TETRAD is causal analysis system develop at Carnegie Mellon University (CMU). Researchers at CMU and the UWF Institute for Human and Machine Cognition (IHMC) are currently using TETRAD to mine terabytes of NASA satellite data to discover new causal relationships in the data. Causal models discovered by TETRAD are integrated into the Ecocasting archictecture for testing and verification. The TETRAD software system contains a library of algorithms for inferring features of causal structure from background knowledge and statistical constraints. The new Tetrad IV architecture allows for modular addition of new models and editors.

The TETRAD project is an interdisciplinary examination of what can and cannot be learned about causal structure from statistical data. Beginning from a manipulability theory of causation, we axiomatize the connection between causal structure and statistical independence (Spirtes, Glymour & Scheines, 1993), specify and prove correct several algorithms for inferring features of causal structure from background knowledge and statistical constraints, provide procedures for computing the effects of interventions in incompletely known systems, and implement these algorithms, along with facilities for simulation, parameter estimation and hypothesis testing, in practical software (TETRAD II, III, and IV) intended to help scientists with real causal questions. An accessible introduction to the theory can be found in (Scheines, 1996). The TETRAD programs have different capacities. TETRAD III is the most flexible, but has a command driven interface. TETRAD IV has a graphical interface in which users can specify a flow chart of procedures. Elaborations of the TETRAD IV suite algorithms are planned to include Bayesian and other scoring searches, computation of effects of interventions, and improved methods for clustering and inferring latent structure.

TETRAD software has been successfully applied to Naval manpower research, to calibrating satellite data, to discovering the causes of high college dropout rates, to the problem of measuring grief, to the problem of predicting which pnuemonia patients are at risk, to identifying mineral composition from spectra, and to the question of whether prolonged exposure to environmental lead lowers IQ scores in children. For more information, please see http://www.phil.cmu.edu/projects/tetrad/index.html.