Event history analysis has become an important analytical tool in many fields of the social sciences. This course examines the standard tools used for event history analysis-things like life tables, Kaplan Meier estimates, Cox proportional hazards model, and parametric survival models. Additionally, the course focuses on building a tool kit for developing custom models that involve "non-standard" methods like subgroup heterogeneity, incorporation of "immune" individuals, mixture models, models for clustered observations, multi-state models and social diffusion models.
This course is not specific to any field within the social sciences, although many of the examples in this course are taken from demography.
After completing this course you will have: (1) a working familiarity with the tools and concepts for solving quantitative problems in the statistical analysis of time to events; (2) developed the skills and background to evaluate the use (and misuse) of event history analysis in contemporary social science research; (3) built a tool kit for developing custom event history models.
The class is Tuesday and Thursday, 10:30 pm - 12:20 pm in 138 Savery Hall.
Office hours: I will typically be available after class for office hours. Other times can be arranged. Call or email (djholman@u.washington.edu) me with questions or to set up an appointment.
The textbooks are
Additional readings and handouts will supplement the text. These readings will illustrate principles discussed in lecture and the text, and will also be used as the basis for some class discussions. Readings are available here. A selection of readings (largely collected by past students of this course) are available here.
Grades: There will be 5 problem sets (12% each) that will make up 60% of your final grade, and a final project (40%). There are no exams.
The five problem sets will consist of analytical exercises and other short problems. Frequently, the problems will require the use of computer software.
I recommend that you get an account on the CSDE Windows network. The CSDE systems have many useful programs for doing event history analysis (request a Windows account here). Data sets for this course will be available on both the course web site course web site (here) and the CSDE server.
You can use books, readings, notes, and web pages to help you work on the problems. In fact, you can work in groups on most exercises. Grades for late problem sets will depreciate by 10% per day, including any fraction of a day late.
You can use any software that works for you and gets the job done. For example, when we work with the Cox proportional hazards regression model, almost any standard statistical software will work. For other assignments only a few "packages" will be able to easily perform the analysis. One option, and one I encourage, is that you begin working with a statistical programming language. Perhaps the best overall statistical programming language is R. However, the language mle written by your instructor is a good choice as well for advanced modeling. If there is sufficient interest, I will offer optional weekly sessions in a computer lab that introduces mle programming. There are a number of short courses and online tutorials that introduce R (or S-plus).
The mle package is freely available from http://mlelabs.com for use on your Windows or Linux computer. Extensive documentation is available online. You can download a pdf version of the documents for browsing or printing. The mle program is also installed on the CSDE terminal servers. Most of the exercises that you can do in mle can also be done in other statistical programming languages (S-plus, R, Matlab, Gauss, Octave). You are free to use any of these for your work under the idea that learning one such language will help you understand any other.
Projects: 40% of your course grade will be based on a project. This project can take one of several forms:
Week 1: Introduction to Event History Analysis (Jan 3, 5)
Reading: Box-Steffensmeier and Jones (BSJ) Ch 1, 2; Allison Ch 1 Handout: Lecture notes Handout: Notes on Writing an Event History Analysis Paper (Tuma) Overheads: Jan 3 Overheads: Jan 5
Week 2: Parametric Survival Models (Jan 10, 12)
Reading: BSJ Ch 3; Allsion Ch 2 Handout: Notes on distributions Handout: Notes on likelihoods Overheads: Jan 10 Overheads: Jan 12 Problem set 1 distributed (Tuesday).
Week 3: More Parametric Survival Models (Jan 17, 19)
Reading: BSJ Ch 3; Allsion Ch 4 Handout: Messy Data Handout: Covariates Overheads: Jan 17 Jan 19. "Operations suspended" for snow today Problem set 1 due (Tuesday) Problem set 2 distributed (Tuesday).
Week 4: Empirical and seemingly empirical models (Jan 24, 26)
Reading: Gehan (1969); Blossfeld and Rohwer Ch 3; Allsion Ch 3. Optional reading: Bohoris GA (1994) Comparison of the cumulative-hazard and Kaplan-Meier estimators of the survivor function. IEEE Transactions on Reliability 48(2):230-232. Overheads: Jan 24 Overheads: Jan 26 Problem set 2 due (Tuesday) Problem set 3 distributed (Tuesday).
Week 5: Cox Proportional Hazards Models (Jan 31, Feb 2)
Reading: BSJ Ch 4; Allsion Ch 5. Overheads: Jan 31 Overheads: Feb 2 Problem set 3 due (Thursday)
Week 6: More Cox Model and Piecewise Models (Feb 7, 9)
Reading: BSJ Ch 4; Allson Ch 7. Overheads: Feb 7 Overheads: Feb 9 Problem set 4 distributed (Thursday).
Week 7: Model Selection and Diagnostics (Feb 14, 16)
Reading: BSJ Ch 6, 8; Wood et al. (1994). Handout: Selecting an event history model Overheads: Feb 14 Overheads: Feb 16
Week 8: Models of Mixed Populations (Unobserved Heterogeneity I) (Feb 21, 23)
Reading: Holman (2004) Overheads: Feb 21 Overheads: Feb 23 Problem set 4 due (Tuesday) Problem set 5 distributed (Thursday).
Week 9: Continuous Unobserved Heterogeneity (Feb 28, Mar 1)
Reading: BSJ Ch 9; Vaupel and Yashin (1985). Overheads: Feb 28 Overheads: Mar 1
Week 10: Some Advanced Models (Mar 6, 8)
Reading: BSJ Ch 10; Allsion Ch 6, 8; Wood et al. (1994); Strang and Tuma (1993) Overheads: Mar 6 Overheads: Mar 8 Problem set 5 due (Thursday)
Paper Due: Wednesday, March 14, 12:30 pm in my department mailbox