The statistical analysis code for this research project can be found here: ### A. Analysis of Recall Accuracy We aggregated the recall accuracy for each of the conditions in the four experiments, before examining the distribution of accuracy across the responses in whole-report. **ConjunctionWholeReport_Analysis.m**: This script aggregates the recall data of all conditions, and examines the distribution of correct responses in each condition. **ConjunctionWholeReport_ErrorAnalysis.m:** This script aggregates the recall data by ordinal response, and calculates the proportion of each response error type (both correct, only color correct, only orientation/shape correct, both wrong) across the ordinal responses ### B. Generation of Synthetic Probability Distributions To conduct the formal model comparison, we needed to generate synthetic probability distributions for each of the competing models. These distributions were generated through simulation – the functions for the simulation are **generate_wholeReport_dist_pointerModel.m**, **generate_wholeReport_dist_pointerWithIndependentDropping.m**, **generate_wholeReport_dist_featureModel.m** for the three models. The **distributionFitting.m** makes use of saved synthetic probability distributions – the **independentFeatureModel_probDists.mat** for fitting of the independent features model, and **objectPointerModel_probDists.mat** for fitting of the object-based pointer model with general feature dropping rate. ### C. Formal Model Comparison **distributionFitting.m**: This script conducts the formal model comparison for the conjunction condition of the experiment, comparing the all-or-nothing slot ("strong object") model, the independent features model, and the object-based pointer model with a general feature dropping rate. With the object-based pointer model with independent feature dropping having a 4-dimensional space, I went with a Bayesian Optimization method to search for best-fitting values. The script for that is **bayesianOptimization_M4.m**. ### D. Model Comparison Results The results of the formal model comparisons for each experiment are saved as .mat files.