Event-locked responses and encoding
Use this category when events or intervals are supplied independently of the fluorescence outcome. First decide whether the estimand is an average response, a jointly estimated response to overlapping predictors, or predictive value.
Choose the workflow
| Scientific question | Workflow | Output |
|---|---|---|
| What is the response around one event definition? | Peri-event inference | auditable session estimates and paired or independent animal-level time course |
| Does a repeated event/condition contrast differ between animal groups? | Group-by-condition interactions | animal-level peri-event difference in differences |
| What is associated with each of several overlapping events? | Event-kernel encoding | regularized kernels with complete-group validation |
| Does the result survive plausible design choices? | Event-kernel model multiverses | paired design alternatives and failure ledger |
| Does one predictor family add held-out information? | Predictor-family contributions | paired full-minus-reduced held-out score |
| Are time-course intervals calibrated? | Interval calibration | scenario-wise coverage evidence |
| How should bouts be filtered, merged, split, or overlapped? | Interval and bout policies | transformed intervals and lineage ledger |
Interpretation boundaries
- A peri-event average is not a unique causal response when events overlap.
- A kernel coefficient is conditional on all included predictors and basis choices.
- In-sample fit does not establish transport to a new animal.
- Pointwise intervals do not cover an entire time course simultaneously.
- Separate significance tests within each group do not test a group-by-condition interaction.
- Full-versus-reduced prediction measures conditional predictive contribution, not biological necessity.
Coverage gaps this category exposes
- production-ready functional mixed models for whole time courses;
- calibrated selective inference after tuning and model selection;
- richer nonlinear and interaction bases with group-safe validation; and
- reference reproductions of commonly used published event-analysis figures.
See the event-locked worked examples for simulations, public data, duration-varying behavior, and retained negative validation results.