This e-mail address is a graphics file in order to protect me from spam mails. Sorry for the inconvenience -- you have to type it into your e-mail client.
TDA, one of the most powerful programs for the analysis of event history data, was developed by Götz Rohwer, University of Bochum. I appreciate his continuous efforts at explaining to me the program as well as the underpinnings of event history analysis. Although he provided many valuable comments on earlier versions of these pages, any remaining errors are mine. Note that since a couple of years, Götz Rohwer is co-authoring the program together with Ulrich Pötter.
I am also grateful to Prof Kazuo Yamaguchi, Chicago, for his kind permission to use a data set from his book on event history analysis (Yamaguchi 1991) to introduce TDA.
None of these persons is responsible for the contents of these pages, however.
These pages give a short didactical exposition of analysis of transition data with TDA - a program for Transition Data Analysis, also known as event history analysis, survival data analysis or analysis of failure time data. In fact, TDA can accomplish many more things, but these are not covered in this short introduction.
But even as far as transition data analysis is concerned, these pages cannot be considered as an exhaustive manual; rather, their purpose is to serve as a kind of introductory tutorial. After describing how to work with TDA on a PC under DOS or MS-WINDOWS (Overview: Working with TDA) and a short description of the essential parts of a TDA command file (Command Files for TDA), I proceed by discussing some problems arising in the analysis of a concrete example data set (Analyses of an Example Data Set). Finally, some examples for creating plots related to event history data are given (TDA plots).
In the "analysis" and the "plot" sections, I will argue mainly by reference to example command files which (hopefully) can be downloaded from this server. The "analysis" section will deal quite extensively with one special problem of transition data analysis, namely, time-dependent covariates. This is one of the most fascinating aspects of transition data analysis, yet also one that sometimes is not very easy to deal with. In addition, while TDA provides the opportunity of episode splitting, the TDA manual gives only a few hints as to how accomplish this. Without being able to discuss this problem thoroughly, I do hope that simply providing an example may help other users getting along a little better.
Throughout, I use a very simple, though common, case, namely, data that are stored in an ASCII-File where each line in the file represents one case. TDA can read other files, most importantly a certain type of archive (where data are compressed to save disk space). These are problems that are not at all dealt with in this paper. I also say nothing about all the other statistical procedures built into TDA. That is, only a very small part of TDA is covered here.
In sum, this is indeed a very elementary introduction. Complete mastering even of the procedures I describe here requires studying the TDA manual. Why, then, this introduction? Mainly because in my view the TDA manual fails to give such a short elementary overview which may lead to a quick, even though incomplete, idea of how to work with TDA.
Since I wrote the first (paper) version of this introduction, the book by Hans-Peter Blossfeld and Götz Rohwer: Techniques of Event History Modeling. New Approaches to Causal Analyses, Mahwah, NJ: Erlbaum, 2nd ed. 2002 has appeared. This gives a very thorough introduction and yet is much better organized than the TDA manual. Thus, a very good way to learn TDA is to start to work with this introduction, then use this book to achieve a deeper understanding and finally use the TDA manual to learn about additional features.
By the way: The TDA manual can be found on the TDA server at Bochum (see below at "Downloads"). The manual consists of a number of papers that can be downloaded each in turn or as one single (zipped) file (tman.zip). These papers are in PostScript format. To have them printed, you need a printer capable of understanding this format. On a PC under MS-Windows (or DOS), they are sent to the printer with the (DOS) command "copy", e.g: "copy d080302.ps lpt1" (this will print the file d080302.ps, provided that the printer is connected to the "lpt1" interface, which usually will be the case). With freeware program ghostview -- or GSview for PCs running under Windows --, you can view such files on your screen and also send them to your printer. (Again, see section "download" below.)
Note that I use TDA in a Windows environment, and I will refer to this environment in the following. Actually, most parts of this paper -- i.e. all those parts that show how to set up TDA command files -- should be helpful for users with a different operating system as well.
It is assumed that the reader already knows what transition data or event history analysis is all about, for instance, from the articles by Carroll (1983), Teachman (1983), Hutchison (1988a, b), Kiefer (1988), and Petersen (1991a), or from the books by Collett (1999 [1994]), Hosmer/Lemeshow (1999), Kalbfleisch/Prentice (1980), Lawless (1982), Allison (1984), Blossfeld/Hamerle/Mayer (1989), Yamaguchi (1991, especially chaps. 5 and 6), Blossfeld/Rohwer (1995) or Collett (1994). German readers are referred to the papers by Diekmann (1988), Diekmann/Mitter (1990, 1993), and Meinken (1992), as well as to the books by Diekmann/Mitter (1984), Blossfeld/Hamerle/Mayer (1986), and Andreß (1992); some may also wish to use my own very elementary introduction (Ludwig-Mayerhofer 1994) (see list of references below).
