Observation Level

We can get into observation level diagnostics as well

While the process is technical, we can use the simple visualization to note ‘outliers’⁸

• Think of it as leave-one-out (LOO) cross-validation error for a single data point

Look for values above .7 (though this default can be changed)

plot(loo(attendance_brms))

---

8. Pareto smoothed importance sampling. See Vehtari, Gelman, and Gabry (2015) and Vehtari, Gelman, and Gabry (2017) for details.↩