Grid of treatment effects covering a binomial posterior

The posterior of the difference in response rates only has mass where the treatment arm likelihood does, i.e. where theta + v falls in the bulk of Beta(treatment_shape1, treatment_shape2) for some control rate node v. The grid spans that region, clipped to \([-1, 1]\), with a spacing fine enough to resolve the narrowest of scales.

Usage

binomial_effect_grid(
  treatment_shape1,
  treatment_shape2,
  control_nodes,
  scales,
  points_per_scale = 20,
  min_points = 401L,
  max_points = 20001L,
  tail = 1e-12
)

Arguments

treatment_shape1, treatment_shape2

Shape parameters of the treatment arm likelihood, as a Beta density in the treatment rate.

control_nodes

Quadrature nodes on the target control rate.

scales

Standard deviations of every factor the grid must resolve.

points_per_scale

Grid points per narrowest standard deviation.

min_points, max_points

Bounds on the number of grid points.

tail

Tail probability left outside the grid on each side.

Value

An increasing numeric vector of treatment effects.