def test_canonical_frontdoor(self):
model_str = SAMPLE_DAGS["canonical_frontdoor"]
model = parse_model_string(model_str)
is_valid = check_backdoor_criterion(model,
treatment="X",
outcome="Y",
conditioning_set={"Z"})
assert not is_valid
def test_simple_collider(self):
model_str = SAMPLE_DAGS["Collider"]
model = parse_model_string(model_str)
is_valid = check_backdoor_criterion(model,
treatment="X",
outcome="Y",
conditioning_set={"Z"})
assert is_valid
def test_m_bias(self):
model = parse_model_string([
"U₁[unobserved] -> Z",
"U₁[unobserved] -> Y[outcome]",
"U₂[unobserved] -> A[treatment]",
"U₂[unobserved] -> Z",
"A -> Y"
])
path = utils.node_path_to_edge_path(["A", "U₂", "Z", "U₁", "Y"], model.graph)
is_blocked = model.is_path_blocked(path, conditioning_set=None)
assert is_blocked
def test_shries_platt_coach_fitness(self):
model = parse_model_string(sample_dags.SHRIER_PLATT_2008)
model.update_node("Coach", NodeAttribute.ADJUSTED)
model.update_node("Fitness Level", NodeAttribute.ADJUSTED)
biasing_paths = model.get_biasing_paths(as_edge_list=True)
assert len(biasing_paths) == 0
def test_simple_chain(self):
model = parse_model_string(["x->b", "b->y"])
path = utils.node_path_to_edge_path(["x", "b", "y"], model.graph)
is_blocked = model.is_path_blocked(path)
assert not is_blocked
def test_shries_platt_no_adjustment(self):
model = parse_model_string(sample_dags.SHRIER_PLATT_2008)
biasing_paths = model.get_biasing_paths()
assert len(biasing_paths) == 3
def test_collider_descendant(self):
model = parse_model_string(["x->b", "y->b", "b->c", "b->d"])
path = [('x', 'b'), ('y', 'b')]
is_blocked = model.is_path_blocked(path, conditioning_set={'c'})
assert not is_blocked
def test_simple_collider(self):
model = parse_model_string(["x->b", "y->b"])
path = utils.node_path_to_edge_path(['x', 'b', 'y'], model.graph)
is_blocked = model.is_path_blocked(path)
assert is_blocked
def test_fork_chain_mix(self):
model = parse_model_string(["x<-a", "a<-b", "c->b", "c->d", "d->y"])
path = utils.node_path_to_edge_path(["x", "a", "b", "c", "d", "y"], model.graph)
is_blocked = model.is_path_blocked(path)
assert not is_blocked
def test_simple_fork(self):
model = parse_model_string(["b->x", "b->y"])
path = utils.node_path_to_edge_path(['x', 'b', 'y'], model.graph)
is_blocked = model.is_path_blocked(path)
assert not is_blocked
def test_long_chain(self):
model = parse_model_string(["x->a", "a->b", "b->c", "c->d", "d->y"])
path = utils.node_path_to_edge_path(["x", "a", "b", "c", "d", "y"], model.graph)
is_blocked = model.is_path_blocked(path)
assert not is_blocked
def test_canonical_frontdoor(self):
model_str = SAMPLE_DAGS["canonical_frontdoor"]
model = parse_model_string(model_str)
adj_sets = get_adjustment_sets(model)
# front door can't be identified using backdoor adjustment
assert len(adj_sets) == 0
def test_schrier_and_platt_2008(self):
model_str = SAMPLE_DAGS["Shrier&Platt, 2008"]
model = parse_model_string(model_str)
adj_sets = get_adjustment_sets(model)
assert len(adj_sets) == 7
def test_big_m(self):
model_str = SAMPLE_DAGS["big-M"]
model = parse_model_string(model_str)
adj_sets = get_adjustment_sets(model)
assert len(adj_sets) == 4
assert all("Z₁" in s for s in adj_sets)
def test_m_bias(self):
model_str = SAMPLE_DAGS["M-bias"]
model = parse_model_string(model_str)
adj_sets = get_adjustment_sets(model)
assert len(adj_sets) == 1
assert adj_sets[0] == set() # no adjustment must be performed
def test_simple_collider_backdoor(self):
model = parse_model_string(["x[T]->y[O]", "w->x", "w->y"])
adj_sets = get_adjustment_sets(model, treatment="x", outcome="y")
assert adj_sets == [set("w")]
