken to reconstruct the advice data from the raw values instead of relying on the reported summaries.
A.0.0.7 Experiment 5
Several versions of this study were run in the course of developing the manipulation. The initial version had a bug that prevented the groups from being visually distinct. Later versions introduced a clearer manipulation, equivalent rather than genuinely misleading advice for the Sometimes misleading advisor, and rating of advice deceptiveness, respectively.