***N.B.*** Please navigate lower to learn about differences we have discovered some differences between R and Matlab in analyzing the data from Study 2. Please read below.
We provide the data and code for the behavioral and process level analysis in Pleskac, Cesario, and Johnson (2018).
Please note data for studies 1 and 2 were collected at the University of Chicago under the supervision of Josh Correll. He has given us permission to use the data and post it here. In Study 2, half the participants completed 16 practice trials. For the other half, only the experimental trials were recorded so it is unclear if they had 16 practice trials or not.
Data from studies 3 and 4 were collected by Joseph Cesario and David Johnson at Michigan State University.
The files are organized in terms of the studies as reported in the paper. Each directory contains the error data and response time (inverse transformed) used in the ANOVAs. The .html files summarize the ANOVAs from JASP (version 0.8) with both a frequentist ANOVA and a Bayesian ANOVA.
Please note in Study 3, which has 4 factors, JASP has an error that is mistakenly flips the labels of the third and fourth factors. For instance the third factor in the analysis corresponds to the third factor as entered into the analysis. However, the label corresponds to the 4th factor. It also drops the labels of the codnition.
Each directory also contains the trial level data which was used for the Drift Diffusion Model. Note we used this data to calculate the hit and false alarm counts used in the signal detection analysis.
The model code for the hierarchical SDT and DDM are provided in each folder.
The composite folder collapses the data for the common conditions (Race X Object in neutral, clear conditions).
**Matlab vs R for Study 2**
In January 2024 Franziska Henrich contacted Tim Pleskac. She reran the DDM model for Study 2 using R and found some differences as reported in the paper. Tim Pleskac and Franziska Henrich corresponded and together reran the models. By reruning the models using Matlab (as was originally done) they were able to reproduce the results as reported in the paper. But, if they run the model in R they get different values. It is unclear why this is happening. As of June 17, 2024, the best that can be identified is that there is a difference between Matlab and R. Here are the key differences in terms of results:
Threshold: With R, there is no credible threshold difference between Black and White in the safe condition. The pattern is also different in that in the safe condition the threshold for Black suspects is lower for Whites (though not credibly so): M = -0.01 [-0.04, 0.02]. In the dangerous condition the relationship is different (though not credibly) M = 0.01 [-0.01, 0.04]
Drift for Gun: With R there is no credible race difference in drift rate for guns. The trend is the same M = 0.20 [-0.10, 0.52].
We are uncertain the exact cause, but are posting these findings here until we know more. Code and output from Matlab and R are in the Study 2 folder.
