Class OSetTest

java.lang.Object
edu.cmu.tetrad.search.OSetTest

public class OSetTest extends Object
Unit tests for OSet (Henckel–Perković–Maathuis O-sets).
  • Constructor Details

    • OSetTest

      public OSetTest()
      Default constructor for the OSetTest class. This class includes a set of unit tests designed to validate the behavior of causal graph-based methods for computing O-sets in both directed acyclic graphs (DAGs) and their equivalence-class representations (CPDAGs). The tests focus on verifying the correctness of O-set computations under various causal structures and scenarios, such as the presence of confounding variables or the absence of backdoor paths.
  • Method Details

    • testOSetDag_HenckelExample

      public void testOSetDag_HenckelExample()
      Tests the OSet.oSetDag(Graph, Node, Node) method using a classic Henckel–Perković–Maathuis example DAG.

      The DAG structure is:

        U -> X -> Z -> Y
        U -> Z
        W -> X        (instrument)
      

      For the causal pair (X, Y):

      • cn_G(X -> Y) = {Z, Y}
      • pa_G(cn) = {X, U, Z}
      • O_G(X -> Y) = {U}

      The test verifies that:

      • The computed O-set equals {U},
      • and that it excludes mediators (Z), outcomes (Y), and instruments (W).
    • testOSetCpdag_matchesDag

      public void testOSetCpdag_matchesDag()
      Tests the behavior of OSet.oSetCpdag(Graph, Node, Node, int) compared to OSet.oSetDag(Graph, Node, Node) using a minimal confounding example.

      The DAG structure is:

        U -> X -> Y
        U -> Y
      

      For this graph:

      • O_G(X -> Y) in the DAG is {U}, since U is a parent of both X and Y and lies on a backdoor path from X to Y.
      • When the DAG is converted to its CPDAG, the edges become undirected (Markov equivalence class).
      • In this case, O_G(X -> Y) in the CPDAG differs from the DAG’s O-set, because the direction of causal flow is no longer uniquely identified.

      The test therefore asserts that:

      • The O-set in the DAG equals {U},
      • The O-set in the CPDAG is not equal to that of the DAG, since the CPDAG loses orientation information.
    • testOSetDag_simpleChainNoConfounding

      public void testOSetDag_simpleChainNoConfounding()
      Tests OSet.oSetDag(Graph, Node, Node) on a simple causal chain with no confounding variables.

      The DAG structure is:

        X -> Z -> Y
      

      In this graph:

      • There are no common causes (backdoor paths) between X and Y,
      • and all paths from X to Y are directed and causal.

      Therefore, the O-set for (X, Y) should be empty:

      • O_G(X -> Y) = ∅

      The test verifies that OSet.oSetDag() returns an empty set, confirming correct handling of unconfounded causal chains.

    • testOSetCpdag_agreesWithDag_inSimpleChainNoConfounding

      public void testOSetCpdag_agreesWithDag_inSimpleChainNoConfounding()
      Tests that OSet.oSetCpdag(Graph, Node, Node, int) agrees with OSet.oSetDag(Graph, Node, Node) in a simple unconfounded causal chain.

      The DAG structure is:

        X -> Z -> Y
      

      Because there are no common causes (no backdoor paths) between X and Y:

      • All paths from X to Y are directed and causal,
      • The total effect of X on Y is identified without adjustment,
      • Hence O_G(X -> Y) = ∅ in both the DAG and its CPDAG.

      This test verifies that the O-set computed from the CPDAG matches that from the DAG, confirming consistency of OSet.oSetCpdag() in the unconfounded case.