p-value computation for two-tailed test with distribution fitting

[clescoat]

Calcul à revoir pour des distributions skewed

[tigloo123]

A mon avis la version two-tailed devrait etre celle par défaut.

[tigloo123]

The statistically correct solution for two-tailed test is to correctly compute the area under the curve for the both sides of the distribution:

q_val_right = best_dist.ppf(0.95, **args)
q_val_left = best_dist.ppf(0.05, **args)

p_right = 1 - best_dist.cdf(expr, **args)
p_left = best_dist.cdf(expr, **args)

[tigloo123]

@Claire : please make a test using normal distribution for two-tailed outlier testing