Léo Belzile
@leobelzile.bsky.social
Associate professor at HEC Montréal. Gaspé peninsula native. Statistics of extremes. Mountain enthousiast, theatre aficionado.
Yes. I have seen this used in the context leave-one-out cross-validation (LOO CV), taking the LOO posterior predictive distribution to be compared against the observation in P-P and Q-Q plots in Vethari et al. (2017) doi.org/10.1007/s112..., for example. Not sure it has a name.
Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC - Statistics and Computing
Leave-one-out cross-validation (LOO) and the widely applicable information criterion (WAIC) are methods for estimating pointwise out-of-sample prediction accuracy from a fitted Bayesian model using th...
doi.org
June 16, 2025 at 2:53 PM
Yes. I have seen this used in the context leave-one-out cross-validation (LOO CV), taking the LOO posterior predictive distribution to be compared against the observation in P-P and Q-Q plots in Vethari et al. (2017) doi.org/10.1007/s112..., for example. Not sure it has a name.
I have a couple of datasets from an experimental design course bundled in an R package: lbelzile.github.io/hecedsm/. Maybe `LJLSFBM20` would work? It is a within-subject three-way ANOVA (2x2x2), but it can be cast to a within-subject one-way (8 categories). Plenty of contrasts to test!
Data Sets for Experimental Design and Statistical Methods
A collection of curated data sets and utility functions for the course MATH 80667A at HEC Montreal.
lbelzile.github.io
May 21, 2025 at 3:18 PM
I have a couple of datasets from an experimental design course bundled in an R package: lbelzile.github.io/hecedsm/. Maybe `LJLSFBM20` would work? It is a within-subject three-way ANOVA (2x2x2), but it can be cast to a within-subject one-way (8 categories). Plenty of contrasts to test!
The hazard is fitted via splines, so in practice easier to view this via the Bayesian lense.
April 26, 2025 at 5:45 PM
The hazard is fitted via splines, so in practice easier to view this via the Bayesian lense.
If you are willing to go away from the Cox model (by replacing the nonparametric hazard with a semiparametric component), you can use the life table approach. You bin the time axis and run a Poisson regression: see doi.org/10.2307/2346..., doi.org/10.2307/2986..., doi.org/10.1080/0162...
April 26, 2025 at 5:44 PM
If you are willing to go away from the Cox model (by replacing the nonparametric hazard with a semiparametric component), you can use the life table approach. You bin the time axis and run a Poisson regression: see doi.org/10.2307/2346..., doi.org/10.2307/2986..., doi.org/10.1080/0162...
Fully agree with that. The methods extends beyond linear regression models in a natural way. An excellent reference, especially for the sensitivity diagnostics: students are always surprised to see that the effect could be spurious with only 5-10% correlation due to omitted variables.
March 14, 2025 at 1:21 PM
Fully agree with that. The methods extends beyond linear regression models in a natural way. An excellent reference, especially for the sensitivity diagnostics: students are always surprised to see that the effect could be spurious with only 5-10% correlation due to omitted variables.