Issue predicting RTs in a go/no-go task

[Chase Mackey]

Hi everyone, my research group is using stimulus coding to model performance in a go/no-go task. One research group that did something similar  (de Gee et al.) have made their code publicly available here:  https://github.com/hddm-devs/hddm/blob/ master/hddm/examples/gonogo_demo.ipynb

publication here: https://elifesciences.org/articles/54014

To fit my data, I used a slightly different model: "model = hddm.HDDMStimCoding(dataframe,stim_col='stim',split_param='v', drift_criterion=False, bias=True, p_outlier=0, include=('z','a','v','t'),depends_on={'v':['NMF2'],'t':['if_catch_trial'],'a':['if_catch_trial']})" This means the drift rate depends on the stimulus type (NMF2), and t and a depend on trial type (whether go or no-go was the correct response. 

The model predicts accuracy well (figure below panels C and D), but not reaction times (panels G and H, dashed = model, solid = behavior)

Other relevant information: we emulated de Gee et al. by using 10 runs as an input to model.optimize, and we used 3,000 samples with a 500 sample burn-in. We had a minimum of about 60 RTs in each condition, often more like 80-100. 

Chase