Live demo: http://liangcj.github.io/interactivehazard/
A D3.js demo using the well-worn PBC Mayo data (n=312). Non-parametric conditional hazard estimation with interactive marker-bandwidth selection! Time-bandwidths are hard-coded for now.
Potential uses:
- Univariate data exploration
- Evaluating the proportional hazards assumption
- Bandwidth selection
Computations are all done in the browser. Estimation is done by calculating Nelson-Aalen cumulative hazard, then smoothing it using the Epanechnikov kernel.
Scaling up to ~700 observations did not affect performance. Scaling up to ~18000 noticeably affected performance but was still usable. This was on a laptop with an Intel i7-3517U processor (2012 consumer-grade mobile processor). As would be expected, RAM usage and disk speed did not appear to be limitations.
To-do list:
- Labels for graphs
- Add more annotations/descriptions
- Option to vary time-bandwidth
- Change marker-bandwidth to Epanechnikov too (with accompanying alpha shading!)