You may possibly have enough understanding of event history analysis if you know some of the following terms: survivor function, hazard rate, life-table estimator, Product-limit (or Kaplan-Meier) estimator, semi-parametric analysis of hazard rates (or Cox regression model), exponential model, Weibull model, log-logistic model.
Added section "TDA plots" with four subsections ("General", "Kaplan Meier", "Life table" and "Checking residuals") and made some minor corrections throughout. Also added link to download possibility for Ghostview/GSview. Finally, added "History" paragraph (the paragraph you are reading) to help users keeping track of changes. (At the moment I am not aware of any future changes, but let's wait and see ...).
Before June 2002Everything else.
Throughout, an example data set from Kazuo Yamaguchi, Event History Analysis, Newbury Park: Sage, 1991, Chapter 6, Table 6.6 (p. 152-7), will be used (by courtesy of Prof Yamaguchi). Although I will refer to that data set more fully in section "Analyses of an Example Data Set", it might be helpful if the reader has got an idea what the data set is about from reading the appropriate pages in Yamaguchi's book (pp. 139 et seq.). I can only give a very superficial description here.
The data set investigates possible determinants of drop-out from college. It consists of nine variables:
Col 1: ID number
Col 2: Duration until drop out of college or until end of observation period
Col 3: A status variable, indicating drop out (1) or a censored observation (0)
Col 4: Respondent's sex (0 = male, 1 = female)
Col 5: Average high school grades, from 1 (mostly As) to 5
Col 6: Part-time student (1; 0 otherwise)
Col 7: Time elapsed between leaving high school and entering college
Col 8: Time of marriage in "calendar time" (with 1 indicating January 1980, 2 February and so forth)
Col 9: Time entering college in "calendar time" (see Col 8)
The TDA program can be downloaded at http://www.stat.ruhr-uni-bochum.de/tda.html or from an anonymous FTP-Server at ftp://ftp.stat.ruhr-uni-bochum.de/pub/tda/. The program itself (zip- and zoo-archives for several operating systems) is to be found in the binaries subdirectory. It's up to you to choose the version you need corresponding to your local operating system. The README subdirectory in binaries indicates which operating system the different archives are assigned to.
The TDA manual can be found in subdirectory "documentation" (or "doc" on the FTP server). Also, a large number of example command files can be obtained from subdirectory "examples".
Kurt Schmidheiny, Universitat Pompeu Fabra, has written a nice application, aptly called WinTDA, that facilitates working with TDA in a Windows environment. It can be downloaded at http://research.schmidheiny.name (see section "software" at the bottom).
GSview (for Windows users) or Ghostview are programs that will enable you to read the TDA manual on the screen and to print it. The same goes for plots created with TDA. At http://www.cs.wisc.edu/~ghost/ you should find a possibility to download the program of your choice together with Ghostscript which is necessary for GSview or Ghostview to run on your computer. Also, helpful information is available there. If this link should not work, simply using any search engine with keywords "gsview (or ghostview) download" should do.
Downloading this TDA Introduction on your local computer may be worthwhile when you use the Introduction frequently.
Before starting the download procedure you are recommended to create a new directory for the TDA Introduction on your computer. To download the Introduction means to receive a selfextracting zip-archive (tdaintro.exe) to be stored on your computer; you might wish to use the directory created prior to the download. By opening this file (i.e. double-clicking on tdaintro.exe), the entire guide is unpacked into a directory of your choice, preferrably the very same directory.
For working with the TDA Introduction at home, open the file tda_h.htm with your Web Browser (double clicking on the file will usually do as well).
It may be worth repeating the download procedure from time to time for profiting from updates of content and form. We will indicate on which date the latest version of the download archive was created.
And now you are invited to download the TDA Introduction (version June 2002, 106 KB).
This page is a process initiated and maintained by
Prof. Dr. Wolfgang Ludwig-Mayerhofer
Universität Siegen
FB 1 - Soziologie
57068 Siegen
Homepage at the University of Siegen
E-Mail:
This e-mail address is a graphics file in order to protect me from spam mails. Sorry for the inconvenience -- you have to type it into your e-mail client.
Last update: 14 Apr 2007