RS719:

Logistic and Log Linear Models

Table of Contents

Click here to jump to Exercises and Sample Programs.

For information about other methods courses offered in Rural/Development Sociology, see Methods Courses available in Rural/Develoment Sociology.

Overview

Rural Sociology 719 is an advanced level graduate course in research methods and statistical applications focusing primarily on issues of analysis and prediction under conditions of constrained or limited information IVs and DV (logistic regression models). Secondly, this course examines models for assessment of underlying association and/or correlational structure in a data set (log-linear models). It can be conceived of as a second course in regression.

I like to think of the course as the second part of an advanced level graduate methods sequence; the first part of the sequence, RS 718, deals with multidimensional measurement and classification techniques: factor analysis, MDS, cluster analysis. Enrollment in the first part of this advanced sequence is not requisite to enrollment in this course, but "it wouldn't hurt."

Prerequisites

Being an advanced methods course, this course has prerequisites. Specifically, It presumes you have completed two previous courses in statistics: one of which should have included a study of multiple regression and correlation involving more than one independent variable, and one course which provides some exposure to the analysis of categorical variables. It also assumes more than a cursory experience with at least one of the major statistical packages (e.g., SPSS, SAS, SYSTAT).

Topics Covered

Rural Sociology 719 focuses on the use of some advanced features and extensions of multiple regression for analyzing nonlinear relationships among several variables. After a quick review of multiple linear regression, the majority of the semester is concerned with techniques for handling categorical and other "limited information" DVs. Using the linear probability model as a transition from traditional linear regression theory and methods, logit models will be introduced. Topics will include the simple and multivariate logistic regression models, interpretation, model building, goodness-of-fit, assumptions and diagnostics. Both dichotomous and polytomous logit models will be examined. The second major topic area is log-linear models.

Lecture material is designed to review and highlight the major points in the readings for a given topic as well as introduce new developments and interrelationships between concepts or procedures that might not be obvious through reading alone. Numerous examples will be presented to illustrate the main concepts and procedures.

Because the emphasis is on applied analysis, it is expected that the student will complete several laboratory assignments using the computer. Accordingly, some laboratory time will be devoted to assisting you in tailoring the use of computer package programs to the assignments.

Throughout the course you will have the opportunity to enjoy using actual data and canned computer programs (SAS, SPSS) to familiarize yourselves with data-handling techniques and available computer programs. Many find this experience is helpful in thesis work and other course work. It is anticipated you will complete 7 or 8 assignments by the end of the semester.

Organization of the Course

One of the goals of this course is to simulate real life research through a combination of reading, discussion of principles, and application of these to analysis of a genuine study such as you might be involved with. As you can see by glancing at a subsequent section of the prospectus, Outline of Topics, Readings, and Exercises, corresponding to each lecture topic, are a selected set of key readings which introduce or elaborate the major concepts integral to that topic. These provide the background to class discussion on that topic. Additional to these readings, typed lectures may be distributed which integrate the readings, highlighting the main points, and providing additional discussion or viewpoints. Also note that for each topic there is a laboratory exercise developed which permits you to synthesize these concepts and principles and apply them to actual analyses but will incorporate the concepts and principles applicable to that lab as found in your readings and class discussion. That is, your laboratory assignments will include both theoretical background and conceptual discussions as well as interpretations and comments about the data analyzed. This integration and synthesis is requested in lieu of formal tests.

Laboratory Assignments

It is anticipated that you will complete 7 or 8 assignments during the semester. Some may involve hand calculations. The rest involve the use of computer programs. There will be approximately one assignment due every two weeks.

Grading Policy

Your grade for this course will be based on your class participation and performance in completing the laboratory assignments. Also homework in addition to the labs may be assigned. Labs are to be written and handed in along with the appropriate computer printouts. Performance is evaluated in terms of (1) how well you understand and are able to "reason with" the methodology principles discussed in lectures or presented in the readings, and (2) how proficiently you can apply these principles to an evaluation of the results obtained in analyzing a data set. Please note that no extra credit is given for selecting a "good" data set as opposed to a "bad" set--i.e., one which yields the results you desired. Disproving a hypothesis is just as important as producing confirming evidence.

