Stabilized IPW Estimate

Problem

Return the stabilized IPW ATE: (1/n) * (sum(t_i*y_i*p1/e_i) - sum((1-t_i)*y_i*p0/(1-e_i))), where p1 and p0 are the observed treated and control shares. Stabilized weights stay bounded when the arms are unbalanced. Return 0.0 for invalid input or when either arm has no units.

Hint

Multiply treated weights by p1 and control weights by p0.

Test cases

case 1

in: [[1,1,0,0],[10,12,4,6],[0.6,0.4,0.3,0.7]]

expected: 9.047619047619047

case 2

in: [[1,0],[5,1],[0.5,0.5]]

expected: 3

case 3

in: [[1,1],[2,4],[0.5,0.8]]

expected: 0

case 4

in: [[1,0],[5,1],[0.5,1]]

expected: 0

case 5

in: [[],[],[]]

expected: 0

Starter codePython

def stabilized_ipw_ate(treatment, outcome, propensity):
    # Your code here
    pass

5 test cases · run your code against them in the editor