Parameter interpretation and feedback on modeling approach

[arellag]

Hi all,

I have an experiment where participants view two images side-by-side on screen and then use the keyboard to indicate their choice of picture (‘g’ key for picture on the left, ‘h’ key for picture on the right). There are two experimental conditions: 1) “certain“ - a visual cue indicates which picture participants must choose. 2) “uncertain” - participants are free to choose whichever picture they'd like. The goal is to see how drift rates and thresholds differ between conditions and particularly look at individual differences profiles.

As a first step I am attempting to fit very simple models separately per participant. I have coded response as the side choice: -1 (left) vs 1 (right), and allow drift and threshold to vary by condition, with no intercept. If necessary to capture properties of the stimulus itself beyond what side it’s on, I can also add predictors for word frequency and word length of the picture names (we actually run this also as a language task with verbal naming of the target), but I’d rather keep the models very simple for now. To clarify, I want to capture the drift rate towards the chosen target picture and the threshold separation between the two picture options, but don't care as much about what the particular picture choice is.

Here’s the basic code:

 model = hssm.HSSM(
        data=df,
        include= [v_in_condition, a_in_condition],
        model="ddm",
        extra_namespace={"lvl": lvl},
        prior_settings="safe",
      p_outlier = 0
        )

lvl  = [‘certain’, ’uncertain’]

v_in_condition =   {
    "name": "v",
    "prior": {
        "condition": {"name": "Normal", "mu":0,  "sigma": 1.0},
    },
    "formula": "v ~ 0 + C(condition, levels = lvl)", 
    "link": "identity",
}

a_in_condition =   {
    "name": "a",
    "prior": {
        "condition": {"name": "Gamma", "mu": 1.5, "sigma": 0.75},
    },
    "formula": "a ~ 0 + C(condition, levels = lvl)", 
    "link": "log",
}

Now, here’s an example subject who appears to display the expected results: higher drift rate for the certain condition, higher threshold for the uncertain condition. But I am not sure I am modeling this correctly, nor how to interpret the signs — what does a negative drift rate here represent? Is this capturing any left/right side biases people may have (which I do not care to capture in the drift)? My understanding from other posts of people using stimulus coding is that I may need a custom link function, but unsure how to go about that. Same question for interpreting the negative threshold (sounds similar to this issue).

And bonus example, a subject whose results don’t quite make sense — higher threshold but also higher drift (huge difference) for the uncertain condition.

Any advice on how to approach and/or interpret this is welcomed! I am new to modeling so it’s been difficult to assess if I’m in the right direction.

Many thanks!