As a graduate course, students are expected to perform at that level of ability: assignments thoughtfully written and typed, turned in when due, with computer output attached. Generally, an assignment is due every two weeks, as you can determine from examining the Outline of Topics, Readings, and Exercises in this prospectus. To help you pace yourself in completing all the assignments, I have a policy of deducting one letter grade from any paper turned in a week after the "due date", two letter grades if two weeks late, etc. A due date will always be publicly announced for each assignment. Accordingly, after 2:30 one week later, your grade is dropped one letter; two weeks, two letters, etc.

Grading will be conducted in accord with university policy. No S/U options will be given, however. Anyone wishing to visit or "sit in on" the course should enroll as AUDIT. As the grading scheme used in this class is somewhat complex, let's take a moment to portray it.

Each laboratory assignment is generally worth 100 points. Hence, all 7 or 8 of the labs are worth approximately 700-800 points (sometimes bonus points are offered for extra work). For each lab, each procedure in the assignment is assigned a "point value" which generally ranges from 5 to 15 points, depending on the procedure's importance and the number of procedures to be performed in the lab. Points will be deducted from the announced "point value" of each procedure if it is judged you did not fully meet expectations of the analysis, interpretation, or write-up of the procedure.

To help you gauge your relative performance on each assignment, both the total points assigned on that lab plus a letter grade will be reported to you. The letter grades will be assigned approximately as follows: A+ = 97-100 points; A = 93-96 points; A- = 89-92 points;
B+ = 85-88 points; B = 81-84 points; B- = 77-80 points; C+ = 73-76 points; C = 69-72 points; C- = 65-68 points, etc.

The points you receive on each lab will be cumulated toward your final grade. Hence, your final grade will be a function of the sum of the points you received on each lab plus any bonus points received. That sum will be divided by the number of labs and translated to letter grades analogously to the above point categories used for grading each individual lab.

Class Times

Class periods for this course are of two types: lecture periods and lab periods. Lecture periods are convened at 12:20-2:20 T-R. Lab periods will be generally held during the last hour Thursday afternoons.

Teaching Staff

Joe Francis is located in 334A Warren Hall, Phone 5-1687, e-mail jdf2. Office hours will be from 2:30 to 3:30 Tuesday and Thursday. Amy Guptill, the teaching assistant, is located in 336 Warren Hall, Phone 5-2065, e-mail aeg9. Her office hours will be announced.

Readings and References

Demaris, Alfred
1992 Logit Modeling. Sage Publications. Sage University Paper #86.

Hosmer, David W., and Stanley Lemeshow
1989 Applied Logistic Regression. New York: John Wiley & Sons.

Ishii-Kuntz, Masako
1995 Ordinal Loglinear Models. Sage Publications. Sage University Paper #97.

Liao, Tim Futing
1995 Interpreting Probability Models. Sage Publications. Sage University Paper #101.

Menard, Scott
1995 Applied Logistic Regression Analysis. Sage Publications. Sage University Paper #106.

Aldrich, John H., and Forrest D. Nelson
1984 Linear Probability, Logit, and Probit Models. Beverly Hills, CA: Sage Publications. Sage University Paper # 45.

Breen, Richard
1996 Regression Model: Censored, Sample Selected, or Truncated Data. Sage Publications. Sage University Paper #111.

Eliason, Scott
1996 Maximum Likelihood Estimation. Sage Publications. Sage University Paper

Hagenaars, Jacques A.
1995 Logliner Models with Latent Variables. Sage Publications. Sage University Paper #94.

Neter, John, William Wasserman, and Michael H. Kutner
1985 Applied Linear Statistical Models. Homewood, IL: Irwin.

Neter, John, William Wasserman, and Michael H. Kutner
1989 Applied Linear Regression Models. Homewood, IL: Irwin.

Pedhazur, Elazar J.
1952 Multiple Regression in Behavioral Research. New York: Holt, Rinehart, Winston.

Rudas, Tamas
1997 Odds Ratios in the Analysis of Contingency Tables. Sage Publications. Sage University Paper #

Questions, Comments, or Problems

If you have any questions or problems concerning this course, contact Joe Francis. He can be reached by e-mail at jdf2@cornell.edu or by phone at 255-1687. This page last modified 19 January 1998.