## Data The data used for all analysis and modelling is contained in cleaned_data.csv in long form (and "img/cleandats.RData"). The data transformations and exclusions necessary to re-obtain this from the raw data set are run in 1-extract.R (the first uncommented lines of code). The key for each column is below: - s is "subject", indexes participant number - sess is "session", indexes which session it was for the participant (1,2 or 3) - block indexes which block it was for that session (one, two, or three) - cond indexes which experimental condition it was (Low Reliability= AUTO_L, High Reliability = AUTO_H, Manual=MANUAL) - failtrial indexes whether the automation was correct or incorrect. nonf (non-failure) for automation correct, fail for automation incorrect - failtrial_H is a redundant column indexing whether automation would have been in a failure in high-reliability conditions (and so is identical to failtrial for hr trials). Can be safely ignored. - trialnum is the number of the trial within a particular block of a particular session - S is stimulus type (c for aircraft in conflict, n for nonconflict) - R is response type (C for conflict, N for nonconflict) - RT is response time in seconds ## Script Explanations The manifest results were generated using 2-results_writeup.Rmd (with wording edits in ms word). The tables in the supplementary materials were generated with 2.5-supplementary_writeup.Rmd. The cognitive model reported in text was specified in R/3-model_specification_A.R. Originally, a model with a wider prior on A was fitted (R/3-model_specification.R), but this would not converge for two participants. Model results are generated in R/4-model_results_writeup.Rmd, with pre-computing of some results taken care of in 5-individual_diffs.R and 5-posterior_exploration.R. Scripts ending with _27 are similar analyses to those reported in text, but with the extra 3 participants that we accidentally tested beyond the full counterbalance. ## Loading pre-computed model outputs In the scripts, there is a special load function (loadpath()) that I use to get pre-computed modelling results (.RData files) from my dropbox. This function will not run and so is commented out. Instead, download pre-computed results from this OSF page (Reliability_Samples.zip) and load them into R using the standard load() function. ## Response to reviewers Some additional analyses were included in response to reviews. Specifically, we reanalysed individual-difference correlations with accuracies transformed to the probit scale, and mean RT transformed to mean[log(RT)]. In addition, we made nicer looking plots, and examined overall accuracy across conditions (marginalized over automation correct/automation incorrect trials). We also analysed the test-reliability and internal consistency of the trust in automation scale. These additional analyses are available in Reliability_Analysis_ReviewerResponse.zip. For convenience this also includes a precomputed object saving the new (transformed scale) model-based correlation analysis results (which run slowly): "model_correlations_A_lb_TRANSFORMED.RData"