Resources for manuscript *How cognition bootstraps its way to complex concepts*. ## Pre-registrations For pre-registrations, see the "Registrations" tab. ## Material The `experiment_material` folder contains summary pictures of each experiment, explained in each experiment's pre-registration. ## Data The `data` folder contains R data for experiment 1, reported in *Powering up causal generalization: A model of human conceptual bootstrapping with adaptor grammars*, In Proceedings of the 44th Annual Meeting of the Cognitive Science Society (2022). Also in the `data` folder, there are three CSV files: * `subjects.csv`: each row for one participant. This table has columns * `exp_id`: 1-4, for Experiments 1 to 4 * `ix`: pseudo identifier per participant * `condition`: which curriclum the participant is randomly assigned to. Experiments 1 & 2 involve three curricula (construct, decon [short for de-construct], and combine); Experiments 3 & 4 involve two (combine and flip) * `task_duration`: mileseconds a participant took from passing the comprehension quiz to making the last generalization prediction * `age`: participant's self-reported age (>=18) * `sex`: participants' self-reported gender * `report_a`: participant's self-reported causal relation in Phase I * `coded_a`: code for participant's self-reported causal relation in Phase I * `report_b`: participant's self-reported causal relation in Phase II * `coded_b`: code for participant's self-reported causal relation in Phase II * `trials.csv`: each row is a generalization trial for one participant. This table has columns * `exp_id`, `ix`, `condition`: same as above * `phase`: A for Phase I, B for Phase II * `trial`: trial number 1-8. Sorted according to a unified baseline. In actual experiments all generalization tasks were presently in random order in each phase for each participant * `stripe`: number of stripes on the agent object in this trial * `dot`: number of spots on the agent object in this trial * `segment`: number of segments on the recipient object in this trial * `prediction`: number of segments this participant selected in this task * `is_groundTruth`: whether prediction matches ground truth (see manuscript and prereg for definitions of ground truth) * `models.csv`: each row contains model predictions for a generalization task. This table has columns * `condition`, `phase`, `trial`, same as above * `prediction`: similar as above, but contains all possible prediction values (integers 0 - 16) * `n`: the number of participants made this prediction in this task, aggregated over all four experiments * Rest of the columns follow the same naming logic: `raw_X` is the model's raw probability predicting this result in this task, and `fitted_X` is the probability after cross-validation fits (see paper and `analysis/cross_valid.R`). The list of models contain `ag` (naive adaptor grammar), `agr` (adaptor grammar with revisiting), `rr` (rational rules), `sim` (similarity-based), `gp` (gaussian process regression), `mm` (multinom regression), `lm` (linear regression), and `rand` (random selection, 1/17 for each prediction) ## Analysis The `analysis` folder contains two R files. The `cross_valid.R` file has R code for cross validating each model. For models `sim`, `mm` and `lm`, implementation of the model is part of their cross-validation loop; for other models, see Github repository <https://github.com/bramleyccslab/causal_bootstrapping> for implementation, and here cross-validation fits the trembling-hand noise parameter over raw model predictions (see Methods section in the manuscript). The `analysis.Rmd` file is a Rmarkdown document containing all the analyses and plots reported in the manuscript.