TETRAD 3:

 

Tools for Causal Modeling

 

 

 

User's Manual

 

 

 

 

 

 

Program by:

 

Peter Spirtes, Richard Scheines,

Christopher Meek, Thomas Richardson, Clark Glymour,

Herbert Hoijtink and Anne Boomsma

 

 

 

 

User's Manual by:

 

Richard Scheines, Peter Spirtes,

Clark Glymour, Christopher Meek and Thomas Richardson

 

 

 

 

 

 

 

 

 

 


  • Table of Contents
  •  

    Getting TETRAD 3

     

     

    Tetrad 3 runs on Sun sparcstations or PCs. You can download it, or run it remotely.

     

    Downloading via FTP

     

    For

    1) FTP to: unix.andrew.cmu.edu

    2) User ID: philoso

    3) Password: guest

    4) cd /afs/andrew/hss/phi/tetrad3

    5) set the mode to binary:

    6) To get the Sun version: get suntet3

    7) To get the PC version: get wintet3.exe

     

     

    Running Remotely

    1) telnet to: sun4.andrew.cmu.edu

    2) User ID: philoso

    3) Password: guest

    4) The program is the executable file: /afs/andrew/hss/phi/tetrad3/suntet3

     

     

    Acknowledgments

     

    Research for this work was funded by the Navy Personnel Research and Development Center and the Office of Naval Research, under grants #N00014-88-K-0194, #N00014-89-J-1964, #N00014-91-J-1361, and #N00014-93-1-0568. We thank Steve Sorensen, Susan Chipman, Jan Callahan, and Laurie MacDonald, all of whom have contributed ideas that influenced the design of the program.

    We thank Amy Culbertson, Phil Craiger, Jeff Houston, Bob Morrison, Mike Nikada and Rich Reimer, all of whom have patiently tried early versions of TETRAD 3 on their data and alerted us to bugs and conceptual problems. We thank Steve Klepper for lots of statistical advice and for trying to convince us to avoid a philosopher's jargon. Undoubtedly we were too stubborn and this text will still be in a language unfamiliar to the average reader.  We thank Peter Bentler, Marek Druzdzel, Gregory Cooper, Frank Faulbaum, Steve Fienberg, Brian Junker, Kevin Kelly, Michael Meyers, Judea Pearl, Thomas Richardson, James Robins, Teddy Seidenfeld, and Nanny Wermuth for their challenges and suggestions.

    Finally we thank our spouses, Jessica Feldman, Martha Harty, Alison Kost, and Mary Meek, for their patience.


     

    How to Use this Manual

     

     

     

    TETRAD 3 is a collection of tools, or modules, to help build causal models of statistical data. We have organized the manual into two parts. The first (chapters 1-4) describes information common to the entire enterprise, and the second (chapters 5-17) contains detailed descriptions of each of the 13 modules.

    All of the modules can be operated after reading only the first few pages of the chapters that describe them, but more sophisticated uses of the program require reading chapters 1 and 3, and the first five sections of chapter 4; in addition, a reading of chapter 2 is strongly recommended. Chapter 1 describes what the program does in general terms, gives an overview of the modules, and describes several common research contexts in which TETRAD 3 can help. Chapter 2 provides an abridged glimpse at the theoretical foundation upon which TETRAD 3 is built. It describes the basic connection between causal hypotheses and probability distributions, several key theorems that characterize when causal inference is possible, and the two kinds of statistical models that TETRAD 3 is meant to handle: recursive linear structural equation models and Bayesian networks. The full theory is given in Causation, Prediction, and Search (Spirtes, Glymour, & Scheines, 1993). Chapter 3 describes how to operate the program. It documents how to enter and exit the program, and gives several examples of how to use the procedures. Chapter 4 explains the input to TETRAD 3, that is, how to construct input files, how to enter commands, the syntactical restrictions on variable names, numbers, and so on. The beginning of the chapter describes information that is essential for using any module, but most of the chapter is intended to be a reference for questions that come up in the course of extended use. Part 3 (chapters 5-14) contains descriptions of each of the modules that comprise TETRAD 3. R- check thisThe Appendix describes some of the algorithms used and characterizes their reliability. 

    Throughout the manual we give examples of actual TETRAD 3 sessions. Instead of using "suntet3" or "wintet3" which are the names of the Sparcstation and PC versions, respectively, we use "tetrad3" throughout.  For example, a session from chapter 3 illustrating how to enter and exit TETRAD 3 is given here:

     

    Session 3.1: Entering and Exiting TETRAD 3

    ********************************************************************

    tetrad3> pwd

    /afs/andrew.cmu.edu/hss/phi/tetrad3

    tetrad3> tetrad3

     

                        TETRAD - Version 3.0

                    Beta Version - October, 1996

                        COPYRIGHT (C) 1996 by

            Peter Spirtes, Richard Scheines, Clark Glymour

               Christopher Meek, Thomas Richardson,

               Herbert Hoijtink, and Anne Boomsma

     

     

    For help, type "help"

    Initializing Data Structures

    >exit

    tetrad3>

    *******************************************************************

     

    The sessions are titled by chapter number and session number, with the titles underlined. The actual sessions are demarcated above and below with a line of "*************." Text printed by the program is in fixed width font, with user input in bold. Comments that explain the input or the response of the program, but are not actually part of the user input or the program response are indented and in Times italic font, for example:

     

    To exit TETRAD, we type exit.

     

    We also give many examples of input and output files, which are all regular text files. (Often, for brevity, we do not reproduce the entire output file.) We list these files in a fixed width font, and separate them from the manual text with a line of "#######"s that includes the file name. These files are included on the disc containing TETRAD 3 in the R- do this"Examples" directory. The line of "###"s is not part of the file itself. For example, the following shows the input file pub.dat.

     

    #############    pub.dat     #####################

    /Covariance

    162

      ab  gpq  pre   qfj   sex  cite  pubs

    1.00

    .62 1.00

    .25  .09  1.00

    .16  .28   .07  1.00

    -.10 .00   .03   .10  1.00

    .29  .25   .34   .37   .13  1.00

    .18  .15   .19   .41   .43   .55  1.00

     

    /knowledge

    addtemporal

    1 sex

    2 ab

    3 gpq pre

    4 qfj

    5 pubs

    6 cite

    #############    pub.dat     #####################

     

    When TETRAD 3 asks a question, it often provides a default answer that can be invoked by hitting a carriage return. The default is always in square brackets, for example, [y]. We explicitly write the symbol for a carriage return <CR> only when a default answer is chosen, but every command to the program must be followed by a carriage return before it is executed.

    TETRAD 3 is a work in progress; new versions will be produced, and bugs removed. We would appreciate any comments on any part of the program and manual. Please send them to any of the authors at:

    Dept. of Philosophy

    Carnegie Mellon University

    Pittsburgh, PA  15213

    412-268-8568

     

    Email:

    scheines@andrew.cmu.edu

    ps7z@andrew.cmu.edu

    cg09@andrew.cmu.edu

     









     

    Table of Contents