This is the OSF pre-registration and associated analysis and data for the paper *Semantic space is sculpted by episodic associative learning: A behavioral representational similarity account* by Catherine Walsh and Jesse Rissman, Ph.D. Pre-print is located at https://psyarxiv.com/8kq59/. We have provided all files necessary to re-create the analyses and figures presented in the paper, in addition to the development of our novel imputation method. Run time for our primary analyses should take 10-20 minutes on a normal desktop computer (the length is primarily due to the computation time for the shared vector approach validation of imputation). If there is anything missing, please direct inquiries to Catherine Walsh (crewalsh@g.ucla.edu). ### Experimental and Analysis Protocol Overview 1. Data was collected using lab.js: **two_learning_with_arrangement_init.json** and **two_learning_with_arrangement_final.json** 2. Data were processed initially using **process_indiv_data.py** and **process_indiv_SONA.py**. This created a csv of data for each subject. 3. Data pulled into R, organized into lists, subjects who did not meet inclusion criteria were removed. Data are saved in **NEW_final_data.rds** Please note that ordering of the output CSVs from previous scripts may be different from the saved data in this step, which may result in incorrect subjects being excluded. Please use the final datasets in the output folders, or from this .rds file to reproduce our results. 4. Pulled into **data_agg.Rmd** for more refinement, output is saved as csvs in **data_csvs** directory, in addition to an RDS file (that gets used to store all needed data for final analysis). 5. Analysis occurs in **FINAL_for_github.Rmd**. Annotations are provided in workbook. ---------- ### Data included *data* : list of behavioral data information *all_data* : long dataframe of trials, including: - PTID - word pair information: pair label, LSA cosine similarity, word2vec similarity - learning condition (test vs restudy) - relatedness (a priori categorization and 2 relatedness judgements) - correct at learning and final - response at learning and final (and whether response at final was misspelled) - duration of response at learning and final - similarity at initial arrangement - similarity at final arrangement - change in similarity across learning - whether or not the trial type during the final learning opportunity was the same as the previous trial (i.e. did the task switch?) *processed_data*: list of average accuracy data by relatedness, learning condition (not bifurcated), day. Split by normative a priori relatedness and subject specific relatedness judgements. *tidy_processed_data_norm*: (normative) processed_data collapsed into a tidy format dataframe. *bifurc_data*: list of average accuracy data by relatedness, learning condition (bifurcated), day. Split by normative a priori relatedness and subject specific relatedness judgements. *tidy_bifurc_data*: (normative) bifurc_data collapsed into a tidy format dataframe. *tidy_arr* : long dataframe of similarity trial data from only OBSERVED pairs (arr_data[[sub]]$observed) and not explicitly bifurcated, including: - PTID - word pair information: pair label (cue and target) - LSA cosine similarity - word2vec similarity - learning condition (test vs restudy) - relatedness (a priori categorization and 2 relatedness judgements) - correct at learning and final - similarity at initial arrangement - similarity at final arrangement - pairwise change in similarity across learning - change in cue and target across learning (for all and NN) - correlation to word2vec before and after learning - asymmetry (only for NN) *tidy_all_arr*: same as tidy_arr but for ALL potential pairs - no representational change analyses, just pairwise change (not explicitly bifurcated) *observed_arr_avg_all_trials_NN* : representational change measures averaged across (bifurcated) test conditions, split by subsequent recall. Used to create: - corr_lme_all_trials_all_test_conds_NN - avg_change_init_to_final_NN - filtered_final_all_test_conds_NN - diff_to_w2v_all_test_conds_NN *all_arr_trials_NN* : trial level representational change data for each subject, with out tested but recalled incorrectly at day 1 condition. *bifurc_arr* : list of average change in similarity by learning condition (bifurcated) and relatedness for each subject [no additional information] *tidy_bifurc_arr* : bifurc_arr [average change in similarity by learning condition (bifurcated) and relatedness] collapsed into one dataframe *top_unpaired_sim* : change in pairwise similarity for semantically related lures - all potential target words that have stronger similarity to a to-be-learned cue than its associated to-be-learned pair - PTID - to be learned cue - to be learned target - semantically related lure target word - learning condition, relatedness, subsequent memory for associated to-be-learned pair - word2vec similarity, initial similarity, final similarity, change in similarity for the to-be-learned and lure pairs *scatter_data* : similarity data and accuracy data for individual differences correlations across conditions. Excludes tested but incorrectly recalled at Day 1 condition. Included, but not used for primary analyses: - *acc* : average accuracy for each subject across combinations of conditions, including behavioral testing effects - *change_dissim* : dataframe of change in similarity measures (raw and testing effect) by condition, relatedness and combinations. Uses ALL trials.