Question: Go/No go

[Linda Fiorini]

Greetings HDDM Community,

I'm currently delving into analyzing data from a Go/No-Go task, and I find myself uncertain about the best approach.

Initially, I attempted to utilize the code outlined in this tutorial (https://hddm.readthedocs.io/en/latest/tutorial_gonogo_chisquare.html) by computing a model for each subject. I fit the model by inserting 'NaN' values for cases where there were no reaction times.

However, I've encountered two primary issues:

1. The results vary significantly each time I run the model, rendering the analysis unreliable (see images below: each one represents one different fit with same parameters and same data).
2. I am sacrificing the hierarchical structure of the DDM model by fitting a separate model for each subject


In my quest for alternative approaches, I found another paper using the same analysis (found here: https://www.nature.com/articles/s41467-021-23890-7) with associated code available on GitHub (https://github.com/andrillon/wanderIM/blob/master/behav/HDDMStimCoding_GNG.py. The results from this study appeared to make sense, but I still have some questions:

1. If I use NaN values the models does not run: How do I handle cases where there are no-go responses if I can't use NaN values? Unfortunately, I couldn't find the file they used for in the data storage.

2. Similarly, I'm concerned about potentially neglecting the hierarchical aspect of the DDM by fitting the entire model without specifying individual subjects.

Any guidance or insights into resolving these dilemmas would be greatly appreciated.

Thank you very much,

Linda

[Edoardo Pinzuti]

Dear Linda, 

For the second part of your question, I think you should input missing values with 999 and you can also use the hddm.HDDMRegressor with that.

Hope this helps.

Edoardo

--
You received this message because you are subscribed to the Google Groups "hddm-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hddm-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hddm-users/52c6228b-28af-4bdc-b57a-e9fab33afdc0n%40googlegroups.com.

[Katie Alonso]

Hi Linda and Edoardo, 

I'm having a similar issue with coding the Go/No-Go rt for the trials with no response. When coding missing rt's as 999 (either +999 for all or +999 for correct no-go trials and -999 for incorrect no-go trials) hddm.HDDMRegressor gives the zero probability error 'stochastic wfpt.1's value is outside its support, or it forbids its parents' current values.' Whereas when coding missing rt's as NaN, hddm.HDDMRegressor works without error but then the step-out procedure of model sampling fails. Please let me know if you've resolved how to properly run through both functions without error. 

Thank you!

Katie