Question about Lapse Specification in HSSM

[simonarvin]

Hi all, first of all, thank you for building and maintaining HSSM.

I have a few conceptual questions about how lapse/outlier processes are implemented.

1) What happens to outlier trials?

When a trial is assigned to the outlier process (with probability p_outlier):

• Is it effectively excluded from the DDM likelihood and instead evaluated under the lapse distribution?
• Does it assume random responding (e.g., 50/50 choice), or is choice still modeled in some structured way?

In other words: how exactly are these trials treated in the likelihood?

---

2) Random effects for lapse parameters?

It appears that p_outlier (and possibly lapse) are currently global parameters.

However, participants likely differ in lapse tendencies (e.g., attention, fatigue, motor slips).

Is there currently a way to:

• Add subject-level random effects to p_outlier?
• Specify something like p_outlier ~ 1 + (1|id)?
• Allow predictors (e.g., block number, fatigue, pathology) to influence lapse probability?

If not currently supported, is this conceptually compatible with the HSSM framework?

---

Thanks again for the excellent tool, really appreciate any clarification!

[AlexanderFengler]

Hi @simonarvin ,

Re question 1:
The way this currently works, is that you simply apply a mixture distribution globally, so there isn't exactly a trial-wise assignment of 'outlier' vs. 'non-outlier', but rather we a priori define a mixture distribution with some p_outlier setting and all trials are assumed to follow that. You can however still define a covariate-driven model to p_outlier as below.

Re question 2:
Yes HSSM supports specifying a model for p_outlier, it is essentially treated as just another parameter for a model. So you can define the kind of regression as you suggest.

Best,
Alex