psychelzh / cogstruct-dev Goto Github PK

Formerly, the clustering of factor distributions were performed on the averaged results across different n_fact parameters, which might obscure the results. It might be better to search the results with best silhouette score across different n_fact parameters.

Based on our previous studies, data cleaning includes:

• Remove unmotivated participants (e.g., too many errors by binomial test)
• Remove outlying participants

It is extremely complicated to set the rule to label a participant as unmotivated. Nevertheless, it is a necessary thing. For now, we only did the outlier detection part.

Here is the screenshot of the stimuli program code:

It means, all the response times are 100 milliseconds less than the real. The histogram of response times proved this:

So, we need to add 100 to all the response times data.

Based on this article, it would be remarkably interesting to fit an oblique bifactor model. Some preliminary trials found some fitting issues with the current model structure. Needs further inspection.

I should have used time instead of ocassion. But considering the model is extremely time-consuming, I will keep the spelling error but keep a record of the error here.